Addis Ababa University

College of Natural Sciences

Development of Spell Checker for Guragina Language

Mengistu Gebre Bokane

A Thesis Submitted to the Department of Computer Science in Partial

Fulfillment for the Degree of Master of Science in Computer Science

Addis Ababa, Ethiopia

March, 2024
 Addis Ababa University

College of Natural Sciences

Mengistu Gebre Bokane

Advisor: Yaregal Assabie (Ph.D.)

This is to certify that the thesis prepared by Mengistu Gebre Bokan, titled Development of Spell
Checker Algorithm for Gurage Language, and submitted in partial fulfillment of the requirements
for the Degree of Master of Science in Computer Science complies with the regulations of the
University and meets the accepted standards with respect to originality and quality.
Signed by the examining committee:

Name Signature Date

Advisor: Yaregal Assabie (PhD) _________________ ___________

Examiner: Ayalew Belay (PhD) _________________ ___________

Examiner: Mesfin Kifle (PhD) _________________ ___________

 ABSTRACT

A spell checker is an essential tool in Natural Language Processing (NLP). Its purpose is to
identify and correct spelling errors in text, providing suggestions for correct spellings in a specific
language. Spelling errors can be categorized into two types: non-word errors and real-word
errors. Non-word errors are misspelled words that have no meaning in the particular language,
while real-word errors involve words that exist in the language but are used incorrectly in terms
of semantics and syntax.

The research focused on non-word error detection as a strategic decision, given the complexity
and limited resources available for the Gurage language, also known as Guragina. This language
consists of over thirteen varieties and different orthographies, but there is a modern standard.
Currently, there is no existing spell checker for any Guragina Language varieties or the standard.
Addressing non-word errors first provides a solid foundation before tackling the more challenging
task of real-word error detection and correction. This phased approach allows researchers to
make meaningful progress on this under-resourced language, rather than attempting to solve the
entire spell checking problem at once. The intention is to use the non-word spell checker as a
starting point, then leverage that knowledge to progressively tackle real-word error handling.

This work introduce a non-word spell error checker for the standard Guragina Language. The
system detects and corrects errors using Ratcliff algorithms for identification and distance
calculator techniques for correction. The prototype of the system was developed using Python.

We evaluate the performance of the system using metrics such as accuracy of 98.27%, precession
of 98.07%, recall of 97.75%, and F1 Score of 95.45%. Future work includes enhancing rule
definitions by incorporating word classes, handling exceptions, adding supplementary spell
checker functionalities, and expanding the system to encompass real-word errors.

Keywords: Spell Checker; Gurage Language; Morphology, morphological analysis, linguistic

resources, rule-based approach, under-resourced languages
 ACKNOWLEDGEMENT

First and foremost, I want to express my deep gratitude to the Almighty God and Virgin Mary for
blessing me with good health, wisdom, strength, and grace, which enabled me to successfully
complete and present this research study.
I would like to extend my heartfelt appreciation to my advisor, Dr. Yaregal Assabie, for his
invaluable guidance, encouragement, and constructive feedback throughout the entire process.
From the initial selection of my research topic to the final stages of the study, his unwavering
support and motivation have played a vital role in my progress. I would also like to thank Dr.
Fekede Minuta and Mr. Bahiru Lilaga for their invaluable linguistic support, as well as the Gurage
Zone Cultural Tourism Office for providing the necessary materials for data collection.
I sincerely thank Mr. Ayechew Tefeta and the entire management team at my office for their
encouragement and willingness to dedicate time to support my study. Their support has been
crucial in helping me complete this research.
To my beloved wife, Blene (ብሌኔ), and my children, Eyoab and Aminen, I am deeply grateful for
your unwavering support and encouragement throughout this journey, from the initial stages of
topic selection to the presentation of my final defense.
I would also like to acknowledge and express my gratitude to my dear family, especially my
mother Adanech Dereja, and my late father Gebre Bokane (even though you are no longer with
me, this was your dream). Your love, encouragement, and support have been constant sources of
strength throughout my academic life. Additionally, I extend my thanks to Mr. Mitku, all my
siblings, and my friend Zelalem for their kind support.
In conclusion, I am sincerely grateful to all those who have contributed to my research study,
whether through their guidance, encouragement, or support. Your presence in my life has made a
significant difference, and I am truly indebted to each and every one of you.
TABLE OF CONTENTS
LIST OF TABLES ....................................................................................................................... iv

LIST OF ALGORITHMS ............................................................................................................ v

LIST OF FIGURES ..................................................................................................................... vi

ACRONYMS ............................................................................................................................... vii

CHAPTER ONE: INTRODUCTION ........................................................................................ 1

1.1. Background of the Study .................................................................................................. 1
1.2. Motivation ........................................................................................................................ 2
1.3. Statement of the Problem ................................................................................................. 2
1.4. Objectives ......................................................................................................................... 3
1.5. Research Methodology ..................................................................................................... 3
1.6. Scope and Limitations ...................................................................................................... 5
1.7. Application of the Results ................................................................................................ 5
1.8. Organization of the Thesis ............................................................................................... 6

CHAPTER TWO: LITERATURE REVIEW ............................................................................ 7

2.1. Socio-Linguistic Situation in the Gurage Region ............................................................ 7
2.2. Overview of Guragina Language ..................................................................................... 9
2.2.1. Variety of Guragina Language................................................................................ 10
2.2.2. Guragina Language Writing System....................................................................... 12
2.3. Morphology .................................................................................................................... 13
2.3.1. Verb......................................................................................................................... 13
2.3.2. Noun........................................................................................................................ 22
2.3.3. Adjective ................................................................................................................. 29
2.3.4. Pronouns ................................................................................................................. 30
2.3.5. Number ................................................................................................................... 32
2.3.6. Adverbs ................................................................................................................... 32
2.3.7. Conjunctions ........................................................................................................... 34
2.4. Spelling Errors................................................................................................................ 35
2.4.1. Types of Spelling Errors ......................................................................................... 35
2.4.2. Techniques of Spell Checker .................................................................................. 37

i
 2.5. Summary ........................................................................................................................ 45

CHAPTER THREE: RELATED WORK ................................................................................ 46

3.1. Introduction .................................................................................................................... 46
3.2. Spelling Checker for Semitic Language......................................................................... 46
3.1.1. Spelling Checkers for Amharic ............................................................................... 46
3.1.2. Spelling Checkers for Arabic .................................................................................. 49
3.3. Spelling Checker for Afro-asiatic Language .................................................................. 50
3.3.1. Spelling Checkers for Afaan Oromo....................................................................... 50
3.3.2. Spelling Checkers for Kafi Noonoo ........................................................................ 50
3.4. Spelling Checker for Indo-European Language ............................................................. 50
3.4.1. Spell Checker for Bengali Language ...................................................................... 50
3.4.2. Spelling Checkers for English ................................................................................ 51
3.5. Spelling Checker of Austronesian Language ................................................................. 52
3.5.1. Spell Checker for Malay ......................................................................................... 52
3.6. Summary ........................................................................................................................ 53

CHAPTER FOUR: DESIGN OF GURAGINA SPELL CHECKER .................................... 54

4.1. Introduction .................................................................................................................... 54
4.2. General Architecture ...................................................................................................... 55
4.3. Preprocessing ................................................................................................................. 56
4.3.1. Tokenization ........................................................................................................... 57
4.3.2. Vowel Consonant mapping ..................................................................................... 58
4.4. Error Detection ............................................................................................................... 58
4.4.1. Guragina Stem Extraction ....................................................................................... 58
4.4.2. Matching ................................................................................................................. 60
4.5. Correction Suggestion .................................................................................................... 61
4.6. Distance Calculator ........................................................................................................ 62
4.6.1. Suggestion Ranking ................................................................................................ 66

CHAPTER FIVE: EXPERIMENT ........................................................................................... 68

5.1. Introduction .................................................................................................................... 68
5.2. Data Preparation ............................................................................................................. 68
5.3. Implementation............................................................................................................... 69

ii
 5.4. Prototype of the System ................................................................................................. 70
5.5. Evaluation of the Ratcliff / Obershelp Algorithm .......................................................... 70
5.6. Performance Results of Distance Calculator .................................................................. 72
5.7. Comparison of the Proposed System with Existing Ones .............................................. 73
5.8. Discussion ...................................................................................................................... 74

CHAPTER SIX: CONCLUSION AND FUTURE WORK..................................................... 75

6.1. Conclusion...................................................................................................................... 75
6.2. Contribution of the Thesis .............................................................................................. 76
6.3. Future Work ................................................................................................................... 77

REFERENCES............................................................................................................................ 79

APPENDICES ............................................................................................................................. 85

iii
 LIST OF TABLES

Table 2. 1 Derivation of ʧənə for different person. ..................................................................................... 14

Table 2. 2 Derivations of Guragina complex verb stems with tə- ............................................................... 15
Table 2. 3 Reduplication of word with morpheme tə .................................................................................. 15
Table 2. 4 Derivation of Causatives stem in Perfective, Imperfective and Jussive words .......................... 16
Table 2. 5 Guragina Verbal nouns.............................................................................................................. 17
Table 2. 6 The negation of stem bəkər “miss” ............................................................................................ 17
Table 2. 7 Perfective negations ................................................................................................................... 18
Table 2. 8 Negated imperfective ................................................................................................................. 19
Table 2. 9 Negated imperfective and jussive............................................................................................... 19
Table 2. 10 Negative Command with morpheme /ɨn-/ and /at-/ .................................................................. 19
Table 2. 11 Negation with morpheme /e-/ ................................................................................................... 20
Table 2. 12 Different modality of obligation............................................................................................... 22
Table 2. 13 Different morphological inflections ......................................................................................... 23
Table 2. 14 Examples of Guragina’s Plural with third person pronoun .................................................... 23
Table 2. 15 Gender in Guragina ................................................................................................................. 24
Table 2. 16 Guragina’s Gender with “arɨst” and “təbat” ......................................................................... 24
Table 2. 17 Definiteness of nouns ............................................................................................................... 25
Table 2. 18 Definiteness with hʲita and huta ............................................................................................... 25
Table 2. 19 Derivations of Noun ................................................................................................................. 27
Table 2. 20 Guragina Compound word ...................................................................................................... 27
Table 2. 21 Guragina’s root-pattern nouns ................................................................................................ 29
Table 2. 22 Basic Guragina Personal Pronouns ........................................................................................ 30
Table 2. 23 Plural and Singular Objective (accusative) pronoun with ʔə .................................................. 30
Table 2. 24 Pronominal suffixes of possession ........................................................................................... 31
Table 2. 25 Guragina’s Adverbs of time ..................................................................................................... 33
Table 2. 26 Guragina’s Adverbs of frequency ............................................................................................ 33
Table 2. 27 Adverbs of manner ................................................................................................................... 34
Table 2. 28 Adverb of Place ........................................................................................................................ 34
Table 2. 29 Typographic Errors ................................................................................................................. 36
Table 5. 1 Composition of the lexica based on the category of the stem words and affixes ....................... 69
Table 5. 2 Performance Results of Distance Calculators ........................................................................... 72
Table 5. 3 Comparison of Evaluation results with previous similar work ................................................. 74

iv
 LIST OF ALGORITHMS

Algorithm 4. 1 Tokenizer algorithm used by the spell checker .................................................... 57

Algorithm 4. 2 Stem extractor Algorithm. .................................................................................... 59
Algorithm 4. 3 General Matching Algorithm ............................................................................... 60
Algorithm 4. 4 Guragina Word Distance Algorithm (GWDA) used by the spell checker ............ 63
Algorithm 4. 5 Suggestion Ranker Algorithm............................................................................... 67

v
 LIST OF FIGURES

Figure 4. 1 Design Science Research Process Model ................................................................... 54

Figure 4. 2 General Architecture .................................................................................................. 56
Figure 4. 3 Examples of distance calculation's values ................................................................. 64
Figure 4. 4 Distance between characters by family and order ..................................................... 65
Figure 5. 1 Prototype of the algorithm.......................................................................................... 72
Figure 5. 2 Distance between characters by family and order ..................................................... 73

vi
 ACRONYMS

Bi-LSTM Bidirectional Long Short-term Memory

GL Guragina Language
GLSC Guragina Language Spell Checker
DCAUS Direct Causative
DLT Dictionary Lookup Technique
DCAUS Indirect Causative
GWDA Guragina Word Distance Algorithm
IPA International Phonetic Alphabet
IPS Impersonal
MA Morphological Analysis
MT Machine Translation
OCR Optical Character Recognition
NLP Natural Language Processing
QA Question Answering
SC Spell Checker
VC Vowel Consonant

vii
 CHAPTER ONE: INTRODUCTION

1.1. Background of the Study

Natural Language Processing (NLP) is the computer-based approach to text analysis that is based
on a set of theories as well as a set of technologies. The ultimate goal of NLP, as a branch of
artificial intelligence, is to read, decode, understand and make sense of human languages in a
meaningful way [1]. There are too many languages, most of them have different alphabets and a
lot of rules. Their treatment of using a single linguistic tool is practically difficult. These are strong
reasons to develop different tools for various languages including spellchecker.
In NLP, a spell checker (spelling checker) is one of the most important areas in the field, which
focuses on detecting and correcting spelling errors in a text. Currently, many NLP applications use
SC tools either independently or embedded within another system [2].
In different languages, spelling errors are mainly caused by insertion (insertion of the wrong
character), deletion (omission of a letter), substitution (using one letter instead of the other), and
transposition (miss positioning of letters). According to Kukich [3], spelling errors are generally
split into two types: Typographic and Cognitive. Typographic errors occur when the correct
spellings of the word are known, but the word is mistyped by mistake. These errors are mainly
associated with the keyboard and therefore do not follow any language criteria.
The typographical errors fall into one of the four categories. A study by Damerau [4] shows that
80% of typographical errors come from one of the following four categories. Spelling error
detection is the main thing in spell checking to detect the errors in written text. It is used to flag out
misspelled words in a text. To suggest a set of possible corrections, a spellchecker has to identify the
first misspelled words. The most important approaches that are used to develop an SC system are
dictionary look-up and N-gram approaches [3], [7], [8].
Ethiopians speak around 80 different languages [9]. The majority of Ethiopian languages,
including Gurage Language (GL), are under-resourced, as are the majority of global languages, in
that they lack proper language processing tools and methodologies.
The language known as Gurage (GL) is referred to as Guragina and falls under the South-
Ethiosemitic division of the Semitic branch within the Afroasiatic language family. It encompasses
approximately thirteen different variations. The Guragina language is further divided into sub-
groups, namely North Gurage, West Gurage, and East Gurage. The existence of these dialectal
1
variations and other related factors poses challenges when attempting to create a unified natural
language processing (NLP) application, including spell checker (SC). However, despite the
variations, a standard Guragina language and orthography have been recently established. This
study primarily focuses on the development of a spell checking and correction system for the
standardized Guragina language.

1.2. Motivation
Spell-checking is one of the most important applications, used to generate spelling error-free
documents. Spell-checking techniques developed and implemented in languages [5] such as
English, Arabic, Indonesian, Bangla, Indian, etc.
It is very common practice to see a lot of spelling errors in typed Guragina documents. Even fluent
speakers and writers are discouraged to write for fear of being misspelled and/or do not know how
to write using Guragina orography. In addition, most people who type Guragina make mistakes
and do not even notice them. Even after the manual correction, it is still very common to see
misspellings in printed or electronic documents that lead to misinterpretation of the meaning of a
word. This has motivated us to develop an SC for the GL.

1.3. Statement of the Problem

A spell checker is considered one of the most important NLP applications used to enhance the
success rate of NLP applications, like machine translation, search engines, social media, word
segmentation, proofreading, information retrieval, and natural language understanding [10].
Especially, for the languages with low resources and lack of computer tools, like Guragina, SC
plays a vital role.
The GL is one of the morphologically complex SOV and low-resource languages, with many
varieties. Spell checker tool developed and implemented for one language cannot be applied to
others directly [11]. To develop most NLP applications such as SCs for such languages, it may
need to advise the language’s expertise, understand the language morphology (word categories
and their possible inflections, derivation, and compounding), and other its unique characteristics.
For this reason, it is difficult to develop an SC system easily and needs research work for Gurgigna.
On the other hand, most varieties of Guragina Language are morphologically similar and share

2
most of their linguistic features. To take advantage of this, we are interested in doing a pattern-
based spell checker for the Standard Guragina Language.
Although SC is very important to the GL, to our knowledge, there is no study undergone about SC
for Standard GL and any of its varieties. This work is very important not only as an SC but also
will help to develop and preserve the language by encouraging and supporting the language users
to write and communicate by using this language on different platforms.
Generally, the first and great sense that made us this research is that different studies have been
performed on SC worldwide and some attempts of a few Ethiopian languages like Amharic, Afan
Oromo, Tigrigna, etc., but not at all with Standard Guragina Language.

1.4. Objectives
1.4.1. General Objective
The main objective of this thesis is to develop a spell checking and correcting system for the
Standard Guragina Language that leverages a morphology-based approach.

1.4.2. Specific Objectives

To accomplish the aforementioned general objectives, the following specific tasks have been
undertaken:
• Review related research works to understand the morphological structure of Guragina
words and to identify the state-of-the art in spell checking
• Collect a list of Guragina words used for testing.
• Develop an system that check and suggest corrections
• Develop the prototypes of the system
• Evaluate the performance of the system

1.5. Research Methodology

The research methodology employed in this study is the design science research methodology [12].
This methodology focuses on generating an artifact in the form of a model for a Guragina
Language spellchecker. As a result, a carefully developed and tested artifact or prototype is
produced, incorporating an algorithm to identify and correct spelling errors in Guragina.

3
Literature Review

Our research involved an extensive examination of diverse literature sources such as books,
journals, articles, online resources, thesis, and dissertations. The purpose of this comprehensive
review was to acquire relevant information for our study. Additionally, we actively sought input
from linguistics experts to enhance our comprehension of the structure and characteristics specific
to the Guragina Language. By collaborating with these experts, we were able to delve deeper into
the linguistic aspects of the Guragina Language and explore established approaches for detecting
spelling errors across various languages.

Data Collection
To carry out experiments on corpus-based SC collection of various forms of organized words is
necessary. We collect and organize the text corpus for this research work from Welkite University,
various medias, Gurage Zone Culture and Tourism Office.
Afterward, the data pre-processing was the next step in data collection. Real-world data is often
incomplete, inconsistent, and/or lacking in certain behaviors or trends, and is likely to contain
many errors. Data pre-processing is a proven method to resolve such issues. Data pre-processing
prepare raw data for further processing. We collected a list of stem words of verb, nouns,
adjectives, adverbs and various affixes through a combination of manual and semi-automated
methods. This involved the systematic identification and extraction of affixes and derivations from
individual words. By utilizing these affixes and deriving from a single word, we generated a wide
range of derived words, including verbs, nouns, adverbs and adjectives. These derived words
encompass various parts of speech commonly found in Guragina.

Prototype Development
We build a prototype to validate our work using the different tools. To develop the prototype of
the system, appropriate supporting tools are used to facilitate the study. These include a
morphological analyzer, the latest version of the Python programming language, and other free
tools. Python is an interpreter, interactive, object-oriented programming language. Python
combines remarkable power with very clear syntax. It has interfaces to many systems calls and
libraries, as well as various window systems [13]

4
Evaluation
In this research, we use accuracy, recall, precision and F1- score as main evaluation metrics to
measure the effectiveness of the selected approach for a Guragina Word Distance Algorithm
(GWDA).

1.6. Scope and Limitations

This study specifically focuses on detecting and correcting spelling errors within the endorsed
variants of the standard Guragina Language, among its various dialects. It is important to note that
the Guragina writing system did not have consistent guidance on orthography until 2020, resulting
in variations. In this study, we adopt a new standard and indorsed orthography by Gurage zone
administration council from among these variations.

The spellchecker system identifies correctly spelled words as valid and flags incorrectly spelled
words as invalid, while also attempting to suggest similar words as alternatives. However, it is
crucial to acknowledge that the lack of a standardized text corpus and limited digital resources for
the Guragina Language pose limitations to this study. The amount of prepared data for this study
is also relatively minimal due to these constraints.

1.7. Application of the Results

The system, which can detect and correct spelling mistakes in a language, will help significantly
the writing process on both offline and online applications. Users of the GL are can encourage
writing and communicating in the language. This work will be very significant in various areas.
Some of these include the following:
• In the learning and teaching processes of Guragina
• To generate a Guragina document with minimal or no spelling mistakes and save time for
the writer.
• For the development of the language by encouraging writers with the language.
• For Language Standardization: Spell checkers play a role in language standardization
efforts by promoting consistent spelling conventions. They help establish and enforce

5
 commonly accepted spellings within a language, contributing to linguistic cohesion and
clarity.
• In the development of other NLP applications like grammar checker, machine translation,
OCR, information retrieval, text synthesis, text extraction, speech text synthesis, etc.
• For anyone who wants to communicate through the text of GL offline or online.
As a result, this work will represent a big step toward detecting spelling errors in the Guragina
language, as well as a foundation for further NLP applications in the language. In addition to
verifying the accuracy and professionalism of the document. Furthermore, it will save time and
effort for a person to manually find misspellings throughout the document. Not only will this
study's application benefit anyone who wants to process Gurgigna documents, but it will also
provide a good opportunity for other researchers who want to develop other NLP applications
using the language. In general, this work will aid language development through motivating
writers, as well as teaching and learning activities.

1.8. Organization of the Thesis

The remaining part of this thesis is structured as follows. Chapter 2 covers the background,
linguistic characteristics, morphology, and writing system of the Guragina Language.
Additionally, this chapter goes through various types of spelling errors, and spell-checking
techniques. Chapter 3 presents the relevant research on the use of spell checkers and associated
topics. The design of the Guragina Language spell checking system is covered in the fourth
chapter. The findings of the experimental spell-checking system will be shown in the fifth chapter.
The last chapter, Chapter Six, will include a conclusion, a contribution, and future works.

6
 CHAPTER TWO: LITERATURE REVIEW

This chapter gives a description of various spellchecker studies on both the local language and
other languages using different techniques are discussed. It was used to obtain general information
on the spellchecker, the GL writing system and the different approaches of SC techniques. In
addition, presents a description of the state-of-the-art concepts related to our work. From Section
2.1 through Section 2.3 a step-by-step discussion of the sociolinguistics of Gurage region, the
origin of Gurage and Guragina Language varieties, its writing system and its morphological
characteristics is provided. The topics of general spelling errors and the various methods for error
detection and correction are covered in Section 2.4. We provide a brief summary of this chapter in
Section 2.5.

2.1. Socio-Linguistic Situation in the Gurage Region

The term "Gurage" refers to the population that the South Ethiopian-speaking community in the
Central Ethiopia Regional state of Ethiopia, one of the most recent regions to be split off from the
Southern Nation Nationality region. Geographically, Gurage is situated approximately 100 km far
from south of Addis Ababa, which is more or less bordered by the Rift Valley lakes in the East,
River Awash in the North and River Gibe in the West and Southwest in general [9].

The origin of Guragina varieties has been a subject of academic debate for decades among
Semitists. The debate emerges from the enduring lack of clarity about the origin of the Semitic
languages in general[15], [16].

Regarding where the Guragina variety came from, there are three hypotheses. The first one
presumes that the Gurage people originated in modern-day Eritrea, in a region known as Akale
Guzay under the rule of Amdetsion (1314-1344). Consequently, it is believed that the Guragina
Language's forefathers were King Amdetsion's troops. Oral tradition has offered evidence for this
notion. According to the second hypothesis, the ancestors of the Guragina speakers migrated from
east of Ethiopia during the expansion of Ahmed Gragn (1524-1543), a Muslim warrior. The
Islamist militia and army moved to the Gurage region during the expansion. The third theory holds
that the Guragina variations have African roots and were spoken in southern Ethiopia prior to
Amadetsion's rule and the expansion of Ahmed Grange. According to Menuta [14], the notion that

7
Ethiopian Semitic is the ancestor of all Semitic languages also raises the possibility that Guragina
variants are Ethiopian in origin and were spoken there from the start. The origins of the Gurage
people and the Guragina variations have mostly remained a mystery.

There were also different oral traditions regarding the origin of the language and the people of
Gurage. Based upon [17], the term, Gurage is from Gura - somewhere in Eritrea. The oral tradition
is that the suffix "ge" means "land" or "village," whereas the word "Gura" means "the people,"
which suggests that "Gurage" means the land or country of the Gura people. There are other words
along those lines that seem to support that argument.; for example, Harar-ge ‘the land of the Harari
people’, Abessh-ge ‘name of a district in Gurage Zone’, Arat-ge ‘a place in Kistane’ [14]. The
author says that "Gura" is a common word among Gurage communities that means "left", the
opposite of "right". Hetzron [18], opposes this oral tradition concerning the origin of the Gurage
people and language. According to the author, the traditional argument is not supported by any
linguistic evidence.

Socially, the Gurage communities socially organize themselves in different ways. Ethnic hierarchy
is one means of organization which consists of Tib, Bet, Wefencha and Den[14]. Tib is the highest
social hierarchy. It encompasses almost all Gurage clans. Bet is the second broad social hierarchy
which also consists of several Gurage clans. Wefencha is the third social hierarchy which consists
of sub-clans within the clan. Den is the fourth and the smallest social hierarchy which usually
refers to members of a sub-clan [14].

Yejoka is a traditional justice System for high-level societal decisions to be made as well as a
method for the administration of justice, by making meeting in the most common places and
Yejoka. Its main responsibilities include defending the community from external aggression,
enacting administrative and judiciary laws and supervising the implementations of the fundamental
rules of low. This system is takes place under podocarpus tree or Yejoka by elder’s by using
customary law called Kitcha.

With regard to the economic activities, the Gurage people are predominantly farmers [14], [19].
The main crop of the area is ‘Ensete’ from which they make different kinds of food, mainly a kind
of bread called wusa. Wild bananas usually grow in the lowland and semi-highland areas of the
Gurage Zone. The extreme highlanders, such as Muher do not plant wild bananas since they are
not productive in such a cold area [19]. Besides wild bananas, the Gurage people produce different

8
kinds of cereals such as barley, wheat, maize, pepper, lentils, ‘teff’ and commercial crops such as
coffee. Gurage people also rear animals such as cows, sheep, goats, horses, mules and donkeys.
Above all, they are well-known as traders among the remaining Ethiopian communities.

Gurage people are highly mobile [14]. A large number of people live outside Guraage zone. The
current establishment of Gurage is the result of a variety of social and historical reasons. Migration,
the movement of the Gurage people from the Gurage region to other parts of Ethiopia, is called
fannonet amongst the Gurage communities, also has its own influence on the current linguistic
situation in the Gurage region.

Nowadays, many Gurages are businessmen in the major cities of Ethiopia where Amharic is the
dominant language; Amharic has been imported into the Gurage area in part by these traders.
Guragina varieties are used solely as spoken languages (except Silt'e), and are restricted to
domestic and commercial communication. Because of this and other several factor Gurage
community were remaining decentralized; this weakened the people of Gurage and Guragina
Language [14]. Recently, the communities of Gurage were institutionalized together, with a
centralized administration. It being an opportunity to the Gurage community and disaggregated
language and culture together [14].

2.2. Overview of Guragina Language

The linguistic variations discussed in this study are primarily spoken in the southwestern region
of Ethiopia, located approximately 158 km from Addis Ababa. This area is traditionally referred
to as the Gurage area, and the term "Gurage" is used to describe both the linguistic features of the
region and the people residing there [14], [17]. However, it's important to note that not all
communities in the Gurage region are Semitic, such as the Libido and K'abeena communities.
Ethiosemitic languages belong to the Semitic language family and are spoken in Ethiopia and
Eritrea. They are broadly categorized into North and South Ethiosemitic. The South Ethiosemitic
varieties are spoken in the central, eastern, and southwestern parts of Ethiopia. This study
specifically focuses on the Guragina Languages, which are a group of languages spoken in the
southwestern region of Ethiopia.

9
In sociolinguistics and stylistics, the term "variety" refers to any distinct linguistic system. The
Gurage area is linguistically highly diverse, with different language varieties often further divided
into dialects and sub-dialects. This area is situated within the Ethiopian linguistic area, which is
known for its remarkable linguistic diversity [19]. Guragina belongs to the Semitic branch of the
Afroasiatic language family and comprises approximately thirteen varieties [22].

2.2.1. Variety of Guragina Language

Distinguishing between dialects and languages and quantifying the similarities between different
language varieties have been fundamental questions in dialectology. Typically, the measurement
of linguistic distance between two or more language varieties is used to approach these questions
[16]. In general, the greater the structural similarity between two languages (phonetic,
morphological, lexical, or syntactic), the closer they are considered to be. When the structural
similarities are significant, the varieties are often regarded as dialects of the same language [19].
However, the distinction between dialects and languages is more complex, and non-linguistic
factors (such as social, cultural, political, and psychological aspects) often play a crucial role. This
means that determining linguistic distance based solely on structural similarity may not be
sufficient to determine whether two varieties should be classified as dialects of the same language
or as separate languages [23].

Currently, the Gurage zone is administratively divided into thirteen districts: Ezha, Geto, Gumer,
K'abeena, Libido, Mesqan, Kistane, Abeshge, Chaha, Endegagn, Inor, Muhir-Aklil, and Wolane.
It's worth noting that the Gurage region is home to speakers of non-Semitic languages such as
K'abeena and Libido [14].

According to Menuta [14], the Gurage zone is further subdivided into four agroclimatic zones:
alpine (above 3200 meters), temperate (between 2300 and 3200 meters), subtropical (between
1500 and 2300 meters), and tropical (between 1100 and 1500 meters above sea level). The alpine
climate zone surrounds Gurage Mountain, also known as Zebidar, which reaches an altitude of
3600 meters. This mountain acts as a geographical barrier, dividing the Gurage area into east and
west and hindering inter-group contact between the people of Gurage East and Gurage West.
Linguistically, this mountain has been used as a reference point, with the Eastern Guragina
language referred to as "Oriental Gurage" and the Western Gurage language as "Occidental

10
Gurage" [28]. From a linguistic perspective, the sub-groups of Guragina, as described in Section
1.1, can be categorized as follows: North Gurage, which includes Kistane and Dobi; West Gurage,
which encompasses Meskan, Cheha, Ezha, Gumer, Muher, Inor, Endegagn, Enner, and Geto; and
East Gurage, which consists of Welane, Silte, and Zay. Due to the existence of these various
dialects and other related factors, developing NLP application, including a spell checker is
challenging.

A survey conducted by Menuta [14] shows that almost all Guragina speakers are bilinguals. Speakers of
the Guragina varieties have a frequent contact among each other. Many of them have contact with Chaha,
compared to other Guragina varieties. The majority of the participants believe that they understand Chaha
better than other Guragina varieties. They also attribute positive value to Chaha.

Gurage Mountain w a s mentioned as the main geographical barrier to contact among the
Guragina speakers. For instance, Mesqan and Kissane do not have a frequent contact with the
speakers of Chaha, Muher and Inor because of this barrier. Lack of public transport and hostility
among the Gurage clans are indicated as other obstacles [14]. In market places, Guragina is more
frequently used than Amharic. In schools, religious places and administration, Amharic is the
dominant language. According to this study, Amharic (63%) is more frequently used at home
than Guragina (42%).

The reorganization of the Gurage ethnic group and the standardization of linguistic varieties
continued to present challenges. The biggest challenge has been the ethnic diversity within the
Gurages, and the disagreements between the Gurages ethnic groups and the political elites about
language materials to be taken into account in the standardization process [20], [21].

In 2013, the Gurage Zone established a Language Board to promote and standardize Guragina,
and introduced an official Guragina script, and endorsed by Gurage zone administration council
in 2020. According to [24], the aim of improvement of this orthography is to make it simple,
pattern based and also to standardize it.

In the process of Guragina Language standardization, in the endorsed language of Gurage,

particularly Cheha plays a central role, along with incorporating certain vocabulary from other
language varieties. As a result, our detailed analysis primarily centers on the morphology of Cheha,
considering its significance in the standardization efforts.

11
2.2.2. Guragina Language Writing System
Guragina is written left-to-right, separated by white space, which uses Ethiopic script as a
subgroup of Ethiopian Semitic language. For over half a century the Gurgege language has used
the Amharic Fidel with some additional letters developed for sounds which are not found in
Amharic. The Guragina orthography underwent seven separate modernization attempts, mostly in
the palatalized and labialized graphemes, creating short "eras" in the orthography since 1950 [20].
These different writing systems were used to publish some fictional work in Guragina Language
as well as a New Testament, called Geder Gurda, and the complete Bible called Meta’af Qidus.
As the Guragina Language was not in past used in institutions such as media, education and
administration bureaus, the orthography was not officially recognized. Because of this, Guragina
variants are only used orally, with the exception of Silt'e, which has been written since 1982 and
is teaching in primary schools since 1995 [20].
Like the original Ethiopic script, the current Guragina script is an alpha syllabary with symbols
that mainly represent consonant-vowel sequences. On September 14, 2021, the Unicode Standard
version 14.0 was formally adopted, giving the Modern Guragina Orthography set's letter additions
official acceptance on a global scale [27]. We can read many books which are written in this
standardized Guragina language and its modern orthography.
As presented in Appendix 1, the complete modern Guragina character set contains 31 base
characters (consonant) vertically arranged in seven columns (vowels), which gives 217 characters,
and the other 8 characters vertically listed in 5 columns, which gives 40 characters. This shows
that, the Guragina syllabary, or “Fider”, comprises total of 257 syllographs. Even
though many of the letters or orthographies of Amharic and other Ge'ez scripts, there are some
variances. For instance, Guragina’s modern orthography gives new shapes to five velar syllables
that had previously been visually identical under the writing practices of other Ethiosemitic
languages (for example, ጔ ‘gʷ ə’ vs 𞟹 ‘gʷ ə’ …). Second, there is only one 'ha' (ሓ) in Guragina as
opposed to five in Amharic. The usage of words like “አ in place of ኧ, ኣ for አ, and ሐ in place of ኸ
“among others, as part of modern convention are another distinction. Not only that, but "and" also
አ and ዐ has the same sound in Amharic but not in Guragina. Last but not least, there are several
characters like ኸ ‘kʲə’, 𞟠 ‘hʲə’, ጘ ‘gʲə’, ᎄ ‘bʷ ə’, 𞟨 ‘hʷ ə’, ᎀ ‘mʷ ə’, ቈ ‘k’ʷə’, ᎈ ‘fʷə’,
ጐ ‘gʷə’, and ᎌ ‘pʷə’, that are exclusively present in Guragina and not in Amharic. In Guragina,

12
these letters are rather common. A designed version of the Garagina keyboard has just been
accessible for all popular operating systems [4]. The most recent Guragina script (seven orders of
consonants and vowels) was shown in Appendix 1, which is adopted from [20].

2.3. Morphology
Morphology is about the structure of words. Word is the key element in morphology. All languages
have words and in all languages some words, at least, have an internal structure, and consist of one
or more morphemes.

Morphemes are the smallest meaningful unit of the structure of language whereas the biggest
units are words. For instance, the word "cats" is made up of the root morpheme "cat" and the suffix
morpheme "s," which indicating plural.

Guragina is a Semitic language, and Semitic languages generally have complicated morphologies.
It shares linguistic roots with other Semitic languages like Hebrew, Arabic, Amharic, and Syrian.

2.3.1. Verb
A verb is the most crucial element of a sentence is the part that describes the subject of the clause,
expresses an action (cut, halt, search), an occurrence (snow, happen), a change (include, shrink,
dissolve, broaden), or a state of being (e.g., is, be, have). Clauses and sentences cannot be formed
from a set of words without a verb.

The Guragina Language is also characterized by the typical root-and-pattern morphology common
throughout all Semitic languages. The root-and-pattern morphology's fundamental idea is as
follows. First, there is a root, called a radical, comprises of a set of consonants with a basic lexical
meaning (core meaning of a word) and arranged in a particular order [7]. Next, there are various
regular patterns, also known as templates, that control the quality and placement of vowels in
relation to the radicals[10]. Radicals do not naturally appear as words; rather, they must be
incorporated into a template to serve as a base.

It is necessary to be familiar with the verbal systems of the Semitic languages in order to
understand how the root and pattern morphology evolved and persisted. The Semitic verb
morphology is described based on the complex non-concatenative/root-and-pattern system [25].

13
Guragina is one of these languages, and it describes the simple verb stems using a non-
concatenative/root-and-pattern morphological process [10]. According to [26], in many languages,
including Guragina, the verb shows more morphological complexity than any other word class.
Even it shows the most complex form in Guragina [19]. As shown in table 2.1, from Guragina root
word ቸነ/ ʧənə we can drive different word for different person.

Table 2. 1 Derivation of ʧənə for different person.

Example Phonetic Representation Suffix Translation

ቸነ-𞟫ም ʧənə-hʷɨm hʷɨm I came
ቸነ-𞟥ም/ቸነ-ሐም ʧənə-hʲɨm /ʧənə-həm hʲɨm/ həm You came
ቸነ-ችም/ቸነም ʧənə-ʧɨm / ʧənə-m ʧɨm / m s/he came
ቸነ-ነም ʧənə -nəm nəm we came
ቸነ-ማም/ቸነ-ቦም ʧənə-mam/ ʧənə- bom mam / bom they came

The verbs are can be derived from the simple verbs by affixes. The other method is derivation of
verb by interdigitating vowels into the C-pattern to create different verb stems, and the roots (root
radical) are used to make the inflectional verb conjugations.

In Guragina, verbs can range in length from one radical to four radicals [22]. Similar to tri-
consonantal root verbs, mono-radical verbs are also common, particularly in Guragina variants
[10]. For instance, the sentential verbs ː akjə “He chew.”, epə “He worked”, ela “He wished”, amə
“He gave.”,efə “He covered.” ʃə “He wanted" are realized as they have mono-radicals in their
surface structure [14].

Derived Stems
The verbal process of derivation illustrates how root morphemes change in the verbs' meaning in
relation to their lexical and semantic content. This morphological process involves change of
forms, increase number of root-morphemes, and change of syntactic category and semantic usage
[19]. In Guragina, the derivations of complex verb stems are the passive marker /tə-/, causative
marker /a/t-/, and verbal noun /-ot or wə-/ which are exemplified in Table 2.2.

14
 Table 2. 2 Derivations of Guragina complex verb stems with tə-

Root Base Translation tə-stem Translation

s-b-r stem
səpər - break tə-səpər- was broken
b-l- a bəna- eat tə-bəna- was eaten
a-n-t’ ant’- cut t-ant’ə- was cut

Furthermore, the bound morpheme /tə-/ is also used in reciprocal expressions. Here, the
penultimate consonant of the triliteral radical is repeated and in between the first radical, and
penultimate radical the low central vowel /a/ is inserted. Consequently, as exemplified in the in
the Table 2.3, they have the template slot tə-CəCaCəC- in the canonical triradical root morphemes.
For instance, the t-stem təsəpər- becomes təsəpapər-o “they broke each other”.

Table 2. 3 Reduplication of word with morpheme tə

Passive Translation Reduplicative stem Translation

tə-səpər- was broken tə-səpapər- break one another
tə-ʃəgər- was exchanged tə-ʃəgagər- exchange one another
tə-agəz- was helped tə-gagəz- help one another

Causative Stem
In Guragina, there are two types of causatives markers, direct causative and indirect or a double
causative. The intransitive stems are mainly where the direct causative is derived. It is derived with
/a-/. The passive and indirect causatives are both represented by the indirect causative and it
derived with /at-/. The examples in Table 2.4 demonstrate the derivation of causatives (direct and
indirect), in perfective and imperfective aspects, and the jussive mood.

15
 Table 2. 4 Derivation of Causatives stem in Perfective, Imperfective and Jussive words

Perfective Imperfective
Verb Translation Type
Past Present Future Jussive
wəndə Get down BASIC wənd-m jɨ -wərəd jɨ-wərd-te jə-wərd
DCAUS a-wənd-m a-wənd jɨ-wərd-te j-a-wərd
ICAUS at-wənd-m j-at-wərd j-at-wərd-te j-at- wərd
bəna Eat BASIC bə na-m jɨ-bə ra jɨ-bəra-te jə -bra
DCAUS a-bəna-m j-a-bəra j-a-bəra-te j-a-bra
ICAUS at-bəna-m j-at-bəna j-at-bəna-te j-at -bəra
səʧʼə Drink BASIC səʧʼə-m jɨ-səʧ’ jɨ-səʧ’-te jə-st’e
DCAUS a-səʧʼə-m j-a-səʧ’ j-a- səʧ’-te j-a-sʧ’
ICAUS at-səʧʼə-m j-at-səʧ’ j-at- səʧ’-te j-at-sɨʧ’
sɨjə Buy BASIC sɨjə -m jɨ-sɨjə jɨ-sɨjə -te jə -sə jə
DCAUS a- sɨjə-m j-a-sɨjə j-a-sɨjə-te j-a-səjə
ICAUS at- sɨjə-m j-at-sɨjə j-at-sɨjə-te j-at-səjə
kəna Climb BASIC kə na-m jɨ- kə ra jɨ- kə ra-te jə -kra
DCAUS a-kəna-m j-a- kəra j-a-kəra-te j-a-kra
ICAUS at-kəna-m j-at- kəra j-at-kəna-te j-at-kɨra

Verbal Noun
In Guragina, nouns are derived from the jussive stems morphologically. The root-morphemes take
the suffixes /-ot/ in order to change category from verb into nouns. In the canonical triradical root
consonants, the root morpheme and suffixes are linearly appeared in newly derived nouns except
the inserting of the epenthetic vowel to solve consonant constraint in the variety as can be seen
from the data in Table 2.5.

16
 Table 2. 5 Guragina Verbal nouns

Jussive stems Translation Infinitive Translation

bɨhʲ - cry bɨh-ot Crying
dɨrg- Hit d ɨrg-ot Hitting
k’wat’ɨr - tie k’wat’r-ot to tie / tying

Verb Negation
When a verb is negated, a negation marker is added before the conjugated verb. In Guragina, their
three verbal negative markers [10], [19], [27]. In order to do to accomplish this, the marker for the
all-perfective is /an-/, for the Imperfective and Jussive it is bound morpheme /a-/, and for the
prohibitive it is prefix /ɨn-/.

Negated Perfective
The suffixes of the perfective are different for each person. As seen in Table 2.6, the perfective is
negated by adding an- just before the base.

Table 2. 6 The negation of stem bəkər “miss”

Subject Singular Plular

1 an-bəkər-hʷ an- bəkər-nə
2m an- bəkər-hə an-bəkər-xu
2f an- bəkər-hʸ an-bəkər-xɨma
3m an- bəkər-ə an- bəkər-o
3f an-bəkər-tʃ an-bəkər-əma
IPS an- bəkwər -i)

Additionally, the negation in this inflectional verb conjugation is produced by the morpheme /an-
/, which is defined as being symmetric or linear[28]. We provide examples of this verbal negation
in a Table 2.7 for all perfective verb types.

17
 Table 2. 7 Perfective negations

Affirmative Negative
Guragina Meaning Guragina Meaning
sənəqə He stole. an- sənəqə He did not steal.
zəgədə He remembered an-zəgədə He did not remember.
.sefə He sewed. an-sefə He did not sew.
mesə He dug up. an-mesə He did not dig up.
zɨmamədə he Mixed-up an- zɨmamədə He did not Mixed-up
ʒɨpapərə He flipped an-ʒɨpapərə He did not flip
qʷəmə He stood an- qʷəmə He did not stand
qʷəqʷərə He squeezed an-qʷəqʷərə He did not squeeze.

This morpheme /an-/ is also combined with the subordinate morpheme /bə-/to create negation in
conditional (if clauses). When the subordinator and negation morphemes are morphologically
connected, the central mid vowel /ə/ is deleted, realizing the morpheme like from the word sətʃə
“drank” we can construct bə-an-sətʃ “If he did not drink”.

Imperfective Negation
The negative marker morpheme /a-/ is appended to all verb types in the imperfective aspectual verb
conjugations in Guragina [10], [19], [27] , as seen in Table 2.8. It is clear from the table that, negative
marker /a-/ is prefixed and attached with the subject marker morpheme / jɨ-/.

18
 Table 2. 8 Negated imperfective

Affirmative Negative
Guragina Meaning Guragina Meaning
jɨ- sətʃə He drunk. a-jɨ-sətʃə He does not drink.
jɨ-kʷəm He stands up. a-jɨ-kʷəm He does not stand.
jɨ- met’ɨr- He differentiates a- jɨ- met’ɨr- He does not differentiate.
jɨ-banr He demolishes a-jɨ-banr He does not demolish.
jɨ-qat’r He ties a-jɨ-qat’ɨr He does not tie.
jɨ-qʷənɨs He reduces a-jɨ-qʷnɨs He does not reduce.
jɨ-qʷəlf He locks a-jɨ-qʷəlf He does not lock.

Negated imperative and jussive

The other negative marker used by Guragina, are at-, e, ɨn. In which, giving command or the
imperative verbal negation marked by the morpheme /a-/, the jussive verbal negation marked by
the morpheme /e-/, as shown in Table 2.9, and prohibitions, or the negative command marked by
the morpheme /ɨn-/, as shown in Table 2.10.

Table 2. 9 Negated imperfective and jussive

Affirmative Negation
Guragina Meaning Guragina Meaning
sɨt’e- drink! at- sɨt’e- Don’t drink!
bɨra You eat! at-bɨra Don’t break!

Additionally, in many south Guragina the prohibition or the negative command is expressed by
the morpheme /ɨn/ interchangeably with /a/[19], [23] without significant semantic difference, just
like to Amharic atbɨla vs. ɨndatbɨla. The information in Table 2.10 offers an illustration of this.

Table 2. 10 Negative Command with morpheme /ɨn-/ and /at-/

Negated with/ ɨn-/ Negated with /at-/

Guragina Transliteration Translation Guragina
Transliteratio Translation
እን-በና-ሐ ɨn-bəna-hu You shouldn't eat አ-ት-ብር-ኦ n
a-t-br-o (3P) You don't eat
እን-ትማር-ሐ ɨn-tɨmarɨ-hə You shouldn't አ-ት-ማር a-t-mar (3S) You don't learn
እን-ሰነፍ-ሐ ɨn-sənəf-hə learn!
You shouldn't አ-ት-ስረፍ a-t-sənəf (3S) You don't scare
scare!

19
 Guragina uses the morpheme /e-/ in jussive to morphologically inflect negation, as shown in the Table
2.11. The jussive aspectual verb conjugation prefixes this negative marker morpheme to all verb
types[19], [29] .

Table 2. 11 Negation with morpheme /e-/

Affirmative Negation
Guragin Transliteration Translation Guragina Transliteration Translation
a
አ-ስብር ə-sɨbr Let him brake. ኤ-ስብር e-sɨbr Do not let him open.
አ-𞟨ር ə-hwər Let him go. ኤ-𞟨ር e- hwər
3MS.NEG- Do not let him go.
አ-ቶራ ə-tora Let him sit ኤ-አ-ቶራ open.JUSS
e-tora Do not let him sit.
3MS.NEG-
Tense and mood open.JUSS
3MS.NEG-
Tense open.JUSS
Tense is a verb-based approach for expressing the time of an action or state in relation to the
moment of speaking, as well as occasionally its continuance or completion. The three basic tenses
are present, past, and future.

In Guragina, tense is morphologically represented by the auxiliary verbs (banə, enə, nərə, and
tanə) and the bound morphemes -te and -ʃə, and it is classified into three basic distinctions as ːpast
present and future respectively.

Guragina makes a distinction between the three verb bases that make up the three fundamental TAM
forms: Perfective, Imperfective, and Jussive (including Imperative). The imperfective and jussive
verbal forms have future time readings when they are suffixed with the bound morphemes -te and -
ʃə. The perfective and jussive verbal forms denote an action that occurred at the moment of speaking,
which indicates the present. The auxiliary verb banə or -ba “was” can be added to the first two to
indicate past tense, and the -te- and -ʃə suffixes can be added to the last two to indicate alternative
future tenses.

20
Present
The imperfective stem of the primary verb is inflected in Guragina to reflect the present reading,
which is not clearly indicated. Additionally, as shown in example the locative clauses and
predicates use the existence/auxiliary verb nəre, “there is” to indicate present tense readings.

Example 3 E kʷa aʒəmənə nere

አኳ ኣዠመነ ነረ
There is a wedding today.

Huta bɨzə fɨrank nere-n

ሑታ ብዘ ፍራንክ ነረ-ን
He has a lot of money.

ɨja-m agazɨ-m shewa Nane

እያ-ም ኣጋዝ-ም ሸዋ ናነ
I'm traveling to Addis Ababa with Agaz.

Past
Guragina has two morphological descriptions for past tense. The independent auxiliary verbs banə
("there was") and tanə ("when it was") are used to express past readings, and the inflected perfective
verbal forms are required to indicate past reading. For example, zega banə “He was poor, zega tane
“When he was poor” etc. The adjuncts (adverbs) are attached for further temporal time reference; the
perfective verbal forms do not carry a further constraint of verbal activity. səpʷər-ə-n-m “he
brokem”, dənəgʷ-ə-wi-m, bəna-m, bəna-tʃɨ-m etc. This auxiliary verb (banə) is realized as its full
forms or it can be reduced into ba that is interpreted as “there was”. i.e., banə and ba can be used
interchangeably.

Future
In Guragina, the bound morphemes -te or-ʃə are concatenated to the aspectual imperfective verb
forms for the readings of the definite and indefinite futures, respectively, to denote the futurity of
an action. The negative imperative stem of the verbs is denoting the future negation
morphologically and syntactically as: e-sətʃ-ɨ-n, which is translated as "He is not drinking," "He
does not drink," or "He will not drink." However, these future markers are not attached to the
negative imperfective verbal forms. All imperfective aspectual verb forms have the morpheme -te
added, which is used to distinguish the predetermined future under external control. In order to
21
convey future readings in this variety, it must be concatenated into the imperfective verb
conjugations of all verb kinds.

Mood
The morphological or grammatical method of modality expression is mood [27]. It relates to ideas
like “possibility,” “necessity,” “permission,” and “obligation,” among others. In Guragina, verbs
are grammaticalized for expressing the attitude of speakers. The modality of the verbs may have
different realizations that are categorized into types such as agent-oriented, speaker-oriented,
epistemic, and subordinating [23]. Thus, each of these types has their own sub types and there may
be an intertwining of function among these types.

In Agent-oriented modality, there are number of sub-classes including obligation, necessity, ability
and desire. In Guragina, the essence and moral duty for ones or others life is morphologically
distinctive, and they are expressed by the auxiliary/existence verbs nərə-/banə- and the applicative
(object malfactive) marker -b-. The morphological representation of moral obligation used by
speakers of Guragina is shown in Table 2.12. The existence verb nərə- together with the malifactive
marker /-b-/ are employed to define state of obligation. The malfactive marker is followed by the
object pronominal suffixes in this morphological process.

Table 2. 12 Different modality of obligation

Obligation
Subject Singular Plural
Guragina Transliteration Guragina Transliteration
1 ነረ-ብ-ኢ nərə-b-I ነረ-ብ-ንደ nərə-b-ndə
2m ነረ-ብ-ሐ nərə-b-hə ነረ-ብ-ሑ nərə-b-hu
2f ነረ-ብ-𞟥 nərə-b-hʲ ነረ-ብ-ሕማ nərə-b-hɨma
3m ነረ-ው-አ nərə-w-ə ነረ-ብ-ኦ nərə-b-o
3f ነረ-ብ-ኣ nərə-b-a ነረ-ብ-አማ nərə-b-əma

2.3.2. Noun
Nouns are words (not pronouns), according to [30], that can serve as the subject or object of a verb
or the object of a preposition. Simple nouns the simple noun forms are formed by the root-and-
pattern morphological system whereas derivative or morphologically generated nouns are referred

22
to as complex nouns. i.e., complex nouns are derived through affixes, and two or more lexemes
are compounded to form a new semantically unitary concept.

Guragina nouns have both simple and complex forms [31]. They are inflected for number, gender,
definiteness, and case. The discussion of each of these ideas is broken down below into their
individual instances.

Inflection of Noun
Number
The plural nouns of Guragina marked by two different morphological inflections; internal vowel
modification and substitution for kinship words [31], as demonstrated by Table 2.13.

Table 2. 13 Different morphological inflections

Singular Plural
Guragina Transliteration Translation Guragina Transliteration Translation
አርች ərtʃ Boy ደንጛ dəngja Boys
ምስ mɨss man ገማ/ ገመያ gəma/gemja Men
አራም əram Cow አሬ əre Cows
ገረድ gərəd Girl ግሬድ gred Girls
ምሽት mɨʃt Wife እሽታ ɨʃta Wives

The majority of plural numbers, however, are expressed indirectly on verbs. As shown in Table
2.14, they are also indicated by adding the third person pronoun (hɨno with all nouns and hɨnəma
for human (Feminine) related nouns) that interacts with definiteness as demonstrated in example.
Additionally, third person singular male -huta for all noun types and third person feminine -
hʲita for third person singular are both acceptable.

Table 2. 14 Examples of Guragina’s Plural with third person pronoun

Example With Suffix Suffix

Guragina Transliter Transla Guragina Transliteration word-for-word Translati Gura Transliter
ation tion translation on gina ation

እምር ɨmɨr Stone እምር-ሕኖ ɨmɨr-hɨno stone-they Stones ሕኖ hɨno

23
 ወናድ wənad Mare ወናድ-ሕኖ wənad-hɨno mare-they Mares ሕኖ hɨno
ፌቅ Feq goat ፌቅ-ሕኖ feq-hɨno goat- they goats ሕኖ hɨno
ጀበን dʒəbən coffee ጀበን-ሕኖ dʒəbən-hɨno coffee pot- coffee ሕኖ hɨno
pot they pots

In Guragina, feminine and masculine genders are indicated via internal modifications, and these
nouns are mainly representing kinship terms and domestic animals [31]. Kinship word and
domestic animals both lexically denote the masculine and feminine genders, as shown in example
Table 2.15.

Table 2. 15 Gender in Guragina

Masculine Feminine
Guragina Translit Translation Guragina Transliter Translation
eration ation
ምስ mɨss husband ምሽት mɨʃt wife
አርች ərtʃ Boy ገረድ gərəd girl
ኣባ aba Father ኣዶት adot mother
ውር were Ox አራም əram cow
ፈረዝ fərəz horse ወናድ wənad mare

Additionally, to distinguish between animals of the feminine and masculine genders, the adjectives
arɨst “female” and təbat “male” are employed [29]. Before the nouns, these words are combined
like in Table 2.16.

Table 2. 16 Guragina’s Gender with “arɨst” and “təbat”

Generic Masculine Feminine

Gura Translit Transl Guragina Transliterat Translation Guragina Transliterati Translation
gina eration ation ion on
ፌቅ feq Goat ተባት ፌቅ təbat feq male goat ኣርስት ፌቅ arɨst feq nanny goat

ግየ gɨjə Dog ተባት ግየ təbat gɨje male dog ኣርስት ግየ arɨst gɨje bitch

24
 ᎋር fwur Rat ተባት ᎋር təbat fwur male rat ኣርስት ᎋር arɨst fwur female rat

Definiteness
In Guragina, referential restriction of nouns can be categorized into two i.e., definite and indefinite
nouns. Indefinite nouns are referring to the generic word that has not been restricted its function
as in feq goat”. This word is not showing an individual goat but a generic one /any goat; it can be
male, female and unknown [27].

In Guragina, definiteness of nouns is indirectly marked, and they are directly denoted by person
pronouns of both feminine and masculine predominantly the third person as shown Table 2.17.
The former is suffixed to the semantically feminine nouns (only people) and in animate nouns for
diminutive case, and the latter is suffixed with all masculine nouns.

Table 2. 17 Definiteness of nouns

Generic Generic with Suffix

Guragina Transliteration Translat Guragina Transliteration Translation
ገረድ gərəd ion
girl ገረድ-𞟥ታ gərəd-hjita the girl
አርች ərtʃ boy አርች-𞟫ታ ərtʃ-huta the boy
እማር ɨmar donkey እማር-𞟫ታ ɨmar-huta the donkey

Therefore, in Guragina, definiteness is indirectly indicated by the third person masculine

(huta) although the third person feminine singular is used for singular feminine nouns. The
morpheme huta denotes definiteness in all animate and inanimate singular nouns. Table 2.18
shows the grammatical function of the morphemes hʲita and huta in Guragina.

Table 2. 18 Definiteness with hʲita and huta

Grammatical Translation
Guragina Transliteration
ግሬድ-ሕነማ gred-hɨnəma girl.PL-DEF the girls
አርች-𞟫ታ ərtʃ-huta boy-DEF the boy
ደንጛ-ሕኖ dəngʲa-hɨno boy.PL-DEF the boys
Therefore, definiteness in this variety is denoted by the third person singular and plural pronouns
but the third person masculine singular is commonly used for all animate and inanimate nouns.

25
Case
The relationship among inflectional markers and independent nouns has grammatical and
semantic notion. Thus, the syntactic relations of words are referred to as core case whereas the
semantic relations are treated as peripheral case. The grammatical case encompasses subjective
(nominative), objective (accusative), dative (indirect object) and genitive (possessive), and the
semantic case contains locative, ablative, and vocative among other case markers. In Guragina,
both the grammatical and semantic cases are discussed as follows.

Noun derivation
Nouns are also derived from the composition of two or more lexemes that creates a single semantic
concept [31]. However, most of Gurarage nouns are resulted from derivational process by affixes
such as -nət, -i, -ot /wə-, -ənə, -na, and -kʷə. For instance, several adjectives and nouns in Guragina
can be changed into abstract nouns by adding the bound suffix (morpheme) -nət, as listed in table
2.19. In this case if the last latter of the word (before adding -nət) is 4th column latter in Guragina
syllograph, it changes to the 6th latter of its type, like “ጋዋ” to “ጋውነት” or Gawa change to Gaw-
nət.

26
 Table 2. 19 Derivations of Noun

Adjective or noun Abstract noun

Guragina Transliteration Translation Guragina Transliteration Translation
ወሔ wəhe kind ወሔ-ነት wəhe-nət Kindness
ሓሪ Hari Wise ሓሪ-ነት hari-nət Wisdom
ጋዋ Gawa Fool ጋው-ነት gaw-nət Foolish
ሙጥጥ mut’ɨt’ Bad ሙጥጥ-ነት mut’ɨt’-nət Badness
ጌፍ Gef Long ጌፍ-ነት gef-nət Longines
ጀንጅር dʒenedʒer Fat ጀንጅር-ነት dʒenedʒer-nət Fatness
ደንገነ dəngənə Rich ደንገነ-ነት dəngən-nət Richness
ዜጋ Zega Poor ዜግ-ነት zeg-nət Poverty
ቤሻ beʃa Friend ቤሻ-ነት beʃ-nət Friendship
ባሪቅ barik’ Old ባሪቅ-ነት barik’-nət Old age
ወነ𞟨 wənəhwə neighbor ወነ𞟫-ነት fwɨk’wɨnn
ənəh w ət
-nət Neighborhood
ፈንገያ fɨk’wɨr
fəngʲa ‗fat‘
Thief ፈንገያ-ነት fəngʲɨ-nət ‗fatness‘
Theft
ትከ tɨkə Child ትክ-ነት tɨkɨ-nət Childhood
ሰብ səb Man ሰብ-ነት səbɨ-nət Personality

Compound Nouns

Nouns in Guragina are also morphologically created by combining two or more simple nouns, which
are then regarded as a single unified semantic word. Two or more words that relate to a single item,
situation, action, or event are combined. Guragina (Ethiopic scripts), transliteration, and the
translation of each individual word and compound word are all shown in Table 2.20, respectively.

27
 Table 2. 20 Guragina Compound word

Single word Compound word (noun)

Guragina Transliteration Direct Guragina Transliter Translation
translation of ation
each word
ውር ኣጤበ wər at’ebə ox finger ውርኣጤበ wərat’ebə Thumb
ውር ኩታራ wər kutara ox hen ውርኩታራ wər kutara Roaster
ወሐመያ wəhɛ məja good path ወሐመያ wəhɛməja Holiday
የቤት እም jəbet em House mother የቤትእም Jəbetem Wife
ሓኪም ቤት hakim bet doctor house ሓኪምቤት Hakimbet hospital or clinic
ጉነር ፍርጠት gunər fɨrt’ət head spilt ጉነርፍርጠት gunərfɨrt’ət headache
አዳር ሑጅር ɨdar hudʒɨr night cloth አዳርሑጅር ɨdarhudʒɨr Blanket
In addition to that, in this variety, several compound nouns are formed by the combination of ab
“ኣብ” with a simple noun to create a semantically single word, such as ኣበቃጥ “husband”, ኣበሩስ
“family”, ኣበጥብ “same clan”, ኣበጐዳ “friend”, etc…

Root-and-Pattern (Noun Formation)

Guragina uses a root-and-pattern construction for basic nouns, which is a feature of Semitic
morphology exclusively. Here, as one can understand from the Table 2.21, nouns are formed by
combining different vocalic and consonantal affixes into roots with two, three, or four
consonants.

28
 Table 2. 21 Guragina’s root-pattern nouns

Template Root Noun Translation

Guragina Transliteration Guragina Transliteration
CCə ብ-𞟥 b-kʲ ብኸ bɨkʲə mourn
ብ-ኽ s-kʲ ስኸ sɨkʲə run away
COC ም-ት m-t ሞት Mot death
ጥ-ም t’-m ጦም t’om Fast
CəC ድ-ም d-m ደም dəm blood
ብ-ር b-r በር bər river
ᎋ-ስ f ʷ-s ᎈስ Fʷəs Fart
CɨCC ም-ክ-ር m-k-r ምክር mɨkr advice
ድ-ም-ድ d-m-d ድᎃድ dɨmʷd pair
ቅ-ን-ም k-n-m ቅኔመ kɨnemə insult
CəCəC ስ-ን-ፍ s-n-f ሰነፍ sənəf laziness
ድ-ር-ግ d-r-g ደረግ dərəg strength
𞟺-ብ-ዝ g ʷ-b-m ጐበዝ gʷəbəz cleaver
CɨCaC ን-ፍ-ስ n-f-s ንፋስ nɨfas wind
ስ-ክ-ር s-k-r ስካር sɨkar intoxication
ቅ-ም-ር q-m-r ቅማር qɨmar Louse
CɨCCəCa ም-ዝ-ግ-ብ m-z-g-b ምዝገባ mɨzgəb registration
ስ-ብ-ስ-ብ s-b-s-b ስብሰባ sɨbsəba meeting
ም-ር-ም-ር m-r-m-r ምርመራ mɨrməra investigation
CɨCCɨC ም-ስ-ክ-ር m-s-k-r ምስክር mɨskɨr witness
ስ-ን-ጥ-ር s-n-t’-r ስንጥር sɨnt’ɨr splinter
CəCCəC ስ-ን-ብ-ር s-n-b-r ሰንበር sənbər Stripe of stomach

To determine the kind of words in the context of root consonants, the vowel's quality is crucial. For
example, vocalic alteration in the stem template can change the word classes of the terms bɨkʲə “ብኸ”
is noun and mean crying or sorrowful while bəkʲə “ በ ኸ ” He cried. ‟ mʷətə He died. Respectively.
Here, the difference lies on the vowels that are inserted in the radical consonantal roots.

2.3.3. Adjective
Adjectives in Guragigna, despite being a few in number, share morphological and syntactic
similarities with nouns and pronouns. They undergo morphological inflection for number, case,
and definiteness, but not for gender. While singular number is not explicitly marked, plural
numbers are expressed through morphological manifestations. Plural adjectives are formed by

29
reduplicating stems, for example, ብሻብሻ "bɨʃabɨʃa" meaning "red ones" Alternatively, pronouns,
particularly the third person masculine plural suffix "-xɨno," are commonly used to indirectly
denote plural adjectives, such as ብሻሕኖ "bɨʃaxɨno" meaning "red ones" Additionally, adjectives
can be derived morphologically from other basic parts of speech using affixes like “jə-,” "ɨ…jə,"
‘-əma,” "-ənə," and "-w.

2.3.4. Pronouns
In their morphological forms, pronouns can be both simple and complex. They serve the same
purpose as nouns or noun phrases. Specifically, Personal pronouns are used interchangeably with
names of people, places, and things in Guragina, whether in spoken or written language. Unlike other
Ethio-semitic languages like Amharic, the feminine second- and third-person plural is distinct. The
fundamental personal pronouns included in Table 2.22 fall within this variety.

Table 2. 22 Basic Guragina Personal Pronouns

Person Singular Plural

Guragina Transliteration Translation Guragina Transliteration Translation
1 እያ ɨja I ይና jɨna we
2M ኣሐ ahə You Translation ahu
ኣሑ you
2F ኣ𞟥 Ahj You ኣሕማ ahɨma you
3M ሑት/ታ hut/a He ሕኖ hɨno they
3F 𞟥ት/ታ hjit/a She ሕነማ hɨnəma they

On the other hand, the genitive case marker (ʔə-) is morphologically attached to the independent
pronouns for denoting objective (accusative) pronoun. This is demonstrated in Table 2.23. These
complex objective pronouns alter the way nouns behave morphologically. For example, the
generic noun kutara, "hen," might be transformed with these genitives (possessions) pronouns
as ʔəhuta kutara "a hen of him" or pronominal possessive markers, "kutarata, "his hen".

Table 2. 23 Plural and Singular Objective (accusative) pronoun with ʔə

Singular Plural
Guragina Transliteration Translation Guragina Transliteration Translation
ይያ ʔə +ɨja ʔɨja me ይና ʔə +jɨna ʔina us
ያሕ ʔə +ahə ʔahə you ያሑ ʔə+ ahu ʔahu you
ያ𞟥 ʔə+ ahʲ ʔahʲ you ያሕማ ʔə +ahɨma ʔahɨma you
የሑት/ታ ʔə +huta ʔəhut/ta him የሕኖ ʔə +hɨno ʔəhɨno them
የ𞟥ት/ታ ʔə +hʲita ʔəhʲit/ta her የሕነማ ʔə +hɨnəma ʔəhɨnəma them
30
As one can understand from the Table 2.24, similar to the independent personal pronouns,
Guragina also defines possessive pronominal suffixes for person, gender, and number.

Table 2. 24 Pronominal suffixes of possession

Person Pronominal suffixes Example

Guragina Translitera Guragina Transliteration Translation
tion
1PS (አ)ና -(ə)na ኣባ-ና aba-na my father
1PL (አ)ንዳ -(ə)ndə ኣባ-ንዳ aba-nd
mother-ə our father
2PMS ሐ -hə ኣባ-ሐ 1PS.POSS
aba-h
father-ə your father
2PMPL (አ)ኣሑ -(a)hu ኣባ-ኣሑ 1PL.POSS
aba-hu
father- your father
2MS.POSS
2FS (ኣ)𞟥 -(a)hj ኣባ-𞟥 father-j
aba-h your father
-(a)hɨma 2MPL.POSS
aba-h
2FPL (ኣ)ሕማ ኣባ-ሕማ father-ɨma your father
3MS (አ)ታ -(ə)ta ኣባ-ታ aba-ta
2FS.POSS
father- his father
3MPL ሕኖ -hɨno ኣባ-ሕኖ 2FPL.POSS
aba-h
father-ɨno their father
3MS.POSS
3FS 𞟥ታ -hjɨta ኣባ-𞟥ታ aba-hjɨta
father- her father
-hɨnəma 3MPL.POSS
aba-h
3FPL ሕነማ ኣባ-ሕነማ father-ɨnəma their father
1PS (አ)ና -(ə)na አዶተ-ና adot- əna
3FS.POSS
father- my mother
1PL (አ)ንዳ -(ə)ndə አዶተ-ንዳ FPL.POSS
adot- əndə
mother- our mother
2PMS ሐ -hə አዶተ-ሐ 1PS.POSS
adot-ah
mother-ə your mother
2PMPL (አ)ኣሑ -(a)hu አዶተ-ኣሑ 1PL.POSS
adot-ahu
mother- your mother
2MS.POSS
2FS (ኣ)𞟥 -(a)hj አዶተ-𞟥 mother-j
adot-ah your mother
-(a)hɨma 2MPL.POSS
adot-ah
2FPL (ኣ)ሕማ አዶተ-ሕማ mother-ɨma your mother
3MS (አ)ታ -(ə)ta አዶተ-ታ adot- əta
2FS.POSS
mother- his mother
3MPL ሕኖ -hɨno አዶተ-ሕኖ 2FPL.POSS
adot- əhɨno
mother- their mother
3MS.POSS
3FS 𞟥ታ -hjɨta አዶተ-𞟥ታ əhjɨta
mother-
adot- her mother
3FPL ሕነማ -hɨnəma አዶተ-ሕነማ 3MPL.POSS
adot- əhɨnəma their mother
mother-
3FS.POSS
mother-
FPL.POSS
Guragina nouns are given the pronominal possessive affixes in order to display possession forms
similar to English possessive adjectives. The vowel (a) and low central vowel (ə), as seen in the
Table 2.30, are added between nouns and possessive markers. Conversely, if the vowel-final word

31
is joined with vowel-initial words, as in the lexeme’s aba “father” with lexeme na gives abana
“my father”, abanda “our father” etc the first vowels of possessive maker is eliminated.

2.3.5. Number
A number is represented by a numeral, which can be a single symbol or a group of symbols. These
lexical numerals are known as cardinal and ordinal numbers and are used to enumerate things, objects,
etc. In Guragina does not mark singular numbers, however plural numbers are morphologically
expressed.

Cardinal numbers are indicated by the arithmetic sets that identify numerical system of quantity.
Thus, Guragina has decimal numerical system, that is, the class of numbers are expressed in
counting system that they use unit of ten (10) as a base in the multi-digital numeral systems. it is
similar to Amharic numbering system with a few pronunciations difference. From the Appendix
2, it is clear that, in most Guragina varieties, numbers are pronounced as follows: at “ኣት”, hwet,
“𞟪ት” sost, “ሶስት”, arbət “ኣርበት”, amɨst “ኣምስት”, sɨdɨst “ስድስት”, səbat “ሰባት”,
sɨmwɨt “ስᎃት”, ʒət’ə “ዠጠ” and asɨr “ኣስር”. From eleven to nineteen we add
asɨrəm “ኣስረም” as the prefix of each, such as asɨrəm at “ኣስረም ኣት”, asɨrəm hwet,
“ኣስረም 𞟪ት” and asɨrəm ʒət’ə “ኣስረም ዠጠ”.

As one can understand from the Appendix 3, the majority of Guragina varieties, including
Guragina morphologically derive ordinal numbers from cardinal numbers by appending the suffix-
ənə.

2.3.6. Adverbs
Adverbs are one of the grammatical groups of words that provide information about the where,
when, how, how often, and in what way actions of verbs are carried out. Adverbs are less common
in Guragina than other content terms, which are introduced by the relational morphemes such tə-
and bə-. Adverbs are restricted with adverbs that are classified as time, frequency, manner, place
and frequency according to their purposes. These adverbs and their uses in Guragina are covered
in the section below.

32
Adverb of time
The role of the verbs in relation to past, present, and future time is explicitly constrained by
these adverbs. In Guragina, adverbs of time are lexically expressed that are suffixed by particles
like /-ra/ and /-ə/ to denote past and future expressions respectively [14]. This reality in Guragina
is illustrated in Table 2.25.
Table 2. 25 Guragina’s Adverbs of time

Present Past Future

Gura Translit Translat Guragi Transliter Translation Gura Translit Translation
gina eration ion na ation gina eration
አሗ əhwa now ትራማ tɨrama yesterday ነገ nəgə Tomorrow
አኳ ʔəkʷa today ሰስትራ səstəra three days before ሰስተ səstə three days later
ገባት gəbat evening ነበትራ nəbətəra four days before ነባተ nəbatə four days later
The time adverbs tend to form compound nouns to describe function of the action or clause
with nouns or words such as ː qɨrərə “morning”, wana “noon”, ədar “night”, kərə “day”.

Adverb of frequency

Frequency adverb expresses how often an action is happened in relation to time. In Guragina, as
can be realized in Table 2.26, adverbs of frequency are expressed by complete reduplication of an
adverb of time, but the adverb əhʷa “now” is attached with the coordinating conjunction (-m) [32].

Table 2. 26 Guragina’s Adverbs of frequency

Guragina Transliteration Literal translation Translation

አኳ-ም-ነገ-ም əkʷa-m-nəge-m today and today Everyday
አሗ-ም-አሗ-ም əhʷa-m-əhʷa-m now and now Always /every time
ቅረረ ቅረረ qɨrərə qɨrərə morning morning Every morning
የሕርም ሕርም jəhɨrm hɨrm year year Every year
ዋና ዋና wana wana noon noon Every noon
እንም ግዝየ / ɨnɨm gɨzjə or a l l times or all day Always /often
እንም ከረ ɨnɨm kərə
ኣታት ግዝየ atat gɨzjə One one time Some times

33
Adverb of manner
In Guragina, an adverb of manner is described by the relational prefix bə-, and it is linearly
attached to simple or complex nouns. Thus, as shown in Table 2.27, PPs morphosemantically
denotes manner of the verbs and they are headed by the relational morpheme (bə-).
Table 2. 27 Adverbs of manner

Guragina Transliteration Translation

በ-ሩጫ bə-rutʃa By running
ተ𞟠ናሐ tə-hjənahe Carefully
ቦሔነት Bo-wəhenət In good manner

Adverb of place
Adverbs of place are tending to show us where something is performed, placed, or its
whereabouts. In Guragina, the function of an adverb of place is denoted by addpositions.
Therefore, the prepositional morphemes in Table 2.28 represent the position or direction of
object, person and thing that is talked about.

Table 2. 28 Adverb of Place

Guragina Transliteration Translation Transliteration Guragin Translation

አግሬ əgɨre Bottom dən aደን inside/under
𞟲ሪት q ərit
w
Top wəsɨt’ ወስጥ Bottom
ነን nən Up tət ተት Low
ኣንቐ/ኣንቔ anqʲə / anqʲe behind/back jɨft ይፍት in front of
መየ / መዬ məjə / məje around fwər ᎈር On

2.3.7. Conjunctions
Connectors syntactically attach the jumbling of any two phrases, clauses or sentences. In
Guragina, two or more words are connected with the coordinator (-m) and disjunctive wem, like
ərtʃ-m gərəd-m for “boy and girl, mɨs wem mɨʃɨt for “husband or wife” but the hypotactic concept
is expressed by the relational morphemes tə- and bə- and ʔə like bə-tʃən etc...

34
2.4. Spelling Errors
Texts of languages can be generated from different sources either by humans as document typing
or emailing software, or by machines such as optical character recognition (OCR) and machine
translation (MT). In natural language processing, spelling error and spell checking is a very
common task in those generated text. A spell checker (or spelling checker or spell check) is a
computer program that checks for misspellings made by users who type in different applications.
Currently, spelling error detection play a significant role in a variety of computer programs [2] [7],
[33]. They could be used independently or embedded within other systems like word processors,
optical character recognition, search engines, speech recognition, machine translation, browsers,
and so on.

2.4.1. Types of Spelling Errors

Spell checker techniques are designed on the basis of spelling errors. Studies have been
conducted to analyze the types and trends of spelling errors. According to kukich [3], spelling
errors are generally categorized into typographic and cognitive errors based on the error patterns
produced when typing words on a word processing application. The other researcher, Gupta [34]
categorized as typographic, cognitive and phonetic errors

Typographic errors occur when the correct spellings of the word are known, but the word is
mistyped by mistake. The majority of the time, they are caused by technological limitations of the
input device (physical or virtual keyboard, or OCR system), like keyboard adjacencies, therefore
do not follow any language criteria. For instance, when typing quickly, two close keys are
frequently substituted. Although they may be dependent on local keyboard mapping or a localized
OCR system[35].

Typographic errors can also be classified as a Single errors or Multi-errors [36]. A word is
considered to have several errors if there are more than one. On the other hand, single errors are
affecting just one letter. A study by Damerau [4] shows that 80% of typographical errors come
from one of the following four categories.

Insertion: - inserting an extra character or single letter in the word.

Deletion: - missing or omission of a single letter a single character from the word.
Substitution: - replacement of a single letter by another single letter.

35
 Transposition of letters: - mis ordering or swapping of two adjacent letters in the word.
Typographic errors can also be categorized into two based on the meaning they give: non-word
and Real-word errors [5]. Non-word errors are spelling errors that do not give any meaning, while
Real-word (semantic) errors are morphologically valid words that give no sense in context.
Damerau [15] shows that 80% of the misspelled words in English are non-word errors. Table
2.29 shows some examples of typographical errors discussed so far.

Table 2. 29 Typographic Errors

Examples of different spelling errors

Language spelling Single Single letter Single letter Transposition of
letter deletion substitution two adjacent letters
English Correct happy
insertion happy happy happy

Incorrect hagppy Hapy hagpy hpapy

Amharic Correct አለን አለን አለን አለን

Incorrect ደስ አለለን ደስ አን ደስ አበን ደስ አንለ
Guragina Correct ሳረንደም ሳረንደም ሳረንደም ሳረንደም
Incorrect ሳለረንደም ሳረደም ሳረብደም ሳንረደም

The other types of errors are cognitive errors, which are also called orthographic or consistent or
real-word errors[35]. These mistakes occur due to misconceptions or when the correct spelling of
words is unknown to the writer [3],[35]. In other words, the writer may not be aware of the proper
writing style.

This particular type of errors is user- and language-specific because it is more dependent on using
the rules of the language. An example of this types of error would be typing “hiar” instead of
“hair”, “ante” instead of “ant”.

In texts written in the Guragina language, cognitive mistake is common. The primary cause is that
the language's orthography wasn’t standardized until 2020, as indicated in Section 2.5. Even once
the language is standardized in 2020, the majority of the orthography is unfamiliar to the writer.
Examples of this kind of error in this language include typing “ብርሕማ” berehema “instead of

36
“ᎇርሕማ” bʷerehema, “ዘፒትሑ” zəpitɨhu, instead of “ዘፒት𞟫” zəpitɨhʷ “እሐ” ’ɨhə instead
of “እሓ” ɨha, “ቆቆሳ” k’ok’osa, instead of “ቈቈሳ” k’ʷe k’ʷesa, etc.

Phonetic Errors happen when a character or a word is replaced out with a phonetically equivalent
sequence of character or homophone word, respectively. An example of this types of error would
be typing “cheap” instead of “cheep”, “ant” instead of “aunt” for English and typing ሐሐብ “ həhəbɨ
instead of “𞟨𞟨ብ” hʷəhʷəbɨ , “ሐብት” həbɨtɨ instead of “ኸብት” kʲəbɨtɨ “ኰር”kʷərɨ instead of
“ቈር” k’ʷərɨ. Similar to cognitive error, these types of error also common in Guragina language.

2.4.2. Techniques of Spell Checker

Spelling error detection and correction are the two main steps of a spell checker system,
which may be carried out jointly or sparely. Error correction step gives the proper suggestions
for any misspelled words and ranks them according to which terms in the list have the highest
likelihood of being used instead. Different approaches are used to develop SC for different
languages based on the characteristics of the language [6].

Error Detection Techniques

Spelling error detection approaches is the main thing in spell checking to detect the errors in
written text. It is used to flag out misspelled words in a text. To suggest a set of possible corrections,
a spellchecker has to identify the first misspelled words. Different approaches are applied to detect
spelling errors for different language based on their language characteristics. N-gram analysis and
dictionary lookup are the two most well-known methods for detecting errors in spell checker
system development[3], [7], [8].

N-gram Analysis

N-gram analysis is one of the popular methods for detecting spelling errors. N-grams are n letters
subsequences of words or strings taken from a string with a length of whatever n is set to[37].N
stands for one, two or three. One letter n-grams are referred to as unigrams or monograms; two-
letter n-grams are referred to as bigrams; and three- letter n-grams are seen as trigrams [37].

The input word's n-gram sequences are compared to the subsequences that were previously
saved, and N-gram frequencies are stored in an n-dimensional matrix. It is a method to find
37
misspelled words in text and used for non-word errors [38], [39]–[41]. The term is considered to be
misspelled if one n-gram is missed.

In order to pre-compile an n-gram table, a dictionary or corpus of text is usually required. Instead
of comparing each entire word in a text to a dictionary, just n-grams are controlled. The n-gram
algorithm was developed as one of the benefits is that it allows strings that have differing
prefixes to match and the algorithm is also tolerant of misspellings. Each string that is involved
in the comparison process is split up into sets of adjacent n-grams.

The table stores n-gram’s existence or frequencies, any n-grams in an input string that have
nonexistence or low frequencies are classified as probable misspellings. It is said that a set of
positional binary n- gram arrays are able to detect error more accurately. Because each element in
the positional binary n-gram arrays matches the exact position within each word. However, this
raises the storage space problem due to the large capacity of the positional arrays. Since most
misspellings do not contain any impossible n-grams, so n-gram analysis techniques are not good
at detecting human generated errors but good at detecting machine- generated errors like OCR
[39].

A check is done by using an n-dimensional matrix where real n-gram frequencies are stored. In
general, n-gram detection technique work by examining each n-gram is an input string and
looking it up in a precompiled table of n-gram statistics to determine either its existence or its
frequency of words or strings that are found to contain nonexistence or highly infrequent n-
grams are identified as either misspelling. The major advantage of n-gram algorithm is that it
requires no knowledge of the language that is used with, so it is language independent or it is a
neutral string-matching algorithm.

Dictionary Look Up

A dictionary/Wordnet is a lexical source containing a list of correct words for a specific language
[38], [40], [41]. Dictionary lookup approach is one of the methods of developing spell checkers,
which use a dictionary to find words. It is looking at every word in the dictionary of the lexicon, a
corpus, or a combination of lexicons and corpora [35]. The non-word errors can be easily detected
by checking each word against a dictionary. If the word exists in the dictionary, it is considered a
correct word and if the word doesn’t exist in the dictionary, it is flagged out as a misspelled word.
38
Dictionaries are represented in many ways, each with their own characteristics like speed and
storage requirements. Large dictionary might be a dictionary with most common word combined
with a set of additional dictionaries for specific topics such as computer science or economy. Big
dictionary also uses more space and may take longer time to search. storage space requirement and
time-consuming searching

The drawbacks of this method are difficulties in keeping such a dictionary up to date, storage space
requirement and complete enough to include all the words in a text. It is the fact that it will be
impossible to store all words and even if all are stored, they should be updated from time to time.
In addition to this, a known shortcoming of dictionary-based systems is handling so-called real-
word errors [14]. This kind of error is difficult to identify using these methods because the
misspelled word exists in the dictionary.

Moreover, searching for a word in a large knowledge base would not be efficient [6]. To clarify
this, a large dictionary requires more space and may take a longer time to search for a specific
word and a too-small dictionary consider many words as invalid words but the words are valid in
that specific language. Simultaneously, system response time should be reduced. Dictionary
lookup and construction techniques must be tailored according to the purpose of the dictionary. To
solve the problem of inefficiency, a technique called hashing is used.

Hash tables are the most common used technique to gain fast access to a dictionary. A hash table,
commonly referred to as a hash map, is a type of data structure that uses an associative array or
dictionary. It is an abstract data type that associates keys with values. A hash table utilizes a hash
function to create an index. When a word is hashed, the hashing address or key is looked up in a
hashing table, and the resultant hash shows where the relevant value is kept. i.e., compute its hash
address and obtain the word stored at that address in the hash table that has already been built. The
fundamental benefit of hash tables is their random-access nature that eliminated the large number
of comparisons needed to search the dictionary, which makes dictionary searches faster [42].

The main drawback of hash tables is the requirement for creative collision-avoidance hash
algorithm. We compute each hash function for a word before storing it in the dictionary, then we
set the vector elements in the dictionary that correspond to the computed values to true. We
determine the hash values for a word and search the vector to see whether it is a part of the

39
dictionary. If all entries corresponding to the values are true, then the word belongs to the
dictionary, otherwise it does not.

Error Correction Techniques

The second step of spell checking, next to detection, is spelling error correction. Spell correcting
refers to the attempt to endow spell checkers with the ability to correct detected errors, i.e.,
to find the subset of dictionary or lexical entries that are similar to the misspelling in some way.
Error correction consists of two steps; the generation of candidate corrections and the ranking of
candidate corrections [3], [38]. To find one or more viable correction words, the candidate
generation procedure commonly uses a precompiled table of permissible n-grams. To rank order
the candidates, the ranking procedure commonly uses a lexical similarity measure between the
misspelled text and the candidates or a probabilistic assessment of the likelihood of the repair.
Most of the time, these two procedures are viewed as independent processes and carried out in
order.

Spelling error correction categorized in different way based on the way of handling. First it can be
classified as interactive and automatic.

In interactive, the spellchecker can suggest more than on correction for each misspelled word and
the user decide to select for replacement. In case of automatic correction, the spellchecker has to
decide on the one best correction and the error is automatically replaced with it.

In in another way it can also classified as isolated word error correction and context-based error
correction. Isolated term correction can work by identifying spelling errors for each word in a text
whereas context-sensitive correction checks if a word (spelled correctly or not) gives a meaning
context-wise. The former one is mainly used for correcting real word errors. Real word errors are
words which have correct meaning in the dictionary but contextual error in a given text.

Rule-based, n-gram-based, minimum edit distance, similarity keys, probabilistic, neural

networks, and noisy channel approaches are a some of the error correction techniques now in use
[35], [42].

40
Minimum Edit Distance

The spelling correction method that has received the greatest attention and currently most widely
used to spelling correction is the minimum edit distance method. This technique involves
converting one string into another at the lowest possible cost or with the fewest possible editing
steps [35]. In order to transform, this refers to the fewest amount of editing operations that must
be performed between two characters. With the use of this technique, several operations,
including insertions, deletions, substitutions, and transpositions, can be carried out depending on
various potential costs for each of these operations. The
shortest edit distance between a misspelled word in a text and a word in the dictionary is typicall
y what one concerned about.

According to various scholars, including Kuchich [3], [49], the majority of spelling errors might
be corrected by the addition, deletion, substitution, or transposition of one letter or two characters.
The dictionary word is referred to as a plausible correction if a misspelling may be changed into
one by reversing one of the error processes (i.e., insertion, deletion, substitution, and
transposition).

This method is only applicable to typos in single or more useful for keyboard input errors that
phonetic errors[42], [43]. As a result, there are only fewer possible comparisons. By comparing
words with a length of four to six characters to a list of words that appear frequently, the Edit
Distance algorithm can identify spelling errors. If the search term is not included in the list, it is
then looked up in a dictionary that has terms arranged by alphabetical order, word length, and
character occurrence.

Where a word in the dictionary is one character longer than the detected word, then the first
character in the dictionary word that is different is discarded and the rest of the characters are
shifted one-bit position left [35]. If the two words match, then the word in the dictionary is reported
to be the correctly spelled word by a single insertion.

On the other hand, the first character in the detected word that differs from the matching character
in the dictionary word is considered as incorrect if the word in the dictionary is one character
shorter [35]. The remaining characters in the misspelled word are therefore shifted one bit to the
left and that particular character in the detected word is removed. The dictionary term is reported

41
as the correctly spelled word by a single deletion if the new misspelled word and the entry in the
dictionary match.

When the lengths of the word in the dictionary and the misspelled word are the same, but they
differ by one-character position, then the dictionary entry is reported as a candidate correction as
they differ by a single substitution.

Where the lengths of the word in the dictionary and the misspelled word are the same, but they
differ in two adjacent positions, the characters are proposed to be swapped. If the two words
are the same, there is a match by a single transposition [35].

Rule Based Techniques

Rule-based technique works by having a set of rules that capture common spelling and typographic
errors and applying these rules to the misspelled word. In this technique, incorrectly spelt words
are changed into the correct ones by using algorithms that try to express knowledge of typical
spelling error patterns. The information is provided as rules [35]. They work by comparing the
misspelled word to a set of criteria that identifies common spelling and typographical errors

These rules may include general morphological information, the lengths of the misspelled words,
and more. For example, they may explain how to turn a verb into an adjective by adding the suffix
"-ing." [35] . Every right word produced by this algorithm is accompanied by a correction
suggestion. You can rank the ideas by summing the probabilities for the relevant rules, which are
included in the rules. The concept of edit distance can be viewed as a particular application of a
rule-based system with a limitation on the number of rules that can be employed. Edit distance can
be considered as a subset of a rule-based technique with limitations on the possible rules [44].

Similarity Keys

In this technique, every word is mapping of every word into a key [35]. The key to the misspelled
word is computed and words with similar key values to it will be generated as candidate
suggestions for the misspelled word[35], [38]. Words are converted into similarity keys that reflect
relationships between the characters in the words, such as positional similarity, material similarity,
and ordinal similarity, using similarity key approaches [35].

42
Positional Similarity: Its name relates to the degree to which the matching characters in two strings
are located in the same place. It typically appears in an OCR text and literary comparison, and it is
reportedly too limited to be utilized by itself for spelling correction.

Material Similarity: This describes how similar two strings are when the characters are in a
different order. As a gauge of material resemblance, correlation coefficients between the two
strings — the misspelled word and a word made up of the exact same letters but in a different order
— have been utilized. The material similarity is thought to be insufficiently exact for the task of
spelling correction because all anagrams have the same basic elements. For instance, the words
“break” and “baker”, “brush” and “shrub”, etc. are anagrams of each other and have a lot in
common.

Ordinal Similarity: Ordinal similarity, like position similarity, describes character similarity
between two words when they are ordered similarly.

Each dictionary word is given a key, and only dictionary keys are compared to the key generated
for the non-word. The Non-standard word for which the keys were calculated. As a
recommendation, the word with the most comparable keys is chosen. This method is fast because
it only processes words that have similar keys. With a good transformation algorithm, this method
can handle keyboard errors.

N-gram Based Techniques

As we discussed in n-gram analysis for spell checking in Section 2.4.1, it is a subsequence of a

sequence of letters or words with n = 1, 2, or 3 letters. N-grams can be used in two ways, either
without a dictionary or together with a dictionary. It can be used either with or without a dictionary
depends on the situation. N-grams have been employed in a variety of ways, including as trigrams,
bigrams, and unigrams, in text recognition and spelling correction algorithms. N-grams are used
to identify the positions in which the misspelled word error occurs without the use of a dictionary.
If there is a special method for correcting a misspelled word so that it only contains accurate n-
grams, that is what is used. This method's effectiveness is limited. N-grams are used to define the
distance between words along with a dictionary, however the words are always checked against
the dictionary.

43
Probabilistic Techniques

The probabilistic approach is based on statistical features of the language [38]. Initially,
probabilistic approaches such as OCR were utilized for text recognition [35]. The two common
methods are transition or Markov probabilities and confusion or error probabilities [35].

a) Transition or Markov probabilities, like n-grams, determine the probability that a given letter
will be followed by another given letter in a given language. These probabilities can be
calculated by gathering n-gram frequency data from a big corpus.
b) Confusion or error probabilities give the chance of a particular character substituting another
one in a misspelled word. When given a sentence to correct, the system decomposes each string
into letter n-grams and searches the lexicon for word recommendations by comparing string n-
grams to lexicon-entry n-grams. When we have access to a dictionary or index, transition
probabilities are ineffective.

The main difference between the two probabilities is that transition probabilities depend on language,
while confusion probabilities depend on source[35]. To explain, confusion probabilities vary
depending on the source since different OCR devices employ different methodology and features,
such as font types, and each device outputs a unique confusion probability distribution. Human error
can also contribute to confusion probability.

Neural Networks

Neural networks are also an interesting and promising technique, but it seems like it has to mature
a bit more before it can be used generally. Back-propagation networks are now used, with one
output node for each word in the dictionary and an input node for every conceivable n-gram at
every position of the word, where n is commonly one or two [44]. Only one of the outputs should
be active at any one time, showing which dictionary terms the network recommends as a
correction. This strategy works for tiny dictionaries (under 1000 words), but it does not scale
well. The disadvantage of this approach is the time requirements are too big on traditional
hardware, especially in the learning phase.

44
2.5. Summary
In this chapter, we highlighted the sociolinguistic aspects of Guragina Language varieties. There
are more than 13 Guragina Language (also called Guragina) varieties. Linguistically, the sub-
groups of Guragina are North Gurage (Kistane and Dobi), West Gurage (Meskan, Cheha, Ezha,
Gumer, Muher, Inor, Endegagn, Enner, and Geto), and East Gurage (Welane, Silte, and Zay) are
the major linguistic subgroups of Guragina. West Gudrage further classified into Central west
Gurage (Cheha, Ezha, and Gumer) and Peripheral West Gurage (Inor, Endegagn, Enner, and Geto).

We also addressed the morphology of standardized Guragina in this chapter. It has a complex
morphological structure as an Ethio-semitic language. Guragina verbs, nouns, adverb and
adjectives can be simple or complex in structure. Affixes are used to create complex verb forms
such as passive causative, verbal noun, and verbal reduplication stems. Guragina verbs are also
negated by affixes, which correspond to the subject, object, and applicative affixes.

45
 CHAPTER THREE: RELATED WORK

3.1. Introduction
Several studies have been conducted to explore various methods for spell checking as a challenge
within natural language processing (NLP). The field of NLP has extensively researched spell-
checking, resulting in the development of numerous spell-checkers for different languages
worldwide. To provide a comprehensive understanding of spell detection and correction systems,
this chapter presents a background study that examines key works in this area.

Section 3.2 focuses on the spell checkers for specific sub-groups within the Afro-Asiatic language
family, particularly the Semitic languages like Amharic and Arabic. Section 3.3 covers the spell-
checking approaches and strategies employed for Cushitic languages (Afan-Oromo) and Omotic
languages (Kafi Noonoo) within the same language family. In Section 3.4, the discussion shifts to
Indo-European languages such as English and Bengali, exploring the spell-checking techniques
used for these languages. Section 3.5 digs into the spell checker developed for one of the languages
in the Austronesian family, namely Malay. Finally, Section 3.6 provides a summary of the chapter,
highlighting the key findings and contributions related to this field of study.

3.2. Spelling Checker for Semitic Language

3.1.1. Spelling Checkers for Amharic
Yacob [45] attempted to implement the Metaphor algorithm for the Amharic language. In this work,
the author encoded letters with similar sounds into a single letter to have a single representation of
words. In this paper, as much as possible the researcher tried to eliminate most errors and, in the
author’s, experiment used 116 canonically spelled words and 166 misspelled words. According to
the author’s report, the Amharic Metaphone algorithm’s success was up to 90%. The primary
strength of this study is well-planned stylographic redundancies, but its weakness is the usage of a
limited corpus.
In another research, Melaku Tilahun and Tesfa Tegegne [11] develop a spell checker tool for
Amharic texts and demonstrated it by integrating it with open office. In addition, the authors adopted
word formation rules for the Amharic language, which can be integrated into the lexicon used by
an Amharic spell checker. It is also a word-level spell checker, particularly a non-word error
detector spell checker. That is, it does not consider real-word errors and white space. In this work,

46
the authors try to add some new features that are not addressed in previous works, such as internal
inflected words and repeated words. In addition, the usage of Unicode data by the author is supposed
to increase the performance of spell checking by avoiding transliteration. Finally, they try to
measure the performance of the system by taking 5 experiments and calculating the recall and
precision and got the overall performance of the system is 97.27%

In another study[46] , spelling mistakes in Amharic and English are corrected using a corpus-
driven technique with noisy channel. The Damereau-Levenshten edit distance method is used to
assess how closely produced words to a misspelled word are related to one another. The study
focuses in spelling correction for non-word errors. Given that Amharic is morphologically rich,
the authors argue that it would be time-consuming to compile a list of all language-dependent rules
for spelling correction.

They created their own contemporary Amharic corpus (CACO), which was put together from
publicly available Amharic newspapers, legal documents, journals, fiction, short stories, political
books, the Amharic Bible, and children's books. For comparison, they utilized HaBiT, a big text
corpus built from automatically crawled sites. Characters are converted into Latin-based characters
after being extracted from paragraphs. The next steps involve replacing numerals with
placeholders, splitting hyphenated words by deleting the hyphen and substituting a space character,
and identifying and extracting unique phrases. Words are tokenized based on the two-dot separator
character (:) and white space from the detected phrases. The remaining sentences that include
words that only appear once across the whole corpus are then deleted. By presuming that the term
that is only stated once is presumably spelt wrongly, this additional step is introduced.

The KenLM language-modeling toolset is used to train the language models. The error model was
adapted from one developed by Norvig (2009) using 40.000 spelling errors. The most likely split
possibilities based on the related word list and the CACO language model are used to separate
terms with missing white spaces. Based on Amharic test data, the effectiveness of their method is
assessed, and the findings are contrasted with those of the Aspell and Hunspell baseline systems.

89.4% accuracy, 80.6% recall, and 84.8% F1-score were the results of the evaluation for Amharic
spelling error detection. Recall shows language coverage, F1 measure indicates the capacity to
identify spelling errors, and precision flags all misspellings. Additionally, evaluation metrics for
the HaBiT, Aspell, and Hunspell baseline systems were computed and compared. The findings

47
demonstrate that the suggested HaBiT technique improved spelling error detection (F1). However,
when the term list from the HaBiT corpus was applied, the value of recall did not improve.

Additionally, the top five suggestions for the proposed system using CACO contain 77% of the
correct spellings, as opposed to 34% for Hunspell, 62% for Aspell, and 75% for HaBiT. Suggestion
lists were also assessed. The suggested approach excelled HaBiT, Aspell, and Hunspell in the top
initial recommendations categories by 9%, 18%, and 35%, respectively. As long as they are typed
using a QWERTY keyboard with direct mapping between keystrokes and characters, the authors'
suggested approach may be applied to other written languages.

Using a hybrid methodology, Genet Assefa[47] developed an effective automated Amharic spell
checker. The presented study employs the Metaphone and Edit Distance method. While the
Metaphone technique is used to identify spelling errors, the edit distance approach is used to select
the most likely correct word for the misspelled word. The author utilized a 125,000- dictionary
word as well as 500 words for testing. According to the paper, when a phonetic algorithm and edit
distance algorithm combination is used in the development of the spell checker, its error detection
and correction suggestion skills are more than 95% successful. The technique taken in creating
and implementing the spell checker is one of the system's fundamental flaws. The dictionary-based
approach demands that each term be kept in a particular dictionary. Due to time and resource
constraints, it is challenging to store all of the Amharic terms in the dictionary. Additionally,
processing speed and time were not taken into account in the research.

In 2022, Maryamawit Shumetie [51] conducted research on a deep learning-based spell checker
for the Amharic language. The author makes an effort to develop and implement a spell checker
that takes into account the meaning of words both before and after the current word, which is meant
to be corrected.

In the study, a deep learning dual-input model that can take into account the input word's context
in both the right and left branches was presented. The experiment is conducted by setting up a
baseline approach to the edit distance. The same number of data and training environments are
used for the experiment. Since the loss function is based on Sparse Categorical Cross-entropy,
accuracy and loss are used to evaluate the model. 116274 distinct words were collected from
diverse sources for the system's training, testing, and assessment. Python was used to train the
system on Google Collaboratory. It was put to the test by adding the incorrect words in a sentence.

48
It was able to make suggestions while identifying the incorrect word in the phrase that was
provided. In the experiment, the suggested model outperformed the baseline model edit distance
with a smaller loss value and higher accuracy. The precision of the edit distance was 0.68. The
dual input encoder suggested model attained accuracy of 0.9349.

An optimization approach is used during the model training to reduce overcorrection and loss.
Therefore, it can be concluded that the context-based dual-input model is more effective to the
baseline technique for identifying and correcting spelling errors. For improved accuracy and
decreased loss, the system may be improved in the future by using more deep learning algorithms
and more data corpora. The author suggests that this technique may be employed in other
languages as well.

3.1.2. Spelling Checkers for Arabic

Hamza et al. [52] present an Arabic spell checking approach based on morphological analysis and
edit distance algorithms. They utilized a small corpus of 2784 misspelled words for testing. Finally,
they tested evaluate and compared their system to the Levenshtein distance. They reported the
results of the comparison by three error operations, such as insertion, deletion, and permutation, and
discovered that their approach scores 85%, 81%, and 86% success for each error operation,
respectively, while the other Levenshtein distance scores 44%, 81%, and 61% success for each
corresponding error operation. The proposed system's average correction rate is 33.7% higher than
the Levenshtein distance. The average correction time (0.10 ms) is significantly faster than the
Levenshtein distance (0.19 ms).

Another study also carried out for Arabic spell checkers was by Shaalan et al. [49], who proposed
automatic spell check for Arabic. They developed a tool capable of recognizing and suggesting
corrections of misspelled input for common spelling errors. The system was composed of an Arabic
morphological analyzer, lexicon, spell checker, and spell corrector. They limit their system to detect
and correct spelling errors to isolated words. The tool developed tries to add missed characters,
replace an incorrect character, remove the incorrect character and add a space to split words. The
developed tool is helpful for automating the proofreading of the human-typed Arabic texts.

49
3.3. Spelling Checker for Afro-asiatic Language
Both Afan Oromo and Kafi Noonoo are languages that are part of the Afro-Asiatic language
family, although they are categorized under separate branches within the family. Kafi Noonoo is
Gongo sub group of the north Omotic family while Afan Oromo is Cushitic sub-group.

3.3.1. Spelling Checkers for Afaan Oromo

Gaddisa Olani & Dida Midekso [50] studied the design and implementation of a morphology-
based spell checker for the Afaan Oromo language. The authors designed the spell checker using
the dictionary lookup method together with the morphological analyzer. They used a total of 1811
words of test data and tested their system, and they reported using the following three matrices,
lexical recall 88.62%, error recall 100% a, and precession 28.62%. The limitation of this work is
the lexicon coverage of their knowledge bases is small.

3.3.2. Spelling Checkers for Kafi Noonoo

Fikru Tafesse and Yaregal Assabie [39] developed a morphological basis spell checking system
for the Kafi Noonoo language. The suggested architecture of the spell checker in this study consists
of four major components: tokenization, error detection, word suggestion, and error correction. To
implement, dictionary lookup techniques and morphology-based approaches are utilized. The
system prototype is created to test and evaluate the functionality and performance of the spell
checker system. To test and assess the system, 2743 unique words were gathered from various
sources. The evaluation metric was lexical recall, error recall, and precision, and the results were
95.91%, 100%, and 62.76%, respectively. The fundamental limitation of the study is the small
vocabulary of root terms in the root dictionary. The intra-attention mechanism enhanced their
model's overall performance, but it failed in several specific cases.

3.4. Spelling Checker for Indo-European Language

3.4.1. Spell Checker for Bengali Language
In 2017, Mandal and Hossain [53] developed a Bengali (Bangla) spell checker that could only handle
typographic and phonetic mistakes. The authors both real-word and non-word errors in this work. To
deal with these errors, they employ the confusion set approach. The disadvantage of this work is that
it only used a small set without taking into account the morphology the language.

50
Hossain et al. l [37] did another study for a Bangla lexicon with about 1 million distinct terms.
Based on this corpus and lexicon, the authors created a combined spell and grammar checker
program that identifies unique spelling and grammatical errors while also providing suitable
suggestions for both. The spell checker detects all forms of Bangla spelling errors with an accuracy
rating of 97.21% using the Double Metaphone technique and Edit distance based on distributed
lexicons and numerical suffix dataset. The shortcoming of this study is that split-word may not be
detected as a misspelled word.

3.4.2. Spelling Checkers for English

Early research on spelling detection and correction is based on phonetic and string similarities like
Metaphone and Damerau-Levenshtein edit distance algorithms [4]. The Levenshtein
distance measures the similarity between two strings of characters. The Levenshtein distance and
the Damerau-Levenshtein distance are similar, but the latter includes the transposition of two
adjacent characters' type differences (which is a common typographical error). Candidate
corrections are ranked from manually compiled lexicons with the help of these algorithms. GNU
Aspell is a good example that follows this approach
GNU Aspell is a spell checker program developed by Atkinson [51]. It is a free and open-source
spell checker designed to eventually replace Ispell. It can be used either as a library or as an
independent spell checker. Its main feature is that it does a better job of suggesting possible
replacements for a misspelled word than just about any other spell checker out there for the English
language. Unlike Ispell, Aspell can easily check documents in UTF-8 without having to use a
special dictionary. It also supports using multiple dictionaries at the same time. This spell checker
also does not consider the morphological structure of the language and others.

The spell checker application developed by Bhaire et al. [52] uses an edit distance algorithm for
spelling correction and a data structure tree for storing dictionaries. As a result, autosuggest is
possible.

To improve its speed, the spell checker contains features such as multi-word suggestions.
However, if the error is at the beginning of the word, the system does not provide any alternative
spelling word from the list, and if the user-written word has too many errors, it may not offer any
alternative spelling suggestion. The disadvantage of this system is that, because it is dictionary-
based, it must keep all words (inflectional, derivational, and compound words) in the dictionary.
51
This consumes space and increases the likelihood of missing more common inflectional,
derivational, and complex terms in the language.

Seth and Mieczyslaw developed a smart spell checker system called SSCS, which uses adaptive
software architecture to correct users' errors. The system adapts to individual users by incorporating
their feedback to modify its behavior. The architecture is designed like an adaptive controller, with
a feedback mechanism for system adaptation. The spell checker utilizes five sources of knowledge
and text files containing word lists as input. Its goal is to identify incorrect words, attempt to suggest
replacements, and provide the user with options. The SSCS program demonstrates the Adaptive
Software Architecture with nine Knowledge Sources (KSs) across three domains: Input, Error, and
Evaluation. It uses dictionaries, including a user-defined dictionary, to detect incorrect terms. The
system employs various knowledge sources such as character shifting, doubling, appending,
removing, and switching to correct incorrect words. The use of multiple KSs improves system
reliability and expands coverage of spelling problems. Compared to non-adaptive systems, their
method achieved a significantly higher accuracy rate of selecting the correct corrected word in over
80% of cases. However, this system does not consider the morphology of the language, which is a
limitation of the work.

3.5. Spelling Checker of Austronesian Language

3.5.1. Spell Checker for Malay
Basri et al. [49] developed an automatic spelling checker for Malay blogs, which aimed to detect
and correct misspelled words without human input. The previous spell checker relied on a
dictionary, but it could only correct words already present in the dictionary. To address this
limitation, the researchers proposed a novel method consisting of five modules: tokenization,
symbol elimination, removal of English words, stemming, and misspelled word identification.
They tested their methodology using Malaysian-written Malay weblogs, specifically focusing on
11 out of 23 weblog categories. After processing the datasets, they obtained 3,046 unique words,
and after removing English terms, they applied the spell checker to 2,675 words, successfully
detecting and correcting 2,453 of them. However, their method missed 232 words (8.67% of all
words). It's important to note that this work solely utilized a dictionary and did not account for the
morphology of the language for inflectional, derived, or compounded words.

52
3.6. Summary
In this chapter, we conducted a thorough review of various spell checker studies conducted for
different languages families, including Austronesian (Malay), Indo-European (English and Bengali),
Afro-asiatic (Afaan Oromo and Kafi Noonoo) and Semitic (Arabic and Amharic). However, we
couldn't find any specific studies related to spell checking in the Guragina Language. The absence of
a comprehensive approach or methodology for generating root words. There is a lack of an effective
and accurate spellchecker and suggester system for Gurage Language. Our main focus was to identify
advanced techniques that are particularly suitable for handling the morphological complexity found
in languages like Guragina. We examined the strategies, approaches, strengths, and limitations of
each spell-checking system in relation to their respective research. The majority of spell checker
systems rely on dictionary lookup and the Levenshtein edit distance algorithm as their foundation.
Some publications have proposed morphology-based spell checks for languages like Amharic,
Arabic, Kafi Noonoo, and Afaan Oromo. Through our analysis, we have determined that
morphology-based techniques are considered state-of-the-art for addressing the challenges posed by
morphologically complex languages. It is important to note that spell checkers designed for other
languages cannot be directly applied to the Guragina Language due to its complex morphology,
language variations, and other factors.

53
 CHAPTER FOUR: DESIGN OF GURAGINA SPELL CHECKER

4.1. Introduction
In doing research, selecting the appropriate approach and design is essential. This leads to the use
of a research paradigm that can lead to the desired outcome for the problem. A research
methodology is chosen that incorporates procedures and techniques that are the best match for the
research in order to achieve the goals and objectives of the study as well as to offer valid and
trustworthy results. In order to address flaws that have been noticed, a new artifact must be
designed, evaluated, and the findings must be presented. This research therefore perfectly fits the
design science research technique, which is based on the approach described in [12] . In order to
examine the issue and find a solution, many steps were taken as shown in Figure 4.1, which are
covered in this chapter along with the research approach used.

Figure 4. 1 Design Science Research Process Model

54
The spell checker algorithm's backend uses morphological characteristics and a stem and affix
dictionary, which make up the other component of this design. The design and implementation of
a spell checker for the Guragina Language were covered in this chapter.

4.2. General Architecture

Morphologically complex languages like Guragina are known for their extensive set of complex
derivation rules, which poses challenges in defining and implementing all of these rules for tasks
such as morphological generation and derivation. This can lead to performance issues when
attempting to detect spelling errors by going through all the rules. To address this, the system
utilizes an algorithm that resolves the performance concerns.

The proposed Guragina Language spell checker system has three major components to accomplish
this. These are preprocessing, error detection, and correction suggestion components, as illustrated
in the Figure 4.2.

The proposed Guragina Language spell checker system has three main components: preprocessing
(including tokenization and vowel consonant (VC) mapping), error detection (involving stem
extraction, and matching), and correction suggestion (with a Distance Calculator and Suggestion
Ranking system). These components work together as illustrated in Figure 4.1 to accomplish the
overall spell checking task.

The system uses two stored databases in the background. These are list of part of speech tagged
stem word for different verb, noun, adverb, adjectives for matching and affixes are stored on
another database. It also uses Guragina Stem Extraction, Guragina Word Distance Algorithm
(GWDA) and Suggestion Ranking components. Fig 4.2 shows the detailed architecture of the
system. The functionalities of each component are discussed in the following section.

55
 Figure 4. 2 General Architecture

4.3. Preprocessing
Preprocessing is an essential step in natural language processing (NLP) that involves transforming
raw text data into a format suitable for further analysis or modeling. It typically includes several
tasks aimed at cleaning, normalizing, and transforming the text. In In the Guragina language spell
checker system, the preprocessing component primarily consists of tokenization and VC-mapping.

56
4.3.1. Tokenization
Tokenizer is responsible for breaking down a text or input into individual tokens or units, such as
words, phrases, or characters. It segments the input text into meaningful units that can be processed
further by other components. Tokenization is a fundamental step in many natural languages
processing tasks, including spell checking. The tokenization algorithm for the terms is presented
in Algorithm 4.1.

Input፡ PARAGRAPH
Output: WORDS_WITHOUT_PUNCT: A list of tokenized words.

1: Input a paragraph of text (PARAGRAPH).

2: Split PARAGRAPH into individual words using whitespace as the
delimiter. Store the words in a list (WORDS).
3: Initialize an empty list to store the tokenized words without
punctuation (WORDS_WITHOUT_PUNCT).
4: For each word in WORDS:
a: Remove punctuation characters from the word.
b: Append the modified word to WORDS_WITHOUT_PUNCT.
5: Output WORDS (tokenized words) and WORDS_WITHOUT_PUNCT (tokenized
words without punctuation).

Algorithm 4. 1 Tokenizer algorithm used by the spell checker

In our system one of the preprocessing tasks is tokenization. Tokenization is a technique for
dividing a block of text into separate tokens. Therefore, these tokens might be sentences, phrases,
paragraphs, punctuation, or single words (Lines 1). Our method is designed to assess a word's
accuracy for the Guragina Language at the word level, hence in our instance for spell check, we
take tokenization at the word level into consideration. Boundary delimiters like white space and
punctuation are used to separate the text. Punctuation marks that appear at the end of a word are
ignored and treated as separate tokens.

Words are separated from one another by white space. If a space is present, the word that follows
it becomes a token. Therefore, in our text processing example, we consider blank space for
tokenization (Lines 2 -5).

57
4.3.2. Vowel Consonant mapping
This program separates vowels and consonants in a word by iterating through each character and
classifying it as either a vowel or a consonant. A vowel is a speech sound produced with an open
vocal tract, typically represented by the letters 'አ'’ə’,‘ኡ'‘u’, 'ኢ'’i’, 'ኣ'’a’, 'ኤ'’e’, 'እ’'i’, and 'ኦ'’ o’ in
Guragina Language. Consonants, on the other hand, are speech sounds produced with a
constriction or closure in the vocal tract, including letters such as 'ሕ'‘hɨ’, '𞟥' ‘hʲɨ’, '𞟫'‘hʷɨ’, 'ል'‘lɨ’,
'ም'‘mɨ’, 'ᎃ'‘mʷɨ’, 'ር' ‘rɨ’, 'ስ'‘sɨ’, and so on. A vowel-consonant mapper takes a word as input and
separates the vowels and consonants into different groups or lists. The purpose of a vowel-
consonant mapper is to analyze the phonetic structure of words and to differentiate stem word and
affix of each word. For example, the verb ‘ቸነነም’’ʧ ənənəm’, which means we came, map (convert)
to ‘ችአንአንአም’‘ʧ ənənəm’.

4.4. Error Detection

The Error Detector component identifies errors in the input text. It analyzes the tokens or units
provided by the Tokenizer, VC mapper and flags misspelled words. The Error Detector often relies
on stem extraction, affix rule and matching to detect these errors.

4.4.1. Guragina Stem Extraction

Developing a stemmer for the Gurage language, a Semitic language with rich and complex
morphology, is an important contribution, as stemming is a crucial step in various natural language
processing (NLP) applications such as information retrieval, machine translation, and text
classification.

For the Guragina language, one approach to represent the word structure could be through
character embeddings. Character embeddings capture the sub-word-level information, which can
be particularly useful for morphologically complex languages like Gurage.

The proposed stemmer has two key functionalities:

1. Generating the derivation pattern: This component is responsible for identifying the
probable stem word by analyzing the word structure and extracting relevant prefixes,
suffixes, and hyphens.

58
 2. Reversing the pattern: This functionality allows the stemmer to reconstruct the original
word from the identified stem and affix information.

The stemming algorithm, as outlined in Algorithm 4.2, is designed to process a word by removing
specific prefixes, suffixes, and hyphens to extract the stem. This step is crucial for downstream
NLP tasks, as it helps to normalize word forms and improve the performance of various
applications. This step is crucial for subsequent NLP tasks, as it helps to normalize word forms
and improve the performance of various applications.

Input: WORD

Output: STEM WORD

1: Read the word (WORD) to be processed.

2: Iterate over each prefix in the list of Guragina prefixes

(GURAGINA_PREFIXES):

3: If the word starts with the current prefix:

4: Remove the prefix from the word by slicing it from the beginning.

5: Break out of the loop.

6: Iterate over each suffix in the list of Guragina suffixes

(GURAGINA_SUFFIXES):

7: If the word ends with the current suffix:

8: Remove the suffix from the word by slicing it from the end.

9: Break out of the loop.

10: Remove any remaining hyphens "-" from the word by replacing them with an
empty string.

11: Return the processed word.

Algorithm 4. 2 Stem extractor Algorithm.

59
4.4.2. Matching
The spell checker follows a workflow that involves several steps. First, it accepts a block of text
or single word, splitting it into words and remove additional punctuations. Then, it maps it to its
vowel-consonant pattern.

A matching component is crucial for a spell checker, as it verifies the validity of a word by
searching a dictionary of correctly spelled stem words in VC format. It accepts tokenized words
from VC mapping or stems extraction and maps them accordingly, as shown in Algorithm 4.3.

The process starts by searching for the word in the stem database. If found, it is marked as correctly
spelled (Lines 2). If not, the spell checker searches the affix database for matching affixes (Lines
4). If matches are found, the word is derived to its stem form and checked in the stem dictionary.
If the stem word is found, the original word is marked as valid (Lines 4).

In cases where no matching affixes are found, the spell checker checks if the word is found in the
stem dictionary. If it is found, the word is marked as correctly spelled. If it is not found, the spell
checker flags it as incorrectly spelled and marks the position of the error (Lines 5).

Input፡ a word (WORD) to be spell-checked

Output: found or not found in dictionary

1: Convert WORD to its vowel-consonant (VC) representation

2: Search for WORD in the dictionary
3: If WORD is found:
- Set result = "found in dictionary"
- Mark WORD as correctly spelled
4: Else:
- Search for matching affixes
- If affixes match:
- Drive WORD to its stem form
- Search the generated stem in the dictionary
- If the stem is found:
- Set result = "found in dictionary"
- Else:
- Set result = "not found in dictionary"
- Mark the error position
5: Return result

Algorithm 4. 3 General Matching Algorithm

60
If none of the derivation patterns from the matched rules match the word, the spell checker marks
it as incorrectly spelled and notes the position of the error. This information is later used to generate
corrections (Lines 15-16). If the affixes do not match or are not close to each other, the spell
checker marks the error position and labels the word as invalid. If the result is an incorrectly spelled
word, the spell checker replaces the characters at the error position with the probable correct
characters and generates suggestions. It also searches for closer stem words and performs the same
replacement and suggestion generation process (Lines 17-22). This workflow outlines the steps
taken by the spell checker to determine the correctness of a given word and provide suggestions
or corrections if needed.

4.5. Correction Suggestion

The Correction Suggester component provides suggestions or recommendations to correct the
identified errors or improve the input text. It leverages the information from the Error Detector
and the results of the Distance Calculator to generate a set of potential corrections or improvements
by adopting the Ratcliff algorithm.

The Ratcliff algorithm, also known as the Longest Common Subsequence (LCS) algorithm, is a
sequence matching algorithm that can be represented mathematically.

Given two sequences, such as words or strings, the algorithm finds the longest common
subsequence (LCS) between them. A subsequence is a sequence derived from the original by
deleting some or no elements without changing the order of the remaining elements. The algorithm
calculates the length of the LCS between the two sequences. The longer the LCS, the more similar
the sequences are considered to be. Based on the length of the LCS, the algorithm computes a
similarity ratio, which is a value between 0 and 1. This ratio represents the similarity between the
two sequences, where 1 indicates an exact match and 0 indicates no similarity. The algorithm sorts
the sequences based on their similarity ratios, with higher ratios indicating closer matches. Let's
break down the algorithm mathematically step by step:

Given two sequences, A of length m and B of length n, where A = [a₁, a₂, ..., aₘ] and B = [b₁, b₂, ...,
bₙ], the Ratcliff algorithm can be defined as follows:

61
Compute the Longest Common Subsequence (LCS) matrix, C, of dimensions (m+1) × (n+1),
initialized with zeros.

Iterate through each element ai in sequence A and each element bj in sequence B:

If ai is equal to bj, set C[i+1, j+1] = C[i, j] + 1.

Otherwise, set C[i+1, j+1] = max(C[i, j+1], C[i+1, j]).

The length of the LCS between sequences A and B is given by C[m+1, n+1].

Calculate the similarity ratio between A and B as the ratio of the LCS length to the maximum
length of either sequence, max (m, n): similarity ratio = LCS_length / max (m, n).

Sort the sequences based on their similarity ratios, with higher ratios indicating closer matches.

This algorithm finds the longest common subsequence between two sequences and computes a
similarity ratio based on the length of the LCS. The higher the similarity ratio, the more similar
the sequences are considered to be.

4.6. Distance Calculator

The Distance Calculator computes the similarity or dissimilarity between different units or tokens
in the text. It quantifies the distance, or measure of dissimilarity, between two units, often using
metrics like edit distance, cosine similarity, or Levenshtein distance. In this study we designed
distance calculator called Guragina Word Distance Algorithm (GWDA) used for Guragina
Language, as shown in the Algorithm 4.5.

This component has the task of determining the distance between two words in the Guragina
language. It utilizes a specially designed algorithm that provides to Guragina characters. A
preliminary experiment was conducted to compare different distance calculation algorithms, and
it was found that none of them yielded the desired result in terms of accurately measuring the
difference between Guragina words.

62
 Input፡ two word
Output: distance of the word
1. Initialization:

• Read the two words, word1 and word2.

• Initialize two variables: comparisons to 0 (to count the number of
character comparisons) and distance to 0 (to accumulate the difference
score).

2. Block Identification:

• Identify the longest matching block of characters between the two

words.
• Identify the remaining unmatching blocks of characters.

3. Character Comparison:

• Iterate through each unmatching block:

• For each character in the block:
• Compare the corresponding characters from word1 and word2.
• Increment comparisons by 1.
• Calculate a partial difference score based on the character
families and orders:
o If both characters belong to the same family and have the
same order, add 1 to distance.
o If both characters belong to the same family but have
different orders, add 0.7 to distance.
o If both characters belong to different families but have the
same order, add 0.5 to distance.
o If both characters belong to different families and have
different orders, add 0.3 to distance.
o If one of the characters is an empty string (''), add 0.1 to
distance.
4. Matching Block Contribution:

• Iterate through the matching block:

o Add the length of the matching block to distance.
5. Difference Calculation:

• Calculate the final difference score by dividing distance by

comparisons.
• Return the difference score.

Algorithm 4. 4 Guragina Word Distance Algorithm (GWDA) used by the spell checker

63
The distance calculator employs the algorithm presented in Algorithm 4.2. The algorithm begins
by identifying matching blocks of characters between the two words. It also detects unmatched
blocks of text in both words. Taking into account the matched blocks, it compares the combinations
of characters from the unmatched blocks in the corresponding positions and calculates the average
distance. Characters that belong to the same row of the Guragina alphabet are assigned a distance
value of 0.7. This is based on the assumption that if a user mistakenly types a character within the
same row, they have likely pressed the correct key for the character's family and only made an
error with the vowel. For example, in Abyssinica SIL font, entering the character ሐ (hä) requires
pressing the 'H' key followed by 'E', while entering ሑ (hu) requires 'H' followed by 'U'. Therefore,
if the user types 'hu' instead of 'hä', the difference between the two is very small, resulting in a
higher distance value. The distance calculator is designed so that a distance of one indicates that
the compared words are exactly the same. Consequently, the value decreases towards zero as the
difference between the words becomes larger (Fig 4.3). Thus, in this algorithm, a higher distance
value signifies that the words are closer in similarity.

A distance of 0.5 is assigned to characters that belong to the same column in the Guragina alphabet,
as it is considered more likely that the user made a mistake with the family rather than the order.
Characters without a matching family and order are given a distance of 0.3, as this substitution is
deemed less probable. Comparing a character to a missing character results in a distance of 0.1, while
identical characters are assigned a distance of 1. These distance assignments are supported by the
findings discussed in Section 5.2.1 and are depicted in Figure 4.4.

Figure 4. 3 Examples of distance calculation's values

64
To determine the weights assigned to Guragina characters based on their family and order, an
analysis is performed using two words, namely "ቀፓም" and "ቀፕር". These words are chosen
because they contain pairs of characters that share the same identity, characters from the same
family, and characters in the same order.

Initially, the analysis relies on the following assumptions:

1. The distance between two words ranges from 0 to 1, where a value of 1 indicates that the
words are exactly the same.
2. Fig. 4.5 illustrates five conditions, denoted as a, b, c, d, and e, each assigned a value between
0 and 1.

a b c d e

0 1

Any family, any Different family Different family Same Same family Same family
order empty order
Different order Different order Same order

(Same characters)

Figure 4. 4 Distance between characters by family and order

• Represents any family, any order, and is assigned a value of 0.

• Represents different families, and is assigned a value greater than 0 and less than 1.
• Represents different orders within the same family, and is assigned a value greater than the
average.
• Represents different families and different orders, and is assigned a value less than the
average.
• Represents the same family and the same order (same characters), and is assigned a value
of 1.
3. The distance between characters of different families and orders is considered to be in the
middle, resulting in a distance value of 0.5.
4. The distance between characters of the same family but different orders are greater than the
average value, while the distance between characters of different families and different orders

65
 is less than the average value. This is because the probability of making the first mistake (same
family, different order) is considered to be higher than the second mistake (different family,
different order).

By incorporating these assumptions, the weights between Guragina characters based on their
family and order can be determined.

4.6.1. Suggestion Ranking

This component first calculates the distance between each suggestion and the candidate word using
the Guragina word distance algorithm (GWDA) function. As shown in the Algorithm 4.5, It then
filters out the suggestions that are too far away based on the distance_threshold parameter, and
sorts the remaining valid suggestions by their distance from the candidate word. Finally, it returns
the top max_suggestions suggestions.

66
Input:

- candidate_word (str): The word to be checked for

spelling errors.
- suggestions (list): A list of suggested words.
- distance_threshold (int, optional): The maximum
allowable distance between the candidate word and a
suggestion. Defaults to 2.
- max_suggestions (int, optional): The maximum number of
suggestions to return. Defaults to 5.
Output:

- A list of tuples, where each tuple contains a suggested

word and its distance from the candidate word, sorted by
distance in ascending order.
1. Initialize an empty list called `distances` to store the
distance between each suggestion and the candidate word.
2. For each word in the `suggestions` list:
a. Calculate the distance between the `candidate_word` and the
current word using the `gwda` function.
b. Add a tuple containing the current word and its distance to
the `distances` list.
3. Initialize an empty list called `valid_suggestions` to store
the suggestions that are within the `distance_threshold`.
4. For each tuple in the `distances` list:
a. If the distance in the tuple is less than or equal to the
`distance_threshold`, add the word from the tuple to the
`valid_suggestions` list.
5. Sort the `valid_suggestions` list by the distance value in
ascending order.
6. Return the top `max_suggestions` suggestions from the `
valid_suggestions` list.

Algorithm 4. 5 Suggestion Ranker Algorithm

Together, these components form an architecture that aims to process and improve the input text
by tokenizing it, detecting errors, calculating distances, and suggesting corrections.

In summary, the algorithm and architecture provide a solution for efficiently handling the
complexities of morphologically complex languages, specifically in the context of spell checking
and word suggestion generation.

67
 CHAPTER FIVE: EXPERIMENT

5.1. Introduction
The primary objective of the assessment is to assess the performance of the proposed algorithm.
This chapter provides an overview of the data collected to develop the GLSC prototype and test
its effectiveness. It also discusses the tools employed in building the algorithm. Additionally, we
present the prototype that was developed, along with the evaluation and test results obtained from
the experiment. Lastly, we present our findings and conclusions based on the assessment.

5.2. Data Preparation

In contrast to other Ethiopian languages like Amharic, Afaan Oromo, and Tigrigna, the Guragina
Language has extremely limited resources and is challenging to acquire in digital form. It does not
have publicly available annotated corpus text for spell checker and any other NLP applications.

Source data

To build a dictionary lexicon and assess the spell checker system, this study primarily relies on a
lexicon and textual data from the Gurage zone cultural and tourism office. Additional secondary
data sources include books, children's storybooks, and the New Testament of the Bible.

Data description

The collected data forms a comprehensive linguistic resource that captures the rich vocabulary of
the Guragina language. This dataset contains 656 words, including 256 unique adjectives, 881
nouns, and 1,300 verbs. It also includes detailed morphological information, such as 18 adjective
prefixes, 18 adjective suffixes, 4 noun prefixes, 70 noun suffixes, 18 verb prefixes, and 340 verb
suffixes. This extensive data allows for systematic analysis of word formation patterns in the
Guragina language.
The researchers worked with linguistic experts to understand the structure and characteristics of
the Guragina language and to prepare the dictionary lexicon and affix dictionaries with rules.
adverb and their affix.

68
Table 5. 1 Composition of the lexica based on the category of the stem words and affixes

Verb Noun Adjective Verb Verb Noun Noun Adjective Adjective

Category

prefix suffix prefix suffix prefix suffix

1,300 881 256 18 340 4 70 18 18

Count

5.3. Implementation
Using the free version of a Google Collaboratory, the Guragina spell checker prototype was created
and tested on Python. Python was chosen as the programming language for combination of an
open-source nature; object-oriented architecture, straightforward syntax, rich NLP libraries, cross-
platform compatibility, simplicity, efficiency, and beginner-friendliness make it a suitable choice
for developing the Guragina spell checker prototype. Microsoft Office 2019 for documentation
and Sublime Text are two other software tools utilized in the development of GLSC. Sublime Text
is a popular text editor that is widely used by developers and researchers. It provides a range of
features and functionalities that make it a valuable tool for writing and editing code, including
Python scripts, as well as preparing and manipulating data in various formats. The Windows 11
operating system was used, providing improved speed, security, and productivity enhancements
for a hybrid work environment.
Google Colab was employed for its extensive resources and faster results. It allows writing and
executing Python code in any browser without configuration, providing access to free GPUs and
easy sharing of work. Researchers can import datasets and evaluate models using Colab notebooks,
utilizing the power of Google's hardware, including GPUs and TPUs, regardless of the researcher's
own machine's capabilities. The Guragina Keyman keyboard and Abyssinica SIL font were used
to type and prepare necessary documents in Guragina.

69
5.4. Prototype of the System
The Python programming language is used to implement the Guragina spell-checking system,
which looks for misspelled words and suggests a list of proper ones in their place. The created
prototype has the capabilities of identifying incorrect words, offering suggestions, and, depending
on the calculated distance, replacing incorrect words with right ones from the proposed list.
When a user types a word and clicks the spell check button, the spell checker verifies the word's
accuracy; if the word is misspelled, it is highlighted in red. After that, a selection of suitable
alternative terms for the misspelled word is shown, and the user may select one of these words to
take its place. The misspelled word is then instantly changed with any of the proposed words the
user has chosen from the created list by clicking the replace button.
The user then selects any of the suggested words from the generated list and clicks the replace
button, then the misspelled word is immediately replaced with the word. Figure 5.3 shows the
prototype that displayed after the user makes a replacement for a misspelled word by choosing
from the suggestion list generated by the prototype. As shown in Figure 5.2, there are buttons used
for different purpose as explained in the following

Figure 5. 1 Prototype of the system

5.5. Evaluation of the Ratcliff / Obershelp Algorithm

The main idea behind the Ratcliff algorithm is to find the longest common subsequence (LCS)
between two sequences and use it to calculate the similarity or distance measure. The algorithm
works by iteratively finding the LCS and recursively applying the same process to the remaining

70
portions of the sequences. It has a time complexity of O(N*M), where N and M are the lengths of
the input sequences. This makes it suitable for comparing long sequences efficiently. Additionally,
the algorithm can handle sequences with gaps, which is useful in scenarios where sequences may
have insertions or deletions.

Figure 5. 2 Evaluation distance calculation algorithm using histogram

The results of this study indicate that the Ratcliff algorithm has achieved the following.

Table 5.2 Evaluation distance calculation algorithm

71
5.6. Performance Results of Distance Calculator
This experiment involved a comparison of eight distance similarity measure algorithms applied to
the verb ሴፈ (Sefə) and its inflections. The algorithms evaluated in this experiment include cosine
similarity, Euclidean distance, Manhattan distance, Minkowski distance, Jaccard similarity,
Python's sequence matcher, Levenshtein distance, and Ratcliff/Obershelp. Table 5.2 presents the
calculated distances for each word pair obtained by each distance calculation algorithm.
Table 5. 2 Performance Results of Distance Calculators

The results obtained from Table 5.2 indicate that the Euclidean distance, Manhattan distance,
Minkowski distance, and Levenshtein distance algorithms yield a distance of zero when comparing
identical words. On the other hand, a distance value of one implies similarity for the cosine
similarity, Jaccard similarity, and Python's Sequence Matcher algorithms.
It can be observed that the first five algorithms produce identical distances for the derivations or
inflections of the verb ሴፈ (Sefə). Although these derivations are closer to the stem word, the degree
of closeness is not uniform. This suggests that the Euclidean distance, Manhattan distance,
Minkowski distance, cosine similarity, and Jaccard similarity are not suitable for word-level
comparison in Amharic. Instead, they are more appropriate for comparing documents written in
English or similar languages.
Python's Sequence Matcher algorithm yields better results compared to the previous algorithms.
However, since it employs an exact matching technique, it fails to identify the similarity between

72
words like አትሴፈች and ታትሴፈዋ, despite their close relationship to the stem verb. On the other
hand, the Levenshtein distance algorithm is commonly used for string comparisons in language
processing, counting the number of operations required to transform one word into another.
However, when applied to Guragina, it does not effectively capture the closeness or relationship
between characters within the same family and in the same order. The distances obtained for
‘ትዩትሴፎ’, ‘አትሴፈች’ and ‘ታትሴፈዋ’ are significantly higher than expected. This indicates that
Sequence Matcher and Levenshtein distance are not fully suitable for Guragina string comparison.
Based on these findings, none of the algorithms evaluated are suitable for Guragina. Therefore, a
new distance calculation algorithm is developed and implemented. The verb ሴፈ and its inflections'
distances calculated using the distance calculator are shown in Table 5.2 demonstrates the
distances computed for the same word pairs using the new distance comparison algorithm,
highlighting its ability to identify the closeness between characters within the same family and in
the same order.

5.7. Comparison of the Proposed System with Existing Ones

The main aim of this test is to evaluate the number of correctly accepted inflected and derivate
words and correctly flagged out misspelled words by the spell checker system of Guragina
Language. We also evaluate the performance of the spell checker algrithm by using evaluation
metrics which are represented in Table 5.1
The system shows an average precision and recall of 96% and 99% respectively. Table 5.3
presents a comparison of the evaluation results of the current experiment with the results from a
few other similar previous works focused on Semitic languages. From the figures shown on the
table, Predictive Accuracy of 95% shows that the system has highest possibility of handling any
candidate word accurately, whether the word is valid or invalid. 99% Recall shows that the system
has the highest completeness to label correctly spelled words as valid. Lexical and Error Recall
values of 96% and 89% designate the inclusiveness of the lexicon of the algorithm [39]. It also
shows the efficacy of morphological analysis performed by the spell checker [39].

73
 Table 5. 3 Comparison of Evaluation results with previous similar work
Research Work Accuracy Precession Recall F1 Score
[46] 89.4% - 80.6% 84.4%
[50] 88.62% 28.62% 100% -
[52] 44% 81% - -
[55] 87% 70% 89% 78%
This Work 98.27% 98.07% 97.75% 95.45%

5.8. Discussion
Based on the experimental results presented in Table 5.3, we achieved accuracy of 98.27%
precession of 98.07%, recall of 97.75%, and F1 Score of 95.45%. These findings indicate that our
system performs well with certain limitations.
One of the identified reasons is the limited lexical coverage of the dictionary used. It does not
encompass all the words in the language, including personal names, place names, and scientific
terms. Consequently, some correct words are flagged as invalid due to their absence from the
dictionary. To address this issue, enhancing the stem dictionary by including the stem forms of
these words would improve the system's performance.
The second reason is the mismatching of affixing rules. There is a possibility of applying an affix
rule to a word of a different affix class, resulting in the recognition of meaningless words as correct.
For instance, the prefix ኣ- works correctly with words like ኣ + ጣፈ = ኣጣፈ. However, when applied
to the stem word ኣ+ ኣመ, it produces a word that does not exist in the language (ኣኣመ). Similarly,
the suffix -ኦት functions properly with words like ወረ +ኦት = ወሮት. However, when applied to the
stem word ከና + ኦት, it generates a word that is not part of the language (ከኖት). Addressing these
affixing rule mismatches would contribute to the system's accuracy.
Furthermore, the complexity of Guragina morphology should be taken into consideration.
Currently, there is no developed morphological analyzer or generator specifically designed for
Guragina. Developing a comprehensive morphological analyzer and generator would enhance the
performance of the spell checker system.
In summary, improving the lexicon coverage of the stem dictionary, ensuring accurate
classification of each stem word with the appropriate affix class, and implementing a full-fledged
set of affix rules for the language would significantly enhance the system's performance.

74
 CHAPTER SIX: CONCLUSION AND FUTURE WORK

6.1. Conclusion
As a community, writing using by their own language is very important for individuals,
government and non-governmental organization. Specially, the languages with very low resource
and not well standard, like Guragina the use of writing and spelling checker is more than
facilitating document preparation. The spell checker support for the language development and
standardizing effort by encouraging the user to write by using the languages alphabet.
To achieve this, this study has been done to design a model, implement and develop a prototype
for Guragina Language error spelling checker. It involved study morphology, word derivation,
and spelling errors that can occur in Guragina text writing and development of development spell
checker. In addition, we adopted word formation rules for Guragina Language which can be
integrated to the lexicon used by Guragina spell checker.
In the field of natural language processing, there are several research areas for different languages
of the world. Among these areas, spell checking is one of the research areas where NLP is applied.
Spell checking is about detecting misspelled words in a document and correcting misspelled words
using possible suggestions provided during spell checking. Spell checker performs two main tasks,
namely error detection and error correction.
Various techniques have been proposed to develop a spelling checker for different languages.
Among them, N-gram analysis and dictionary search approaches are applied for error detection
and edit distance, similarity key, neural network, N-gram, rule-based, probabilistic and noisy
channel model are techniques for error correction.
During the process of creating the Guragina Language spelling checker, this study used a rule-
based approach for error correction along with the dictionary search method and the morphological
analyzer (a morphology-based spelling checker) to identify error words and lengths. Only
typographical errors, or non-word errors, in the language are intended to be detected by this spell
checker. The focus on non-word error detection in this research was a strategic decision given the
complexity and limited resources available for the Guragina language. Addressing the non-word
errors is a crucial first step that will provide valuable insights and lessons that can then be applied
towards the more complex task of real-word error detection and correction. Addressing non-word
errors first provides a solid foundation before tackling the more challenging task of real-word error
75
detection and correction. This phased approach allows meaningful progress on this under-
resourced language, rather than attempting to solve the entire spell checking problem all at once.
The intention is to use the non-word spell checker as a starting point, then leverage that knowledge
and capability to progressively tackle real-word error handling in the future, as the abstract
outlines.
As a result, real word errors are not taken into account in this study. Future research will need to
analyze the language's semantics and grammar before introducing real word errors. You can use
the created GLSC independently or as a component of other word processing programs.
By manually entering text in the text editor's provided space or a user can provide a block of text.
The tokenizer separates the text block into individual words, and the morphological
analyzer examines each one using stem and affix dictionaries. The Guragina spelling checker was
developed using more than 6000 stem words from this root dictionary.
Lastly, evaluations and tests are conducted to verify the spelling checking system using sample
data of Guragina Language words randomly collected from papers and books. To verify the
effectiveness of the system's lexical recall, error recall and precision evaluation metrics were
conducted. Based on these evaluation metrics, we obtain accuracy of 98.27% precession of
98.07%, recall of 97.75%, and F1 Score of 95.45%.
In general, we conclude that our morphology-based spelling checker can still perform well, which
needs to be improved, especially more work on the morphology of the language.

6.2. Contribution of the Thesis

The main contributions of this study are described as follows.
• We made up Guragina Stem Dictionary
• We prepared affix pattern of different word categories.
• We introduced a hybrid algorithm by combining both Ratcliff algorithms with a new
distance calculator algorithm that has advanced key features select highly approximate
suggestions, which can be adapted to other morphological complex languages.

76
6.3. Future Work
There are various research areas within natural language processing (NLP) that can be explored
for local languages, and spell checking is one such task. A spell checker serves as a foundation for
other NLP applications, including part-of-speech taggers, grammar checkers, machine translation,
question-answering systems, text-to-speech synthesis, speech-to-text synthesis, anaphora
resolution, text summarization, dialogue systems, and more. For future work in this field, we
recommend the following key activities:
• Increasing the coverage of the stem dictionary's lexicon size would enhance the
performance of the spell checker. This expansion would involve including more stem
words in the dictionary.
• The proposed spell checker system, which relies on morphological analysis, would
benefit from the development of a comprehensive morphological analyzer specifically
designed for the Guragina Language. Currently, a simple morphological analyzer is
used due to the absence of a fully developed analyzer for Guragina.
• Expanding the spell checker to detect and correct real-word errors by considering the
semantics and grammar of the Guragina Language would make the system more
interactive. This would involve incorporating language-specific knowledge to handle
errors beyond non-word errors.
• Extending the work by testing and evaluating the spell checker with a large corpus
that encompasses diverse data sources would provide a more comprehensive
assessment of its performance.
• Developing a standard Guragina Language corpus would be beneficial, as it is
currently a resource-scarce language. Such a corpus would motivate researchers to
explore various NLP applications beyond spell checking and reduce the time required
for corpus collection when evaluating system performance.
• While this work presents a spell checker system as a demo, further projects could
involve developing a fully functional Guragina spell checker that can be seamlessly
integrated with other NLP applications such as machine translation, text
summarization, and more.

77
By pursuing these recommendations, future research in Guragina spell checking and NLP
applications can make significant advancements and contribute to the development of language
technology for local languages. In this work, the researcher concluded that the implementation,
testing and evaluation of system modeling for non-standard word Guragina spelling checker and
correction has fulfilled the objectives of providing a tool with a reasonably good suggestion
support in Guragina language for spelling checker text entry.

78
 REFERENCES

[1] R. Egger and E. Gokce, “Natural Language Processing (NLP): An Introduction,” in

Applied Data Science in Tourism: Interdisciplinary Approaches, Methodologies, and
Applications, R. Egger, Ed. Cham: Springer International Publishing, 2022, pp. 307–334.
doi: 10.1007/978-3-030-88389-8_15.
[2] C. I. Ratnasari, S. Kusumadewi, and L. Rosita, “A Non-Word Error Spell Checker for
Patient Complaints in Bahasa Indonesia,” Int. J. Inf. Technol., p. 5, 2017.
[3] K. Kukich,“Techniques for automatically correcting words in text,” ACM Comput. Surv.,
vol. 24, no. 4, pp. 377–439, Dec. 1992, doi: 10.1145/146370.146380.
[4] F. J. Damerau, “A technique for computer detection and correction of spelling errors,”
Commun. ACM, vol. 7, no. 3, pp. 171–176, Mar. 1964, doi: 10.1145/363958.363994.
[5] S. Singh and S. Singh,“Systematic review of spell-checkers for highly inflectional
languages,” Artif. Intell. Rev., vol. 53, no. 6, pp. 4051–4092, Aug. 2020, doi:
10.1007/s10462-019-09787-4.
[6] R. C. de Amorim and M. Zampieri, “Effective Spell Checking Methods Using Clustering
Algorithms,” Proc.Recent Adv. Nat. Lang. Process.Pages 172–178 Hissar Bulg. 7-13
Sept. 2013, p. 7, 2013.
[7] A. A. Patil and P. R. Sharma,“Study and Review of Selective Spell Checking
Approaches,” vol. 8, no. 2, p. 8, 2019.
[8] S. Sooraj, K. Manjusha, M. Anand Kumar, and K. P. Soman,“Deep learning based spell
checker for Malayalam language,” J. Intell. Fuzzy Syst., vol. 34, no. 3, pp. 1427–1434,
Mar. 2018, doi: 10.3233/JIFS-169438.
[9] Derib Ado, Almaz Wasse, and Janne Bondi, Eds., “The languages of Ethiopia: Aspects of
the sociolinguistic profile,” in IMPACT: Studies in Language, Culture and Society, vol.
48, Amsterdam: John Benjamins Publishing Company, 2021, pp. 1–12. doi:
10.1075/impact.48.int.
[10] S. Völlmin,“Towards a Grammar of Gumer - Phonology and Morphology of a Western
Gurage Variety,” University of Zurich, 2017. doi: 10.5167/UZH-149973.

79
[11] A. Melaku Tilahun and Tesfa Tegegn,“AUTOMATIC SPELLING CHECKER FOR
AMHARIC LANGUAGE,” p. 92, 2017.
[12] K. Peffers, T. Tuunanen,M. A. Rothenberger, and S. Chatterjee, “A Design Science
Research Methodology for Information Systems Research,” J. Manag. Inf. Syst., vol. 24,
no. 3, pp. 45–77, Dec. 2007, doi: 10.2753/MIS0742-1222240302.
[13] K. Kiran, B. V. Kiran, D. C. Sai, G. V. Vamsi, and P. R. Salomi, “Face Mask Detection
Using Machine Learning.” Rochester, NY, Sep. 17, 2021. doi: 10.2139/ssrn.3925736.
[14] “(PDF) Inherent Intelligibility among Guragina Varieties.”
https://www.researchgate.net/publication/312093428_Inherent_Intelligibility_among_Gu
ragina_Varieties (accessed Apr. 12, 2022).
[15] R. Meyer, “Gurage,” Stefan Weninger Ed Semit. Lang. Int. Handb. Berl. N. Y. Gruyter
Mouton 1220-1257, Jan. 2011, Accessed: Jan. 03, 2023. [Online]. Available:
https://www.academia.edu/5533734/Gurage
[16] T. L. Feleke, “Ethiosemitic languages: Classifications and classification determinants,”
Ampersand, vol. 8, p. 100074, Jan. 2021, doi: 10.1016/j.amper.2021.100074.
[17] R. Meyer, “Non-verbal predication in East Gurage and Gunnän Gurage languages,” Crass
Joachim Ronny Meyer Eds 2007 Deictics Copula Focus Ethiop. Converg. Area Afr.
Forschungen 15, Jan. 2007, Accessed: Jan. 03, 2023. [Online]. Available:
https://www.academia.edu/467045/Non_verbal_predication_in_East_Gurage_and_Gunn
%C3%A4n_Gurage_languages
[18] G. Gragg, “The Gunnän-Gurage Languages. Robert Hetzron,” J. East. Stud., vol. 41, no.
3, pp. 231–234, Jul. 1982, doi: 10.1086/372958.
[19] B. Araya Keleta,“Gura Documentation and Description of Morphology and Syntax,”
Thesis, AAU, 2020.Accessed: Dec. 23, 2022. [Online]. Available:
http://etd.aau.edu.et/handle/123456789/24901
[20] Daniel, “(12) A Review of Shifts in Gurage Orthography | Daniel Yacob and Fekede
Menuta - Academia.edu.”
https://www.academia.edu/87133918/A_Review_of_Shifts_in_Gurage_Orthography
(accessed Nov. 28, 2022).

80
[21] R. Meyer, “Language standardization efforts in Gurage,” in 20th International
Conference of Ethiopian Studies, Mekelle, Ethiopia, Sep. 2018. Accessed: Sep. 16, 2022.
[Online]. Available: https://hal.archives-ouvertes.fr/hal-02047237
[22] F. Menuta, “Verbal Extension and Valence in Gumer Variety of Gurage,” Stud. Afr.
Linguist., vol. 51, Aug. 2022, doi: 10.32473/sal.v51i1.121891.
[23] W. LESLAU,“TOWARD A CLASSIFICATION OF THE GURAGE DIALECTS,” J.
Semit. Stud., vol. 14, no. 1, pp. 96–109, Mar. 1969, doi: 10.1093/jss/14.1.96.
[24] “Gurage Adds | PDF | Typefaces | Orthography,” Scribd.
https://www.scribd.com/document/504623447/21037-gurage-adds (accessed Mar. 22,
2022).
[25] A. K. Simpson,“The Origin and Development of Nonconcatenative Morphology,” UC
Berkeley, 2009.Accessed: Nov. 04, 2022. [Online]. Available:
https://escholarship.org/uc/item/7t18f7jw
[26] W. Gewe and M. Gasser, “Learning Morphological Rules for Amharic Verbs Using
Inductive Logic Programming,” May 2012. doi: 10.13140/2.1.5171.2001.
[27] E. Assefa and E. Addis Ababa University,“Major Morphophonemic Operations in Ezha
(Ethio-Semitic),” Macrolinguistics, vol. 7, no. 10, p. 29, 2019.
[28] M. Miestamo, Standard negation: the negation of declarative verbal main clauses in a
typological perspective. Berlin: De Gruyter Mouton, 2005.
[29] C. M. Ford, “Notes on the Phonology and Grammar of Chaha-Gurage,” J. Ethiop.Stud.,
vol. 19, pp. 41–80, 1986.
[30] “17670-EN-morphology-of-the-english-noun.pdf.”
[31] B. W. Debela, “Morphology and Verb Construction Types of Kistaniniya,” Doctoral
thesis, Norges teknisk-naturvitenskapelige universitet, Det humanistiske fakultet, Institutt
for språk- og kommunikasjonsstudier, 2010. Accessed:Dec. 01, 2022.[Online]. Available:
https://ntnuopen.ntnu.no/ntnu-xmlui/handle/11250/244002
[32] F. Menuta,“Morphology of Eža,”Thesis, Addis Ababa University, 2002. Accessed: Jan.
13, 2023.[Online]. Available: http://etd.aau.edu.et/handle/123456789/6370
[33] “IJCTT - Survey of Spell Checking Techniques for Malayalam: NLP.”
https://ijcttjournal.org/archives/ijctt-v17p133 (accessed Jan. 16, 2023).

81
[34] P. Gupta, “A context sensitive real-time Spell Checker with language adaptability,”
ArXiv191011242 Cs Stat, Oct. 2019, Accessed: Mar. 23, 2022. [Online]. Available:
http://arxiv.org/abs/1910.11242
[35] H. L. Liang, “SPELL CHECKERS AND CORRECTORS: A UNIFIED TREATMENT,”
p. 119.
[36] M. M. Al-Jefri and S. A. Mohammed,“Arabic spell checking technique,” US9037967B1,
May 19, 2015 Accessed: Jan. 16, 2023 [Online]. Available:
https://patents.google.com/patent/US9037967B1/en
[37] N. Hossain, S. Islam, and M. N. Huda, “Development of Bangla Spell and Grammar
Checkers: Resource Creation and Evaluation,” IEEE Access, vol. 9, pp. 141079–141097,
2021, doi: 10.1109/ACCESS.2021.3119627.
[38] D. Hládek, J. Staš, and M. Pleva, “Survey of Automatic Spelling Correction,”
Electronics, vol. 9, no. 10, p. 1670, Oct. 2020, doi: 10.3390/electronics9101670.
[39] F. Tafesse, “Morphology Based Spell Checker for Kafi Noonoo Language,” Thesis,
Addis Ababa University, 2018.Accessed: Mar. 28, 2022. [Online]. Available:
http://etd.aau.edu.et/handle/123456789/19584
[40] M. Shimelis, “Amharic Spelling Error Detection and Correction System:,” p. 130.
[41] A. Brhanu,“Tigrigna language spellchecker and correction system for mobile phone
devices,” Int. J. Electr. Comput. Eng., vol. 11, Jun. 2021, doi:
10.11591/ijece.v11i3.pp2307-2314.
[42] R. Altarawneh, “Spelling Detection Errors Techniques in NLP: A Survey,” Int. J.
Comput. Appl., vol. 172, pp. 1–5, Aug. 2017, doi: 10.5120/ijca2017915176.
[43] S.-S. Kang,“Word Similarity Calculation by Using the Edit Distance Metrics with
Consonant Normalization,” J. Inf. Process. Syst., vol. 11, no. 4, pp. 573–582, Dec. 2015,
doi: 10.3745/JIPS.04.0018.
[44] R. Kumar, M. Bala, and K. Sourabh,“A study of spell checking techniques for Indian
Languages,” p. 9, 2018.
[45] D. Yacob, “Application of the Double Metaphone Algorithm to Amharic Orthography.”
arXiv, Aug. 22, 2004. doi: 10.48550/arXiv.cs/0408052.

82
[46] Andargachew Mekonnen, Binyam Ephrem, and A. Nürnberger, “Portable Spelling
Corrector for a Less-Resourced Language: Amharic,” in Proceedings of the Eleventh
International Conference on Language Resources and Evaluation (LREC 2018),
Miyazaki, Japan, May 2018.
[47] G. Assefa, “Automatic Amharic spelling error detection and correction using hybrid
approach,” undefined, 2018, Accessed: Apr. 01, 2022. [Online]. Available:
https://www.semanticscholar.org/paper/Automatic-Amharic-spelling-error-detection-and-
Assefa/ccaf7f78c884c394b8f84f58b4f0433d99118cae
[48] B. Hamza, A. Yousfi, H. Gueddah, and M. Belkasmi,“For an Independent Spell-
Checking System from the Arabic Language Vocabulary,” Int. J. Adv. Comput. Sci.
Appl., vol. Vol. 5, p. 113, Jan. 2014, doi: 10.14569/IJACSA.2014.050115.
[49] K. Shaalan, M. Attia, P. Pecina, Y. Samih, and J. van Genabith, “Arabic Word
Generation and Modelling for Spell Checking,” in Proceedings of the Eighth
International Conference on Language Resources and Evaluation (LREC’12), Istanbul,
Turkey, May 2012, pp. 719–725.Accessed: Jun. 16, 2022. [Online]. Available:
http://www.lrec-conf.org/proceedings/lrec2012/pdf/603_Paper.pdf
[50] G. O. Ganfure and D. D. Midekso, “Design And Implementation Of Morphology Based
Spell Checker,” vol. 3, no. 12, p. 8, 2014.
[51] “GNU Aspell.” http://gnu.ist.utl.pt/software/aspell/ (accessed Jun. 13, 2022).
[52] Vibhakti V. Bhaire, Ashiki A. Jadhav, Pradnya A. Pashte, and Mr. Magdum P.G,
“SPELL CHECKER,” 2017. https://www.ijsrp.org/research-paper-0415.php?rp=P403950
(accessed Jun. 13, 2022).
[53] D. Seth and M. M. Kokar, “SSCS: A Smart Spell Checker System Implementation Using
Adaptive Software Architecture,” in Adaptive Software Architecture. IWSAS, 2005, pp.
187–197.
[54] Banik, Debajyoty and Roy Choudhury, Ritabrata and Ekbal, Asif, “Spelling Checking
Mechanism Based on Layered Language Model Complied with Google Web.” Available
at SSRN: https://ssrn.com/abstract=4517425 or http://dx.doi.org/10.2139/ssrn.4517425

83
[55] Singh, S., Singh, S. HINDIA: “a deep-learning-based model for spell-checking of Hindi
language.” Neural Comput & Applic 33, 3825–3840 (2021).
https://doi.org/10.1007/s00521-020-05207-9

84
 APPENDICES

Appendix 1 Modren Guragina Fider (Alphabet) basic letters

No. Consonant -ə -u -i -a -e -(ɨ) -o

(Transliteration)
1. h ሐ ሑ ሒ ሓ ሔ ሕ ሖ
2. hʲ 𞟠 𞟡 𞟢 𞟣 𞟤 𞟥 𞟦
3. hʷ 𞟨 𞟩 ሗ 𞟪 𞟫
4. l ለ ሉ ሊ ላ ሌ ል ሎ
5. m መ ሙ ሚ ሜ ም ም ሞ
6. mʷ ᎀ 𞟭 ሟ 𞟮 ᎃ
7. r ረ ሩ ሪ ራ ሬ ር ሮ
8. s ሰ ሱ ሲ ሳ ሴ ስ ሶ
9. ʃ ሸ ሹ ሺ ሻ ሼ ሽ ሾ
10. k’ ቀ ቁ ቂ ቃ ቄ ቅ ቆ
11. k’ʲ ቐ ቑ ቒ ቓ ቔ ቕ ቖ
12. k’ʷ ቈ 𞟰 ቋ 𞟱 𞟲
13. b በ ቡ ቢ ባ ቤ ብ ቦ
14. bʷ ᎄ 𞟳 ቧ 𞟴 ᎇ
15. t ተ ቱ ቲ ታ ቴ ት ቶ
16. ʧ ቸ ቹ ቺ ቻ ቼ ች ቾ
17. n ነ ኑ ኒ ና ኔ ን ኖ
18. ɲ ኘ ኙ ኚ ኛ ኜ ኝ ኞ
19. ’ አ ኡ ኢ ኣ ኤ እ ኦ
20. k ከ ኩ ኪ ካ ኬ ክ ኮ
21. kʲ ኸ ኹ ኺ ኻ ኼ ኽ ኾ
22. kʷ ኰ 𞟵 ኳ 𞟶 𞟷
23. w ወ ዉ ዊ ዋ ዌ ው ዎ
24. ʔ ዐ ዑ ዒ ዓ ዔ ዕ ዖ
25. z ዘ ዙ ዚ ዛ ዜ ዝ ዞ
26. ʒ ዠ ዡ ዢ ዣ ዤ ዥ ዦ
27. j የ ዩ ዪ ያ ዬ ይ ዮ
28. d ደ ዱ ዲ ዳ ዴ ድ ዶ
29. ʤ ጀ ጁ ጂ ጃ ጄ ጅ ጆ
30. g ገ ጉ ጊ ጋ ጌ ግ ጎ
31. gʲ ጘ ጙ ጚ ጛ ጜ ጝ ጞ
32. gʷ ጐ 𞟸 ጓ 𞟹 𞟺
33. t’ ጠ ጡ ጢ ጣ ጤ ጥ ጦ
34. ʧ’ ጨ ጩ ጪ ጫ ጬ ጭ ጮ
35. s’ ጸ ጹ ጺ ጻ ጼ ጽ ጾ
36. f ፈ ፉ ፊ ፋ ፌ ፍ ፎ
37. fʷ ᎈ 𞟻 ፏ 𞟼 ᎋ
38. p ፐ ፑ ፒ ፓ ፔ ፕ ፓ

85
 39. pʷ ᎌ 𞟽 ፗ 𞟾 ᎏ
Appendix 2 Numeral (Cardinal) Pronunciation in Guragina

Numeral Translation Tens Translation

at ኣት One asɨrəm ኣስረም ኣት eleven
w
h ɛt 𞟪ት Two as
at ɨrəm ኣስረም 𞟪ት twelve
sost ሶስት Three aswɨrəm ኣስረም ዠጠ nineteen
h ɛt
arbət ኣርበት Four h uja
ʒəʷt’ə 𞟫ያ twenty
amɨst ኣምስት Five sasa ሳሳ thirty
sɨdɨst ስድስት Six arba ኣርባ Forty
səbat ሰባት Seven amsa ኣምሳ Fifty
sɨmwɨu ስᎃት Eight sɨlsa ስልሳ Sixty
t t’ə
ʒə ዠጠ Nine səba ሰባ seventy
asɨr ኣስር Ten səmanəj ሰማነያ eighty
a t’əna
ʒə ዘጠና ninety
bəqr በቅር hundred
Hum 𞟫ም thousand
Appendix 3 Derivation of Ordinal numbers form Cardinal

Cardinal Number Ordinal Number

Transl Guragina Translatio Transliteration Guragin Translati
at
iterati አት one
n atənə አት-አነ
a first
on
w
h
onɛt 𞟪ት two hwɛtənə 𞟪ተ-አነ second
səbat ስባት seven səbatənə ስባት-አነ seventh

86
 Declaration
I, the undersigned, declare that this thesis is my original work and has not been presented for a
degree in any other university and that all sources of materials used for the thesis have been duly
acknowledged.

Declared by:

Name: Mengistu Gebre Bokane

Signature: _________________________________________

Date: _____________________________________________

Confirmed by advisor:

Name: Dr. Yaregal Assabie

Signature: __________________________________________

Date: ______________________________________________

87