*.

SYNTACTIC PARSING

1 Introduction Syntactic parsing

is a fundamental issue in natural language processing and hasa wide range of applications. This issue has
been the subject of intensive research for decades, and as a result, there are parsers in specific fields.
The vastamount of information in the natural language available for access by computershas introduced
the need for the development of computer systems for processing the natural language [6]. Initially,
extracting and interpreting the availablecontent was difficult, but the progressive growth of natural
language processingintroduced easy and systematic solutions. In recent decades, natural
languageprocessing has emerged as an active research field, providing efficient applicationssuch as
machine translation, information retrieval, information extraction, textsummarization, speech
recognition, parsing, etc. Machine learning algorithmsare used to solve these tasks. The way we share or
communicate our feelings isof great importance in processing the text in the interest of analysis [31].

Parsing is the most appropriate approach to interpreting sentences of thenatural language, to

identifying and searching what possible expressions express.It is the process in which the syntactic
structure of the sentence is identifiedusing the lexical and syntactic rules. Identification of syntactic
structure is usefulin determining the meaning of a sentence [31, 17]. Parsing generates a parsetree of
the sentence to eliminate the interpretation ambiguity [31]. Albanianis a language that has a free word
order, like German, which means that thesame forms of the word appearing in different positions in the
sentence, oftenhave different grammatical and/or semantic roles, so identification and
summarysentences in Albanian is a challenging task.

2. Related WorkParsing is a research field of natural language processing, an imperative basis forany
advanced language processing and a key step in understanding a sentence. Itis the process of analyzing a
sentence to discover its phrase structure, accordingto the rules of a grammar.Natural Language
Processing is a field of computer science, artificial intelligence, and linguistics devoted to making
computers understand the expressionsand words written in natural languages. Natural Language
Processing (NLP)encompasses everything a computer, or machine needs to understand the natural
written or spoken language. The main goal of research in NLP is to analyzeand understand the language
[6].

Syntactic parsing is a fundamental field of research in computational linguistics. In the context of NLP,
the term parsing refers to the automated analysis ofa sentence as a sequence of words, to determine its
potential syntactic structurefrom a formal grammar definition [7, 13, 14]. The formal grammar is
generativeand has answered many theoretical issues related to the linguistic structures [9,12]. There are
two forms of parsing. A top-down parser starts with the inputsentence and tries to build a tree whose
leaves match the given input. A bottomup parser starts with input words and tries to build the trees by
finding severalderivations that give the input sentence, applying the grammatical rules [15, 16,

3 Albanian Syntactic ParserA parser is a tool that is responsible for generating the parse tree. It is a
procedural component, which remains the same during the parse tree generationregardless of
language, but the grammar doesn’t remain the same for all languages [31].In syntactic parsing, the
parser seen as search through the space ofall possibles parses to find the correct one for the sentence.
The grammar definesthe search space of the possible parse trees. Most of the parses rely on two
searchstrategies: top-down search that is driven by grammar and bottom-up that isdriven by the data
[15, 16, 22]. Perhaps, the ambiguity is the most serious problem faced by syntactic parsers. At some
point in one pass through a sentence,there will usually be some grammatical rules that can be applied
[17].The model of the syntactic parser consists of various levels of modules. Thefirst level is the lexical
analysis. The purpose of this level is to split the inputinto the sequence of tokens corresponding to the
words. A further step in theanalysis is to map each word with a part of speech tag. The second level, is
thesyntactic parsing, which analyzes the syntactic structure of the sentence. Theanalysis confirms that
phrases are well-formed, and it determines a linguisticstructure represented as a parse tree. The
language analyzer uses knowledgeof the language syntax (grammar), morphology (lexicon), and
identifying thelinguistic relationships provides a structure for semantic interpretation.

3.1 Albanian Part of Speech TaggerThe part of speech tagger used in this work is a trained model based
on handtagged data. The tagging is done using a corpus of words labeled manually withtags. Texts
tagged in the corpus include different word-forms of the same word,

because of the different grammatical characteristics that take a word within thesentence. The corpus is
a 46,306-word collection of samples from written textsof different genres (novels, newspapers, etc.).

3.2 Syntax and Formal Grammar for AlbanianStudying the structure of the sentence is referred to as
syntax. The word syntaxcomes from the Greek word s‘yntaxis, which means “putting together or
arranging”, and refers to how the words are arranged together in a sentence, in thesense of the
grammatical nature with the relationships between the units formedby them, with the related
regularities [20].Parsing requires a mathematical model of the syntax of the respective language, which
is supposed to be a formal grammar. The most common mathematical system for modeling the phrasal
structure in natural languages is ContextFree Grammar (CFG) [16]. Since its introduction by Noam A.
Chomsky (1956),CFG has been the most influential formalism of the grammar to describe thesyntax of
the language and is often used as a basic formalism when describingthe parsing algorithms [13]. Its
purpose is not to describe the details of a particular language, but to formulate the basic principles that
define the naturallanguages, grammar, and their common characteristics [24, 9, 12].

*. Prosody
 INTRODUCTION. Prosody, in
the sense I will be using throughout this review, refers to a structure defined by phrasing and
prominence. Prosody groups syllables and words together and renders some elements more
prominent than others. While some of the phrasing is realized by intonation and rhythm
modulations, others are revealed through phenomena that affect the segments of the utterance.
Prosody plays an important role in the establishment of the phonological, lexical, syntactic, and
semantic representation of the sentence, as well as in retrieving some aspects of the information
structure of the sentence (e.g., what is presupposed or what is not) and the pragmatic
implications that a listeners may infer regarding the speaker’s intentions. For example, people share
the intuition that the segmentally identical phrases “crisis turnip” and “cry sister nip” can be
distinguished from each other based upon the prosodic rendering that the speaker adopts.1 Likewise,
the sentences “Raoul murdered the man with a gun” and “I asked the teacher who left” have more
than one literal meaning and speakers may be able to reveal which they intend through the
sentence’s prosody.2,3,4 The prosody of a sentence can sometimes affect a listener’s conclusion
about the speaker’s beliefs, attitude, or intention.5,6 For example, the sentence “It looks like a
zebra” with an accent on the final word is characterized by a canonical declarative prosodic contour
and is perceived as supporting an affirmative interpretation (“It looks like a ZEBRA. [and indeed, it
is one]”); on the other hand, the same sentence with a so-called ‘contrastive’ accent on “looks” and a
rising intonation supports a How to cite this article: WIREs Cogn Sci 2015.

2 negative or contradictory interpretation (“It LOOKS like a zebra [but it actually isn’t one]”). Thus,
the prosody of a sentence is intimately connected to the sentence’s multiple linguistic structures.
This review provides a summary of some recent advances made in characterizing how prosody
contributes to language comprehension. It is limited to studies conducted on the English language
because these continue to make up the bulk of the research on the topic. In pursuing such
investigation, psycholinguists generate hypotheses about the nature of the information listeners
extract from the utterance and how the information is utilized in the establishment of linguistic
structures. This review is as much a review of the most exciting findings on this topic as it is
about the nature of the hypotheses scholars have considered and the assumptions they entertain
along the way. PROSODY AS AN ABSTRACT STRUCTURE The study of prosody has a long history. Until
the mid 1970s, little consensus had been reached on what determines a speaker’s choice of
intonation and rhythm. Since then, a widely accepted framework has emerged. According to the
‘autosegmental-metrical’ theory, the variations of pitch, loudness, and rhythm of an utterance are
the phonetic realization of an abstract structure composed of phonological objects (see Ref. [7-9]
for reviews). This prosodic structure consists of a succession of intonational elements or targets
that are associated with particular landmarks in the segmental description of the utterance. How
these targets are linked with elements in the segmental string is determined by a ‘metrical grid’
(i.e., an abstract representation that specifies the organization of syllables into constituents of
increasing size, with the smallest constituents nested into constituents of larger sizes) and the level
of prominence that each syllable can receive. The intonation of an utterance, or its tune, consists
of two kinds of discrete elements. Pitch accents are tonal objects that are associated with specific
syllables and that mark them as prominent. Pitch accents are viewed as properties of words.
Edge tones, on the other hand, are pitch movements that mark the right edge of constituents.
The domains of edge tones are the constituents they mark. Both kinds of tonal elements are
further specified by their level, ‘high’ (H) or ‘low’ (L), as opposed to configurations, such as ‘rise’ or
‘fall’. Because the F0 values that these elements can take are variable (a necessary consequence
of assuming two levels), tonal elements are defined as local F0 ‘peaks’ or ‘valleys’ whose specific
phonetic values depend on factors considered orthogonal to their phonological category, such as
the degree of emphasis or the position in the utterance. Finally, the discrete phonetic elements are
assumed to be connected with one another via interpolation or other simple principles. Thus,
within the auto-segmental-metrical framework, a complete theory must specify the inventory of
tonal elements and a series of principles or rules regarding how a sequence of these abstract,
discrete elements translates into a continuous F0 contour. Unfortunately, the field has not reached a
consensus regarding either of these aspects. There is disagreement on how to decide whether a
portion of an utterance’s F0 contour is associated with a tone or is merely a transition, how to
decide whether a pitch accent contains a single target or two (i.e., a bi-tonal pitch accent), and whether
the pitch range of the F0 contour plays a role in this decision. For instance, there is much debate
on whether the F0 range that characterizes the onset of a rise in a pitch accent (i.e., medium or low
in the speaker’s pitch range) can be taken to specify a phonological distinction between two
accent types, L+H* vs. H*, or whether the pitch range marks gradient, non-phonological
variations.10,11 The classification of tonal element as H or L is also contentious. Although H pitch
accents have been loosely described as local F0 peaks associated with accented syllables, and L
pitch accents as local F0 valleys, there is no consensus on defining this locality.12,13 Finally, it
still is an open question whether all intonational categories are expressed through discrete
phonological elements or if some may be conveyed by the continuous variation of a physical
dimension (such as the alignment of a pitch accent with the accented syllable’s vowel).14,15

3 An assumption shared by all phonological approaches to intonation and prosody is that

discrete tonal elements convey meaning, an assumption that Ladd8 dubbed ‘the linguist’s theory of
intonational meaning’: Prosodic elements, alone or combined with other elements to form a tune and
in conjunction with the other linguistic structures of the sentence and a rich pragmatic system, are
assumed to give rise to meaning as we experience it.16-20 However, the specific meanings of tones,
as well as tunes, are debated. It is useful to keep in mind that tonal elements may resemble
segments in this respect: While phonemes are elements of meaning because phonemic contrasts
capture (potential) contrasts in meaning, they themselves have no independent meaning (the /p/ in
‘peach’, in contrast to the /b/ in ‘beach’, convey nothing that differentiates the semantic features of
a peach compared to those of a beach). Because of the assumption that tonal elements are discrete
categories that convey discrete meaning, an approach to establishing the inventory of such
categories has consisted of showing that the interpretation of a sentence changes as a function of
its prosodic characteristics. However, a contrast in meaning (such as an intonation that conveys
uncertainty or incredulity) is not synonymous with a phonological distinction because such
distinction can emerge from contrasting two values located on a continuous dimension.10,21
Phonologists often find it more productive to focus on the phonetic forms that an utterance can
take and tease apart phonetic from phonological variations.10,12,13,22-24 Much debate remains
despite active research on these questions. Nonetheless, many scholars have adopted the
framework offered by Pierrehumbert25 and its later instantiation,26 or the transcription system
known as ‘ToBI’ (an acronym for ‘Tones and Boundary Indices’),27 which substantially departs
from Pierrehumbert’s original theory. To those who are not well versed into the phonological
literature, the popularity of these theories may be interpreted as evidence for a consensus while in
fact, as described above, many of their assumptions have not received firm empirical support or
have been challenged. This brief summary of the state of research on prosody and intonational
phonology highlights the difficulties in describing how the various linguistic structures of a sentence
affect the prosodic structure that speakers assign to their utterance, and how listeners interpret
the continuously varying dimensions of speech into elements that, once integrated with other
linguistic elements of the sentence, give rise to an interpretation. The prosodic structure of a
sentence is a representation that is independent of, albeit related to, the sentence’s other
linguistic structures. It is via these relationships that prosody contributes to the meaning of the
sentence. However, the role of prosody has been conceptualized, and consequently investigated,
differently for different linguistic structures. These disparities, I argue, reflect the division of labor
that scholars have (largely implicitly) assumed between prosody and other markers of a sentence’s
linguistic structures. For the syntax of a sentence, the establishment of the structure is viewed as
depending on the linguistic elements present in the sentence and the grammar, which dictates
both how these elements combine into phrases and how these phrases relate to one another; when
the sentence is spoken, the prosodic phrasing may help determine which of those relationships is
intended. Thus, prosody is conceptualized as complementing syntactic knowledge. By contrast, the
information structure of a sentence, i.e., the way by which the sentence relates to the context in terms
of the discourse status of its content and the attentional and intentional states of the discourse
participants, relies heavily on the pattern of prominence (and perhaps phrasing) that the speaker
chooses to produce. Other devices are available, such as pronouns or syntactic constructions, but the
role of prosody in marking the information structure in English is substantial. Here, the contribution
of prosody to the sentence’s information structure is viewed as leading, as opposed to supporting,
and prosodic elements are often conceived as conveying meaning that bears on information
structure directly. As far as the phonology of a sentence is concerned, the situation is a little
more complex because the structure encompasses many aspects; regarding the establishment of
the segmental content of a

4 sentence, one from which morpho-syntactic elements emerge, prosody is generally viewed as
having a supportive role.28 These often implicit assumptions have had implications on the nature
of the questions that scholars have addressed: When examining the role of prosody in establishing
the phonology and syntax of a sentence, investigators have asked whether or not prosody
contributes to it, and if it does, how to characterize the mechanism by which the contribution
takes place. Regarding the role of prosody in establishing the information structure of a sentence,
on the other hand, scholars have taken for granted that prosody plays a role; instead, research has
revolved around the way prosody conveys the dimensions of information structure and whether
there is a one-to-one mapping between elements of the two structures. Here, I discuss the current
state of knowledge on the role of prosody in establishing each of those structures. THE ROLE OF
PROSODY IN DERIVING THE PHONOLOGICAL REPRESENTATION OF AN UTTERANCE Speech
comprehension requires listeners to make some hypotheses about the linguistic elements present in
the utterance they are trying to interpret. The significant role that prosody plays in this process is
becoming increasingly clear. It has long been recognized that many temporal cues to phonemic
contrasts (such as the length of the delay between the burst and the onset of vocalic vibrations, an
acoustic correlate to the voicing contrast amongst stop consonants) are interpreted relative to global
temporal aspects, such as the syllable duration or speaking rate.29 The phenomenon could be viewed
as confined to a hypothetical early stage of processing, one in which the acoustic signal is
translated into a linear string of phonemic elements. Such phonemic representation, one could
argue, constitutes the sole basis for establishing the lexical and morpho-syntactic composition of
the utterance. This view is still commonly held, despite substantial evidence for its shortcomings. The
assumption of a phonemic string devoid of any phonetic marking of syllable, word, and phrase
boundaries denies a role for these cues in assisting with segmenting and parsing the segmental
string into linguistic units. The absence of systematic cues to word boundaries, as pointed out by
many empirical papers published in the 1970s and 1980s, may have contributed to the prevalence
of this view.30-32 However, a large body of work has since documented the multiple ways in
which the prosodic structure of an utterance is marked in the phonetic characteristics of the
segments themselves (see Ref. [33] for a review). If prosody plays such a fundamental role in
explaining fine-grained systematic variation in the production of individual speech sounds, one may
expect listeners to take advantage of the marking of prosodic constituents in the process of
isolating linguistic constituents. Indeed, while prosodic and lexical constituents are not isomorphic,
their relationship is quite systematic: The edges of prosodic words (and of the prosodic phrases
higher on the prosodic hierarchy) are always aligned with word boundaries. For instance, if an
edge tone is aligned with the syllable “ham”, this syllable must be word final (and, perhaps,
correspond to the monosyllabic “ham”) rather than be parsed as the first syllable of a polysyllabic
word such as “hamster”. Thus, prosodic phrasing could facilitate the segmentation of an utterance
into word-size units by favoring a parsing of the input that respects the prosodic-lexical alignment
and disfavors one that violates it. This hypothesis has now received ample empirical support.34-37
In addition to local cues, segmentation may be assisted by some stress-based patterning or
rhythm, what Dilley and colleagues call ‘distal’ prosody.38-40 People appear to build an expectation
on how syllables-to-come will be grouped based on the pattern of tonal alternation that precedes
them and this expectation affects how they perceive the current grouping, above and beyond the
local acoustic-phonetic cues that may cue grouping. A number of accounts of the mechanism by which
the prosodic structure of an utterance affects segmentation have been proposed. According to
one approach, the acoustic-phonetic characteristics that arise from the

5 prosodic structure at word junctures, such as aspiration, glottalization, lengthening, etc., are
integrated with other cues, such as phonotactic, lexical, and syntactic knowledge, in order to
generate the segmentation that best fit the set of cues.28,41 For instance, the lengthening of a
syllable, given the context, is interpreted as a cue to a word boundary. This approach posits that
the phonetics of an utterance directly mark word boundaries. An alternative account proposes that
phonetic cues to the prosodic structure affect the perceived lexical composition of the utterance
indirectly, via the establishment of a prosodic structure. Such structure is built in parallel to the
phonemic and lexical analysis of the utterance.34,36 In this approach, specific phonetic
characteristics, such as syllable lengthening or articulatory strengthening, cause listeners to posit
elements of prosodic structure. Such prosodic structure would then serve as a source of support
for some parsing of the segmental string over others. For instance, the posited presence of an
intonational-phrase boundary would favor lexical hypotheses whose edges are aligned with the
boundary and disfavor those that straddle it. The proposal of a prosodic structure being built in
parallel to a segmental representation of the utterance implies some independence in the
construction of these two representations, in spite of the fact that the utterance’s phonetics
merges the two. But recent research has revealed some interdependence between the two. In their
seminal study, Dilley and Pitt42 showed that whether or not people perceived the presence of a
short function word in a phrase depended on the perceived duration of the fragment of speech
comprising the syllable that preceded the potential function-word site and the function word
itself, if produced. If the fragment was perceived as too long to be a single syllable given the speech
rate, a function word was posited. Conversely, if the fragment was perceived as too short to also
include a function word, its presence was largely undetected. Duration perception could have been
induced from an estimate of long-range speaking rate, but the effect was also sensitive to a very
local perturbation.43 In the light of this work, one must conclude that the segmental representation of
an utterance is only partially determined by the spectral cues of those segments. Equally
important are the temporal characteristics of the utterance to which listeners try to attribute lexical
content. It is not clear how this finding can emerge from a model in which the prosodic and
phonemic structures of an utterance are independently assessed. Bayesian accounts of speech
perception, however, are in principle compatible with this result and, more generally, with an
approach in which the phonological structure of an utterance depends on the establishment of an
abstract prosodic structure. In the Bayesian framework, the goal of the listener can be
conceptualized as explaining away the data (the phonetics of the utterance) by selecting what is
most likely to have generated them. Listeners evaluate how well a given phonological (both
segmental and prosodic) representation accounts for the phonetics of the utterance based on the
likelihood that this representation would yield the utterance. The key aspect is that the analysis of
the utterance in terms of its segmental and prosodic composition is done simultaneously and in
consideration of the influence that one imposes on the other. In this approach, the distinction
between bottom-up and top-down processing is blurred because some form of loop takes place
between the data and linguistic knowledge: The data give rise to a set of hypotheses which, in
turn and based on a generative mechanism, can be used to evaluate how well they predict the
data. Furthermore, expectations against which the data are evaluated can be adjusted to
incorporate contextual information, a powerful mechanism to account for the fact that the same
phonetic cue can give rise to different percepts across contexts.44,45 The notion of explaining
away the data based on a hypothesized prosodic structure has also been proposed to account for
the role of prosody in establishing the syntactic structure of a sentence, a topic the next section is
devoted to.