Online Machine Translation Efficiency in

Translating Literary Expressions Between

English & Arabic
(Proverbs as a case-in-point)
Ibrahim Jibreel*1
Department of English & Translation, Faculty of Human & Social Sciences, University of Science and Technology,
Yemen

Abstract----Doubtless, Machine Translation has affected translation as a process and a product. This study tests MT's
effectiveness in translating proverbs between English and Arabic. It investigates one important CAT tool device and would
add to the translation technology field as well as guide students of translation and translators, especially literary translators,
to the most effective MT in translating proverbs. It aims to attest which MT will be more communicative, semantic or literal
giving target equivalent and clarifying the error type the MT would make. Furthermore, it aims to test if there are significant
differences between the results of these MTs regarding translation methods. To achieve these aims, thirty proverbs, half
Arabic and half English, have been randomly selected, taken from The Dictionary of Common English Proverbs Translated
and Explained written by Attia (2004) and then translated using five different online MTs: Google, Reverso, Yandex, Systran,
and Bing. As Alabbasi (2015) suggested, the researcher adopted Newmark's (1988) Taxonomy of translation methods,
selecting three major divisions that include the other types in one way or another viz. Literal, Semantic and Communicative.
Analyzing data, Kruskal-Wallis Test and Chi-square were used as well as descriptive statistics. It is found that the most
translation method MT produced when faced with a proverb is the literal, semantic and communicative respectively. Bing is
the most effective MT for translating proverbs providing communicative equivalents. Bing and Google Translate, in the same
rank, provide semantic equivalents. No statistically insignificant differences were found among MT effective translations.
Classifying errors diverge between missing the implied meaning, producing weakly structured translations, choosing wrong
synonyms and distorting the meaning.

Index Terms----Bing Translate, Google Translate, Machine Translation, Proverbs, Reverso Translate, Systran
Translate, Translation Methods, Yandex Translate

1. INTRODUCTION
In such a time when millions of people travel around the planet, by choice or due to economic, business, or political
reasons, the translation of the spoken and written word is now of ever-increasing importance. This planet we are living
on becomes similar to a small village where people are obliged to communicate with other people of different cultures
and languages in order to continue living in one way or another. Therefore, there emerged an urgent need for translation
and translators to link these completely faraway societies together.
Newmark (1988, p. 5) provided us with the most prominent definitions of translation stating that "translation is
rendering the meaning of a text into another language in the way that the author intended the text". Whereas Newmark
stressed the idea of transferring the meaning in the above-mentioned definition, Nida and Taber (1982. p.12) stated:
"translating consists of in the reproducing in the receptor language the closest natural equivalent of the SL message,
firstly in terms of meaning and secondly in terms of style".
Translation is a difficult craft to master, as it requires not only the transferring of meaning but also providing a natural
equivalent with respect to cultural and lexical differences between languages. In fact, culture and translation are
inseparable. However, as the phenomenon of translation is in continuous development, it is almost impossible to know
and grasp all the languages present in the world by human beings. In fact, it is rare for one person to speak more than
four languages fluently, not to mention to have an overall knowledge of these languages' standards, contexts, and the
cultures accompanying them. Therefore, translation researchers are looking for all possible methods and tools to
facilitate this process and provide all people, including those who can only speak their native language, with the ability
to communicate with others and understand them easily, effectively and quickly.

*
Corresponding author, email: [email protected]
 One of these tools is machine translation, a branch of computational linguistics that focuses on the use of e-devices to
render a speech or text from an SL to a TL. Even though there is not any guarantee that machine translation is of high
quality all the time, many programs are able to provide powerful and beneficial outputs within a limited time and
constraints.
Translating cultural expressions is one area of difficulty that translation students and even professionals suffer from.
Nowadays, and with the help of Computer-Assisted Translation (CAT) tools, this problem may be mitigated. It is an
attempt to help translators, especially novices, to find the most appropriate machine translation engine that helps them
in facing this difficulty.
1.1 SIGNIFICANCE OF THE STUDY
English-Arabic translation is becoming a widespread area for work and learners as well as in almost all walks of life,
especially for cross-cultural purposes. Therefore, the use of MT has increased a lot recently, and these machine
translation websites' developers are trying to make them more and more effective every single day. This research is
significant as it aims to ensure these websites' capability to translate cultural expressions as effectively and naturally as
possible.
2.1 AIMS OF THE STUDY
The purpose of this study is to test the efficiency of MT in translating proverbs and provide an equivalent that
affects the target community without distorting the meaning or losing the cultural effect of the source cultural
expressions. It also attempts to decide which of these machine translation websites are going to give the closest natural
and cultural equivalents based on the translation approach utilized by these MTs, which are the literal, semantic, and
communicative methods. Additionally, it explores whether there are any significant differences among MT engines in
proverbs' translations. Furthermore, it analyses the problems found in the translations produced by these MT engines,
identifies and classifies them.

II. LITERATURE REVIEW

A. MACHINE TRANSLATION
The automatic use of computers to translate from one language to another is known as Machine Translation (MT).
Terminologically, it is called Computer-Assisted Translation (CAT) tools belonging to the field of artificial intelligence;
a field specialized in developing software programs that can stimulate human thinking. Sofer (2006) defines MT as
"a term used to describe translation performed by a computer software program, as an alternative to human translation,
performed by a human translator" (p. 83). For Balkan (1992) MT refers to "any system that actually performs a
translation" and classifies "any other computerized translator tool which falls short of translating as a CAT device"
(p. 408). Currently, these technologies are mostly available through websites like Google.translate.com, Microsoft
Bing.com, ReversoContext.com, BabelFish, etc. They translate text algorithmically from an SL to a TL. The main focus
of MT is on employing computers to aid and support humans as they translate from one language to another.
B. MACHINE TRANSLATION AIM
Vauquois (1998) describes MT as "being aimed at enabling a computer to transfer natural language utterances, or to
process a natural language in terms of lexical, syntactic and semantic dimensions" (as cited by Lin & Chien, 2009,
p. 134). Acikgoz and Sert (2006) emphasized that one prominent reason behind the world's globalization is machine
translation.
Many linguistics and scholars who had witnessed the beginning of MT had expected that those laughable outputs of
that time are going to get better and better. They anticipated the future of MT and that one day, humans might have to
just edit, revise or proofread computer translations. However, they never believed that a computer could understand the
text the same way a human could. Although a machine would never understand, Champollion (2001) contended that it
can translate.
Some other scholars were completely against the idea of depending on MT. For Thriveni (2002) one language cannot
adequately convey the meaning of another because speakers of various languages tend to think in distinctly different
ways. She maintains that as an MT cannot easily expose literature, cultural sensibilities in the text, or speeches, cultural
interpretation and identification by a translator should be a more accurate manner to translate.
C. PROCESS OF MACHINE TRANSLATION
Machine translation includes speech translation and word translation. It relies on four main techniques: (a) word
analysis, (b) grammar analysis, (c) meaning analysis, and (d) style analysis. To clarify the process, the sentence is first
divided into each word; then, the meaning of each word is subsequently clarified using the machine database's
electronic dictionary. After that, the meaning of the sentence is analyzed according to grammar rules and then
transformed into concept constructions. Finally, the language model is used to generate the target language. Vauquois
(1968) provided a diagram, called later the Vauquois Triangle. The procedure includes analyzing the ST, transferring
data from a source representation to a target representation, and then creating the TT. This diagram is presented in
Figure one:
 Figure 1. Machine translation, Vauquois Triangle 1968

D. MACHINE TRANSLATION IN TRANSLATION STUDIES

A very important area of translation studies is now machine translation. Several types of research were conducted
regarding MT such as Lembersky, Ordan, and Wintner, (2012), Ali (2016), Anderson (1995), Belam (2003) etc. Several
studies attempted to either compare MT to HT or to MTs' themselves. For the former, Ismajli and Maliqi (2021) studied
the efficiency of HT versus MT. They indicated that MTs:
"have proven to be a truly big breakthrough since such applications employed the practice of post-editing – an MT
system outputs an initial translation and a human translator edits it for correctness, ideally saving time over translating
from scratch" (p. 307).
Li, Graesser, and Cai (2014) evaluated Google English translation comparing it to human English translation and the
original Chinese. They found a significant correlation between Google translation with the Chinese in formality even
though it is not that great. Latent Semantic Analysis (LSA) and Content Word Overlap (CWO) showed that both
translations were highly correlated with one another in terms of cohesion. However, Google Translation had higher
correlations with the Chinese than human translation. Almahasees and Mahmoud (2022) without human translators,
attempted to evaluate the effectiveness of Google Image Translator in translating Arabic-English language on signage.
Besides, the study intended to determine Google's ability to analyze the TT of its picture service in terms of
orthography, grammar, lexis, and semantics using Costa, Ling, Lus, Correia, & Coheur's (2015) Linguistic Error
Analysis Framework. The study demonstrated that Google Translate makes mistakes such as incorrect translation,
omission, addition, incorrect choice, incorrect ordering, subject-verb disagreement, and semantic mistakes. They
concluded that Google Image Translation service helps in configuring the overall message. Thus, a human translation
cannot be excluded for its adequacy and effectiveness.
León Bergasa and Lorés Sanz (2019) conducted a contrastive analysis of tourist text errors in three MTs, namely
Google Translator, Systran, and Bing, on the one hand, and human translation, on the other, in order to compare the
suitability of various MT engines. They discovered Google Translator to be the most reliable in terms of machine
translation systems' accuracy and fluency, and Systran to be the least accurate. There are more spelling mistakes in
human translation than any other source. It is unusual to encounter this kind of inaccuracy in machine translations that
use lexical unit retrieval from the internet. They concluded that MT may be useful as a starting point since the
translations are not fully adequate. Therefore, they insisted on post-edition for obtaining better results in similarity to
the results obtained by Sakre (2019) conducted on business texts.
E. TRANSLATING PROVERBS USING MT
As for Barbour (1963), there are various sources of proverbs. According to Mieder (1994), a proverb is a brief, well-
known expression of the people that encapsulates knowledge, morality, and traditional beliefs in a symbolic, fixed, and
memorably structured form that is passed down through generations. According to Alshammari (2015), this effective
form of communication serves both literary and practical purposes. Besides, they are "special, fixed, unchanged phrases
with particular, fixed, unchanged meanings," according to Ghazala (1995, p. 138).
Generally, in a cultural context, Latief, Saleh and Pammu (2020) looked at the efficiency of MT to create the
technology-based language translation system in Indonesia utilizing Google Translate (GT). The results demonstrated
that written translation outperforms image translation using GT. Consider the cultural context as expressive in
interaction, and GT, as one of the technological communication tools available, Google translation must be used
effectively to overcome language translation systems and cultural barriers. Studying proverbs' translation using online
MT, Al-khresheh and Almaaytah (2018) evaluated GT's accuracy in translating several English proverbs into Arabic.
Limited to a small number of randomly selected English proverbs and using only GT, the findings revealed that when
rendering the same meaning of English proverbs into Arabic, 'Google Translate' had some linguistic issues, particularly
with multiple-meaning words, which were discovered to present numerous challenges and difficulties to online
translation. This was clear in the form of literal translations, incorrect TL equivalents, incorrect word order,
grammatical mistakes and inappropriate lexical words. The researchers concluded that the level of accuracy is not likely
to be accurate and unquestionable. Sharma and Goyal (2011) used Birla, Ahmed, and Shukla's (2009) algorithm to
extract Multiword Expressions from an English text. They used the extraction process for multi-word expressions and
mono-word-meaning to find the appropriate words for the lexical database. The researchers used that algorithm in MT
to render proverbs. Using the Relational Data Approach, they tested over 800 Hindi to Punjabi proverbs.
F. GOOGLE TRANSLATE AND OTHER TRANSLATION ENGINES
Several studies, including Al-Kabi, Gigieh, Alsmadi, Wahsheh, and Haidar (2013), Abdulhaq (2016), and Jabak
(2016), used Google Translate (GT) because it is the most widely used and popular MT for both Arabic and English
(2019). They concluded that MT produces literal translations that do not have the same effect as the original. It can,
however, handle simple sentences and idiomatic expressions. They recommend that MT should be fed with complex
and metaphorical expressions and their solutions in addition to the technical expressions and terms.
Comparing GT with other translation engines, Ali (2020) evaluated the English-Arabic TT translation of selected UN
records using three MTs viz. Google Translate, Microsoft Bing, and Ginger. The ST was divided into 84 meaningful
passages. After correcting the TT using two translation factors: fidelity and intelligibility, the results show that none of
the three translation engines perfectly translated the STs, with a preference for Microsoft Bing translation as the best
one. Comparatively, Ginger’s translation was the most accurate followed by Microsoft Bing and Google translation
respectively.
When Polat, Zakirov, Bajak, Mamatzhanova, and Bishkek (2018) compared Google Translate to Yandex Translate
for translating Kyrgyz proverbs into English and Turkish at the lexical, semantic, and syntactic levels, they discovered
that Google Translate was more accurate than Yandex Translate at the lexical, semantic, and syntactic levels in
translating phrases and sentences of proverbs expressions. The most common errors generated by the two MTs were
verb tense, comma, and spelling, according to error analysis of grammatical items. Using a set of 100 proverbs from
English into Arabic, Hamdi, Nakae, and Okashs (2013) investigated the translation accuracy of Google Translate, Bing
Translator, and SDL free translation. They found that only a few proverbs were accurately translated and that Google is
slightly better than the two.
G. COMMON ERRORS OF MACHINE TRANSLATION
According to Hamdi et al. (2013), when it comes to translating proverbs, online translation tools face a variety of
challenges, such as displaying a literal translation when the accurate translation is not stored in its memory. When
dealing with linguistic structures that require a higher level of accuracy, such as proverbs, several linguistic and
technical issues arise.
For Vilar, Xu, d'Haro, and Ney (2006), the following five points summarize the most common errors made by MT,
namely "(a) missing words, (b) word order, (c) incorrect words, (d) unknown words, and (e) punctuation errors". P, 698.
It indicates that the system was unable to find the correct translation, and this point has five sub-points: sense, incorrect
form, extra words, style, and idioms. The last idiom subcategory is concerned with translating cultural expressions in
general. The MT system does not recognize these idioms and instead translates them literally, resulting in serious errors.
H. TRANSLATION METHODS
Translation methods are presented in three main headings: procedures [Vinay and Darblent (1958)]; methods
[Newmark (1988), Ghazala (1995) and AL-Abbasi (2010)] and strategies [Baker (1992), Venuti (1998), As-Safi (2002)
and Pederson (2007)]. However, most of them classify those procedures, methods or strategies into two main types. It is
worth mentioning that most of the scholars discuss the idea that there are two extreme dimensions. The first, if its
procedures or strategies are followed, leads to literal [Vinay and Darblent 1958, Ghazal1995]; SL-Oriented [Newmark,
1988 and Pederson, 2007]; and what others like Venuti (1998) and As-Safi (2002) have called Domesticating and
General Strategies respectively. On the other extreme, following some procedures lead to Oblique translation [Vinay
and Darblent, ibid]; TL-Oriented [Newmark, 1988 and Pederson, 2007]; free [Ghazala (1995)]; foreignizing [Venuti
(obcit)]; and specific [As-Safi (2002)]. In addition, Baker (1992) classifies the strategies into two types also, but some
strategies work at the word level and others work at the sentence level. In this regard, a researcher cannot escape Nida's
(1964) formal equivalence and Nida and Taber's (1982) dynamic equivalence.
I. NEWMARK'S (1988) TAXONOMY
In general, Newmark (1988) divided translation methods into two categories: SL-oriented and TL-oriented. The
methods used in the case of the first are word-for-word translation, literal translation, faithful translation and semantic
translation. On the other hand, the TL-oriented methods are adaptation, free translation, idiomatic translation and
communicative translation.

III. METHODOLOGY

This study is descriptive-qualitative with a content analysis method. However, quantitative analysis is applied to help
reach accurate results.
 A. SOURCE TEXT SELECTION
The source of data is 30 proverbs, half English and half Arabic, selected randomly from The Dictionary of Common
English Proverbs Translated and Explained by Mohammad Attia, published in 2004. This book provided more than
1500 common English proverbs with their standard translation in Arabic as well as an explanation for the proverb's
intended meaning.
B. MACHINE TRANSLATION SELECTION
As other previous studies focused on Google Translate only, this study intended to test a variety of other MTs in an
attempt to look for better options and outputs. Proverbs were translated using five different machine translation engines,
which are Google Translate, Reverso Translation, Yandex Translator, Systran Translate, and Bing Microsoft
Translator. This selection was based on popularity, free service and most importantly providing the Arabic language.
After the outputs of the proverbs were provided, the analysis was done accordingly and the problems found in these
outputs were presented and discussed.
C. TRANSLATION METHODS ADOPTED
In making a decision for each MT engine translation output for each proverb, the researcher adopted Newmark's
(1988) Taxonomy of translation methods by selecting three major divisions that include the other types in one way or
another, as suggested by Alabbasi (2015), viz. Literal (L), Semantic (S) and Communicative (C).
D. PROCEDURES & ANALYSIS
The 30 ST proverbs (15 are English & 15 are Arabic) have been listed aligned to their Standard Translations as
provided in The Dictionary of Common English Proverbs Translated and Explained written by Attia (2004). Then,
under each Machine Translation engine, the target text translation is written starting with Systran Translate followed by
Google Translate, Reverso.Com, Yandex Translate and Microsoft Bing Translate respectively.
To make a decision for each MT engine output for each proverb, Literal (L), Semantic(S) and Communicative (C)
methods were the categories adopted and written under each MT's translation of the ST proverb.
Using the PSS analysis, the three decisions for each MT output were encoded as 1= Literal,
2= Semantic & 3=Communicative. In the analysis, Kruskal-Wallis Test, Qi-Square and other descriptive statistics were
used.

IV. RESULTS & DISCUSSION

A. MACHINE TRANSLATION VS. TRANSLATION METHODS

TABLE 1
MT ENGINE'S EFFECTIVENESS IN RELATION TO TRANSLATION METHODS
Machine Translation Total
Translation
Systran Google Reverso Yandex Bing Total %
Methods
Freq. 16 13 18 19 12 78
Literal 52.0
% 20.5 16.7 23.1 24.4 15.3 100
Freq. 11 12 8 9 12 52
Semantic 34.7
% 21.2 23.1 15.3 17.3 23.1 100
Freq. 3 5 4 2 6
Communicative 20 13.3
% 15 25 20 10 30
Total 30 30 30 30 30 150 100

As Table (1) shows the most dominant method in all these MTs is obviously the Literal with a percentage of 52.0%,
followed by the Semantic with a percentage of 34.7%, and then finally comes the Communicative method with a
percentage of 13.3%. This result is similar to those of Al-Kabi et al. (2013), Abdulhaq (2016), and Jabak (2019). They
found that MT gives literal translations and cannot produce the same effect as the original.
If, however, focusing on the MT effectiveness of each engine on its own, even though the percentage is not that high,
Bing Translate has the highest rank in providing communicative translations with a percentage of 30%(N=6) and it is
also of the lowest rank in providing Literal equivalents with a percentage of 15.3%(N=12). On contrary, it could be
noticed, as well, that the least effective one is obviously Yandex in providing communicative equivalents, which is 10%
(N=2) and the highest rank in providing Literal translation 24.4%(N=19). It is then followed by Google Translate,
which provided communicative equivalents with a percentage of 25% (N=5). Regarding Semantic equivalents, both
Bing Translate and Google Translate have got the highest rank with a percentage of 23.1%(N=12) each, followed by
Systran at 21.2% (N=11) and Yandex with a percentage of 17.3%(N=9). The least effective one is Reverso Translate
with a percentage of 15.3%(N=8). Comparing these results to that of Hamdi et al. (2013), results were different. Google
Translate provided accurate translation slightly better than the other two. Testing the accuracy and fluency of Google
Translate, Systran and Bing, Bergasa and Sanz (2019) discovered that GT produces the most reliable equivalent and
Systran produces the least reliable one and that Bing is more reliable and accurate than Systran.
 B. SIGNIFICANCE OF TRANSLATION METHODS
TABLE 2
LEVEL OF SIGNIFICANCE

Sig.
Proverbs Translations N Mean Rank Chi-Square df
(P-value)
Machine Translation Literal 78 74.73
Engine
Semantic 52 74.92
.259 2 .879
Communicative 20 80.00

Total 150
a. a *P-value ≥ 0,05 is significant.
b. Grouping Variable: Proverbs Translations
c. Kruskal-Wallis Test: the non-parametric test equivalent of the ONE WAY ANOVA statistic, used for comparing different MTs.

From Table 2, it is clear that 2=.259, df= 2 and the *p-value= (.879) indicate variations among the MT engines
regarding the methods of translations which is statistically insignificant. This result is similar to that of Hamdi et al.
(2013). They pointed out "Although Google seemed to have better scores than others, the difference was insignificant".
This may be due to the fact that most online translation tools depend on internet databases and similar syntactic
architectures.
C. COMPARISON OF THE EFFECTIVENESS OF MACHINE TRANSLATION
(a.) TEST OF NORMALITY
To compare MTs, a test of normality was carried out using Kolmogorov-Smirnov Test.
TABLE 3
ONE-SAMPLE KOLMOGOROV-SMIRNOV TEST

Machine
Translation Proverbs
Engine Translations
N 150 150
Mean 3.00 1.61
Normal Parameters(a,b)
Std. Deviation 1.419 .712
Most Extreme Differences Absolute .160 .325
Positive .160 .325
Negative -.160 -.195
Kolmogorov-Smirnov Z 1.954 3.986
Asymp. Sig. (2-tailed) .001 .000
a Test distribution is Normal.
b Calculated from data.
Table 3 indicates that no normal distribution between the performance of MTs since the p-value is less than (.05).
Therefore, the non-parametric Kruskal-Wallis test was used as an alternative to ONEWAY ANOVA.
(b.) NPAR TESTS: KRUSKAL-WALLIS TEST
TABLE 4
COMPARISON OF MT TRANSLATIONS
Test Statistics (a,b)
Ranks
Machine Translation Proverbs Translations
Engine (N=5) Mean Rank Rank Chi- df Sig.
Square (P-value)
Proverbs Translations Systran Translate 73.43 3
Google Translate 82.33 2
(N=30)
Reverso.com 70.30 4 5.408 4 .248
Yandex Translate 65.73 5
Microsoft Bing Translate 85.70 1

d. *P-value ≥ 0,05 is significant.

 e. Grouping Variable: Machine Translation Engine
f. Kruskal-Wallis Test: the non-parametric test equivalent of the ONE WAY ANOVA statistic, used for comparing different MTs.
Exploring the most effective MT, Kruskal Wallis Test was conducted. Results show that   5.408 , df=4 and the
2

*p-value (.248) is statistically insignificant among MT effective translation performance in favour of Bing Translate.
Observing the mean ranks, Bing Translate produces the highest quality translation m=85.70. It is also clear that Google
Translate is better than other MTs (m= 82.33). In the third rank is Systran m=73.43, followed by Reverso.com
(m=70.30) and Yandex (m=65.73) respectively.
(c.) THE OVERALL JUDGEMENT
To determine the overall degree of Translation Method (TM) produced by each translation, the following grading
rubric was applied.
TABLE 5
GRADING RUBRIC OF TM OF MT
Code 1 2 3

Description Literal Semantic Communicative

Value 1 ≥ 1.6 1.7 ≥ 2.4 2.5 ≥ 3
% 0 ≥ 33.3% 33.4% ≥ 66.7% 66.8% ≥ 100%
After processing data according to the mentioned grading rubric, the following results are obtained as shown in the
crosstabulation in Table six.
TABLE 6
APPROACHES OF TRANSLATIONS VS MT
Mean Std. Deviation Result Rank
Machine Translation Engine (N=5)
Systran 1.57 .679 Literal 3
Google 1.73 .740 Semantic 2
Reverso 1.53 .730 Literal 4
Yandex 1.43 .626 Literal 5
Bing 1.80 .761 Semantic 1
Total 1.61 .712

Table 6 indicates that all the translation engines result either in Semantic or Literal translation. In particular, Bing and
Google are mostly producing Semantic Translation with priority to the former on the latter (m=1.80) vs. (m=1.73). On
the other hand, Yandex, Reverso and Systran generally provide Literal Translation with means values of (1.43, 1.53 and
1.57) respectively.
(d.) ANALYSIS OF DETECTED ERRORS
Having analyzed the outputs of each and all the MTs used in this study, these are some of the errors detected.
The researcher has attempted to classify them due to the reason/s behind the difficulty. Examples are both English
source-based and Arabic source-based.
1. MISSING THE IMPLIED MEANING
TABLE 7
MISSING THE IMPLIED MEANING EXAMPLES

ST Proverb
All work and no play makes Jack a dull boy
MT

Google ً‫كل عمل وال لعب يجعل جاك صبيا ً ممل‬

Bing ً‫كل األعمال وليس اللعب يجعل جاك صبيا ً ممل‬

Yandex ً‫كل العمل وال اللعب يجعل جاك صبيا ً ممل‬

Systran ً‫كل العمل وال اللعب يجعل جاك ولدا ً ممل‬

Reverso ً‫كل عمل وال لعب يجعل من جاك ولداً ممل‬

In English culture, this proverb is said to warn someone that he will not be an interesting person by working all the
time. In Arabic, the possible communicative equivalent is ‫ ( ساعة لقلبك وساعة لربك‬Attia 2004). Literally, it means that you
should divide your time between worshiping God and enjoying your life equally. As the implied meaning is a general
piece of advice, there is no actual relation between this advice and Jack or worshiping. In this example, however, MT
failed to capture the implied meaning and tended to translate the above proverb in an utterly literal translation instead.
A person who has no clue about this proverb or the SL culture might be completely confused if faced with this proverb
in any context. Perhaps, the first question crossing his mind would be 'Who is Jack?'. It is so frustrating that, though the
internet is full of specialized online proverb dictionaries, MT is still incapable of capturing the implied meaning of most
proverbs, even in English.
2. WEAKLY STRUCTURED TRANSLATIONS
TABLE 8
WEAKLY STRUCTURED TRANSLATIONS EXAMPLES
ST Proverb

Beggars must not be choosers

MT

Google ‫يجب اإال يختار المتسولون‬

Bing ‫يجب اأال يختار المتسولون‬

Yandex ‫يجب أن يكونوا شحاذين ال مختارين‬

Systran ‫ال يجب أن يكون المتسولون مختارون‬

Reverso ‫يجب أن يكن المتسولون ال يختارون‬

Table 8 shows that most of the translations are grammatically weak. 'Beggars must not be choosers' is a proverb said
when you know you have no choice but to accept an offer or situation because it is the only one available. Whereas a
communicative equivalent could be )‫ (بلش وقال أوزنه‬or ( ‫) شحات ويتشرط‬, a proper semantic equivalent might be ( ‫ليس للشحاذ‬
‫ ) خيار‬or (‫) السائل ال يختار‬. When translated to Arabic, only Google and Bing provided well-structured translations, though
still somehow literal. However, it is shameful that the rest could not even provide well-ordered sentences. In fact, the
translation provided by Reverso Translation is completely messy and wrongly ordered. A reader can still capture the
intended meaning of a proverb if translated literally in some cases, but it is still unacceptable for MT to generate weakly
structured and wrongly ordered sentences. In some cases, word order and some grammatical errors might change the
whole meaning of a sentence.
3. CHOOSING THE WRONG SYNONYM
Using wrong words is a common error of machine translation. Even though smart technology becomes smarter and
smarter every passing day, a machine is still unable to think the same way as men, at least up to now. Once a human is
faced with a word to translate, s/he is obliged to consider all the possible near-synonyms of this word in the TL and
decide which one is the most suitable and proper to serve the intended meaning. This is decided based on the context,
the translator's understating of that context, and sometimes even the writer's intention. Being that hard and complicated
for a human, is it going to be easier for a machine to understand the context and choose the right synonym?

TABLE 9
CHOOSING THE WRONG SYNONYM
ST Proverb

‫اطلبوا العلم من المهد إلى اللحد‬

MT

Google Seek knowledge from the cradle to the grave

Bing
Ask for knowledge from the cradle to grave

Yandex Order the flag from cradle to grave

Systran Seek science from the cradle to the grave

Reverso Ask for science from the cradle to the limit

 This example in Table 9 shows how MT chose to translate the words )‫ (عِلم‬and the word ( ‫) ا َللحد‬. This proverb, again,
has a similar equivalent in English, which is (Seek learning from the cradle to the grave). The first word, pronounced as
ilm means, in this context, learning or knowledge. However, another word pronounced and spelt just the same has
another meaning, which is "science". The two meanings might look similar to some extent in this context, but both still
have their own proper and suitable contexts. Whereas the proverb meant seeking knowledge and learning in general,
"Science encompasses the systematic study of the structure and behavior of the physical and natural world through
observation and experiment", Oxford Reference. Yandex, on the other hand, went so far as to translate the same word as
"flag". In Arabic, the equivalent synonym of the flag is "‫"عَلم‬, pronounced as Alam, which is a homograph for the
original word ilm. Reverso had also chosen to translate the word (‫) لَحد‬, pronounced as lahd, into (limit) because it is also
a homograph of the word (‫) َحد‬, pronounced as hadd. It also considered the first two letters as the Arabic definite article (
‫) الـ‬. In the beginning, the researcher thought that this error of mistranslation might be solved by adding diacritics to the
original words. Diacritics are marks that can appear above or below letters to alter their pronunciation. However, even
after adding these marks on the two words (‫ )عِلم‬and (‫)اللَحد‬, it still did not help the MT to recognize the intended meaning
and it still translated the two words as (flag) and (limit) instead of (cradle) and (grave).
In other words, near-synonyms and homonyms are a big reason behind many of the errors made by machine
translation and a matter of confusion that machine translation still cannot overcome. A person who does not understand
English might be clueless about this whole change of meaning only because MT could not choose the right and proper
word among many other alternatives. Obviously, it goes back to the fact that a machine still cannot always understand
the context well. It is still unable to weigh all the options available and decide which serves the context the best.
Perhaps, it is also because a machine will always be incapable of understanding the writer's intention. At least, not to
the extent a man can.
4. DISTORTING THE MEANING
TABLE 10
DISTORTING THE MEANING
ST Proverb
‫تحت السواهي دواهي‬
MT

Google Translate Under the causes of disconnect

Bing Microsoft
Still water runs deep
Translate

Yandex Translate Under the sawahi dawahi

Systran Translate Under the coast dwai

Reverso Translation Under Suahi Dohai

The above proverb is actually a well-known Arabic proverb; a person who looks quiet and clueless, called "‫ "ساهي‬in
Arabic, might hide a much smarter, and wiser person, known as "‫"داهية‬. Almost the five MTs failed to understand the
two words and completely distorted the meaning. According to Attia (2004:169), the best cultural equivalent for this
proverb in English is "still waters run deep". Any reader would be completely confused if ever faced with the
translations obtained by Google, Reverso, Yandex, or Systran and the meaning will be either ambiguous or completely
understandable. Most of the MTs chose to treat the two words as proper names and provided a transcription of them.
This sheds light on the fact that the database of these MTs could not recognize these two words, even though of the fact
that both are words used in Standard Arabic and are found in Arabic dictionaries.
Surprisingly, however, Bing provided the best, most accurate, cultural, and communicative equivalent in English
Language even when back-translated. The other four should be fed more with the meaning of the most common
proverbs and their possible cultural equivalents, or at least the meaning and synonyms of all words of any language in
both their singular and plural form.
Based on the explanation provided by Attia (2004), it is said to indicate that a person should not wait for others to do
the work for him; instead, he should do his things himself. Attia considered the English proverb (Paddle your own
canoe) as the best English equivalent for it, considering that it implies the same sense. However, no MT was able to
correctly render the meaning, at least literally. The proverb's words, sense, intended meaning, grammatical structure,
and form were completely distorted and lost. A reader will not be able to even understand the literal idea as it was
grammatically and semantically messed up.
Distorting the whole meaning and structure of a proverb might be one of the worst, most negatively affecting errors of
MT as it leaves the reader entirely baffled and distorts the meaning of the whole context in some cases. Could someone
imagine the confusion of putting one of the translations above in the middle of a dialogue or a context of any kind?
 V. CONCLUSION

The study explores the effectiveness of MTs in translating English-Arabic proverbs and vice versa. In light of the
results, it is found that the most utilized translation method MT resorted to when faced with a proverb is the literal,
followed by the semantic, then finally the communicative method. Among the MTs being investigated, Bing is the most
effective MT in translating proverbs with the highest percentage of communicative equivalents and the lowest
percentage of literal equivalents. Both Bing and Google Translate are in the same rank providing semantic equivalents
among all the others. However, Google is also equal to Reverso with the same percentage of providing communicative
equivalents. The least effective MT among the five was Yandex in providing communicative equivalents and it is also
the highest in providing literal equivalents. If the overall results are taken into account, Bing and Google Translate
produced semantic translations of proverbs while Yandex, Systran and Reverso were more literal. Results indicate
statistically insignificant differences in MT effective translation performance in favour of Bing Translate. Having
analyzed the outputs of each and all the five MTs used in this study, some errors that MT made are detected. These are
(a) missing the implied meaning, (b) weakly structured translations, (c) choosing the wrong synonyms, and (d)
distorting the meaning of the proverb. The author recommended more focus on the pre-translation stage done by online
MT developers and careful post-revision carried out by translators and translation trainers and students. Further studies
are needed to investigate MTs with different text genres so that results acquire strong evidence for being generalized.

REFERENCES
[1] Abdulhaq, S. (2016). Machine Translation: Limits of Accuracy and Fidelity. An MA thesis). An-Najah National
University, Nablus, Palestine.
[2] Acikgoz, F., & Sert, O. (2006). Interlingual Machine Translation: Prospects and Setbacks. Online
Submission, 10(3), 1-16.
[3] Alabbasi, A. (2010) . Introduction to Translation: A Theoretical and Practical Book. Sana’a: Al-Ameen
Publishing and Distribution.
[4] Ali, M. (2020). Quality and Machine Translation: An Evaluation of Online Machine Translation of English into
Arabic Texts. Open Journal of Modern Linguistics, 10, 524-548. DOI: 10.4236/ojml.2020.105030.
[5] Al-Kabi, M., Gigieh, A., Alsmadi, I., Wahsheh, H., & Haidar, M. (2013). An opinion analysis tool for colloquial
and standard Arabic. In The Fourth International Conference on Information and Communication Systems (ICICS
2013) (pp. 23-25).
[6] Al-khresheh, M. H., & Almaaytah, S. A. (2018). English proverbs into Arabic through machine
translation. International Journal of Applied Linguistics and English Literature, 7(5), 158-166.
[7] Almahasees, Z., & Mahmoud, S. (2022). Evaluation of Google Image Translate in Rendering Arabic Signage into
English. World Journal of English Language, 12(1), 185-197.
[8] Alshammari, J. N. (2015). Examining Nida's translation theory in tendering Arabic proverbs into English: A
comparative analysis study. International Journal of English Language and Linguistics Research, 3(8), 45-57.
[9] Anderson, D. D. (1995). Machine Translation as a Tool in Second Language Learning. CALICO Journal, 13, 68-
96.
[10] As-Safi, A. B.(2002) Translation Theories, Strategies And Basic Theoretical Issues. Petra
University. Retrieved From: http://www.uop.edu.jo/download/research/members/424_2061_A.B.pdf.
[11] Attia, M. (2004). The Dictionary of Common English Proverbs Translated and Explained.
[12] Baker, M. (1992). In Other Words, London & New York: Routledge.
[13] Balkan, L. (1992) Translation Tools. Meta, 27 (30), 408-20
[14] Barbour, F. M. (1963). Some uncommon sources of proverbs. Midwest Folklore, 13(2), 97-100.
[15] Belam, J. (2003). Buying up to falling down. In Workshop on Teaching Translation Technologies and Tools.
[16] Birla, V. K., Ahmed, M. N., & Shukla, V. N. (2009). Multiword expression extraction—text
processing. Proceedings of ASCNT-2009, CDAC, Noida, India, 72-77.
[17] Champollion, Y. (2001). Machine translation (MT), and the future of the translation industry. Translation
journal, 5(1). Available on:
https://scholar.google.com/scholar?hl=ar&as_sdt=0%2C5&q=Machine+translation+%28MT%29%2C+and+the+f
uture+of+the+translation+industry&btnG=#d=gs_cit&t=1660262011911&u=%2Fscholar%3Fq%3Dinfo%3At3x6
wX5qm4UJ%3Ascholar.google.com%2F%26output%3Dcite%26scirp%3D0%26hl%3Dar
[18] Costa, Â., Ling, W., Luís, T., Correia, R., & Coheur, L. (2015). A linguistically motivated taxonomy for Machine
Translation error analysis. Machine Translation, 29(2), 127-161. https://doi.org/10.1007/s10590-015-9169-0
[19] Ghazala, H. (1995). Translation as Problems and Solutions: A Coursebook for University Students and Trainee
Translators. Beirut: Dar wa Maktabat Al-Hilal.
[20] Ghazala, H. (1995). Translation as Problems and Solutions: A course-book for university
students and trainee translators(7th ed). Beirut: Dar wa Maktabat AL-Hilal.
[21] Hamdi. S., Nakae. K., & Okashs, M. (2013). Online Translation of Proverbs between Availability and Accuracy-
ISLLLE, Japan,
[22] Ismajli, V. & Maliqi, F. (2021). The effectiveness and efficiency of human translation versus machine translation
Valentina in 1st Alumni Research Conference 2021 (pp. 307-327). Kolegi AAB, Department of English.
[23] Jabak, O. (2019). Assessment of Arabic-English translation produced by Google translate. International Journal
of Linguistics, Literature and Translation (IJLLT) ISSN, 2617-0299.
[24] Latief, M. R. A., Saleh, N. J., & Pammu, A. (2020). The effectiveness of machine translation to improve the
system of translating language on cultural context. In IOP Conference Series: Earth and Environmental
Science (Vol. 575, No. 1, p. 012178). IOP Publishing.
[25] Lembersky, G., Ordan, N., & Wintner, S. (2012). Language models for machine translation: Original vs.
translated texts. Computational Linguistics, 38(4), 799-825.
[26] León Bergasa, A., & Lorés Sanz, R. (2019). A contrastive study of errors in automatic translation and human
translation in tourist texts: an evaluation of Google Translator, Systran and Bing. Available at
https://zaguan.unizar.es/record/85369?ln=fr
[27] Li, H., Graesser, A. C., & Cai, Z. (2014). Comparison of Google translation with human translation. In The
Twenty-Seventh International Flairs Conference.
[28] Lin, G. H. C., & Chien, P. S. C. (2009). Machine Translation for Academic Purposes. Proceedings of the
International Conference on TESOL and Translation 2009 December 2009, pp.133-148
[29] Mieder, W. (1994). Proverbs are never out of season: Popular wisdom in the modern age. Oxford University
Press, USA.
[30] Newmark, P. (1988). A Textbook of Translation. Prentice Hall: China
[31] Newmark, P. (1988). Pragmatic translation and literalism. TTR: traduction, terminologie, rédaction, 1(2), 133-
145.
[32] Nida, E. A., & Taber, C. R. (1982). The theory and practice of translation (Vol. 8). Brill Archive.

[33] Oxford Reference, Science and Technology, from:

https://www.oxfordreference.com/page/scienceandtech/science-and-technology on 15/12/2022.
[34] Polat, Y., Zakirov, A., Bajak, S., Mamatzhanova, Z., & Bishkek, K. (2018). Machine Translation for Kyrgyz
Proverbs—Google Translate Vs. Yandex Translate-From Kyrgyz into English And Turkish. In Сборник
содержит материалы Шестой Международной конференции по компьютерной обработке тюркских
языков «TurkLang-2018»(Ташкент, Узбекистан, 18–20 октября 2018 г.) Данная публикация предназначена
для научных работников, преподавателей, аспирантов и студентов, специализирующихся в области.
[35] Sakre, M. M. (2019). Machine translation status and its effect on business. Journal of the ACS, 10.
[36] Sharma, M., & Goyal, V. (2011). Extracting proverbs in machine translation from Hindi to Punjabi using
relational data approach. International Journal of Computer Science and Communication, 2(2), 611-613.
[37] Sofer, M. (2006). The translator's handbook. Schreiber Pub. Systran: Past and Present.
[38] Thriveni, C. (2002). Cultural elements in translation: The Indian perspective. Available online:
https://translationjournal.net/journal/19culture.htm retrieved on 10/7/2022
[39] Vauquois, B. (1968). A survey of formal grammars and algorithms for recognition and transformation in
mechanical translation. In J.H. Morrell (Ed.), Proceedings of the International Federation for Information
Processing Congress (IFIP-68), (Vol. 2, pp. 1114–1122).
[40] Venuti, L. (Ed.). (1998). Strategies of Translation. In Baker, M. (ed.) The Routledge Encyclopedia of Translation
Studies. London: Routledge, 240-244.
[41] Vilar, D., Xu, J., d’Haro, L. F., & Ney, H. (2006). Error analysis of statistical machine translation output.
In Proceedings of the fifth international conference on language resources and evaluation (LREC’06).
[42] Vinay, J-P., & Darbelnet, J. (1958). Comparative Stylistics of French and English: a Methodology for
Translation, translated by J. C. Sager and M. J. Hamel, Amsterdam/ Philadelphia: John Benjamins.
 ABOUT THE AUTHOR:
Ibrahim Jibreel is an assistant professor in Translation Studies. He is currently the Head of
English & Translation Department at the University of Science & Technology (UST),
Hodeidah, Yemen. In addition, he is the chairperson of the Scientific Research Committee in
the US, Hodeidah and the coordinator of postgraduate studies programs. He obtained his MA in
Applied Linguistics in 2011; and his Ph.D. in Translation Studies in 2017. Furthermore, he was
awarded membership of the International Association of Scientific Researcher on Nov. 2019,
the accredited professional translator (English – Arabic) from the Arab Professional Translators
Society in September 2021, membership of the International Union of Languages and
Translation on July 2021 as well as an active member in the International Association for
Translation & Intercultural Studies (IATIS). He is interested in comparative translation studies, translation and
technology, translation quality assessment, and linguistics; and has published several articles in international journals.