Foundations of Language Science and Technology (FLST) Lecture 4 - - PowerPoint PPT Presentation

▶
foundations of language science and technology flst
SMART_READER_LITE
LIVE PREVIEW

Foundations of Language Science and Technology (FLST) Lecture 4 - - PowerPoint PPT Presentation

Dec 11, 2022 140 likes •497 views

Foundations of Language Science and Technology (FLST) Lecture 4 (28.10.2009): Syntax PD Dr.Valia Kordoni Email: kordoni@coli.uni-sb.de http://www.coli.uni-saarland.de/courses/FLST/2009/ Units of Language Subfields of Linguistics Grammar

slide-1
SLIDE 1

Foundations of Language Science and Technology (FLST)

Lecture 4 (28.10.2009): Syntax

PD Dr.Valia Kordoni

Email: kordoni@coli.uni-sb.de http://www.coli.uni-saarland.de/courses/FLST/2009/

slide-2
SLIDE 2

FLST 09-10 – Lecture 4 (28.10.09)

2

Units of Language – Subfields of Linguistics

Text & Discourse Text & Discourse Grammar Discourse Semantics Pragmatics Sentence Syntax Compositional Semantics Pragmatics Word Morphology Lexical Semantics

  • Sound

Phonetics/ Phonology

  • Use

Pragmatics Meaning Semantics Structure Grammar

slide-3
SLIDE 3

FLST 09-10 – Lecture 4 (28.10.09)

3

Morphology and Syntax

  • Morphology investigates the structure of words
  • Syntax investigates the structure of sentences.
  • In a way, syntax is the morphology of sentence, or,

taken the other way round, morphology is the syntax of words.

  • But: Sentence structure differs from word structure,

in various respects.

slide-4
SLIDE 4

FLST 09-10 – Lecture 4 (28.10.09)

4

Observation 1: Constituents

  • A simple morphological rule of German:
  • The comparative morpheme occupies the first position of

the ending (= the second position of the word)

  • schnell+er+es [ fast+er, n, sg]
  • A simple syntactic rule of English:
  • The finite verb occupies the second position of a

declarative sentence

  • John + gave + Mary + a + book
slide-5
SLIDE 5

FLST 09-10 – Lecture 4 (28.10.09)

5

Constituents [1]

  • Counter-examples (1)
  • Yesterday John gave Mary a book.
  • But John gave Mary a book.
  • Counter-examples (2)
  • The student gave Mary a book.
  • The friendly student gave Mary a book.
  • The friendly student which I told you about yesterday

gave Mary a book.

slide-6
SLIDE 6

FLST 09-10 – Lecture 4 (28.10.09)

6

Constituents [2]

  • Counter-examples (1)
  • Yesterday John gave Mary a book.
  • But John gave Mary a book.
  • Counter-examples (2)?
  • The student gave Mary a book.
  • The friendly student gave Mary a book.
  • The friendly student which I told you about yesterday

gave Mary a book.

  • The verb is still in second place, if we count constituents

rather than words.

slide-7
SLIDE 7

FLST 09-10 – Lecture 4 (28.10.09)

7

Arbitrarily long and complex sentences [1]

  • The mouse escaped into the garden.
  • The mouse that the cat chased escaped into the

garden.

  • The mouse that the cat which Mary owns chased

escaped into the garden.

slide-8
SLIDE 8

FLST 09-10 – Lecture 4 (28.10.09)

8

Arbitrarily long and complex sentences [2]

  • Er hat die Übungen gemacht.
  • Der Student hat die Übungen gemacht.
  • Der interessierte Student hat die Übungen gemacht.
  • Der an computerlinguistischen Fragestellungen interessierte Student hat die

Übungen gemacht.

  • Der an computerlinguistischen Fragestellungen interessierte Student im ersten

Semester hat die Übungen gemacht.

  • Der an computerlinguistischen Fragestellungen interessierte Student im ersten

Semester, der im Hauptfach Informatik studiert, hat die Übungen gemacht.

  • Der an computerlinguistischen Fragestellungen interessierte Student im ersten

Semester, der im Hauptfach, für das er sich nach langer Überlegung entschieden hat, Informatik studiert, hat die Übungen gemacht.

slide-9
SLIDE 9

FLST 09-10 – Lecture 4 (28.10.09)

9

Structural ambiguity

  • Morphology talks about sequences of morphemes.
  • To talk about syntactic regularities requires

reference to constituent structure.

  • Semantic interpretation of sentences also requires

information about constituent structure:

  • Pick up a big red block.
  • in particular, if sentences are structurally

ambiguous:

  • John saw the man with the telescope.
slide-10
SLIDE 10

FLST 09-10 – Lecture 4 (28.10.09)

10

Syntactic ambiguity

  • John saw the man with the telescope
  • John saw the man with the telescope
  • Young students and professors attended the party.
  • Young students and professors attended the party.
slide-11
SLIDE 11

FLST 09-10 – Lecture 4 (28.10.09)

11

Tests for constituency

Substitution test: Word sequences that can be systematically substituted for a single word (e.g., proper name or personal pronoun) form a constituent:

  • The student gave Mary a book.
  • The friendly student gave Mary a book.
  • The friendly student which I told you about yesterday gave Mary a book.
  • Mary gave John a book.
  • Mary gave the student a book.
  • Mary gave the friendly student which I told you about yesterday a book.

Compare with:

  • Yesterday John gave Mary a book.
  • Mary gave yesterday John a book.
slide-12
SLIDE 12

FLST 09-10 – Lecture 4 (28.10.09)

12

Syntactic Categories

  • Constituents that are substitutable for each other can be

subdivided into larger classes that share distribution and structural properties, the Syntactic Categories, e.g.:

  • Noun phrases, consisting of a pronoun, a proper name, or

a complex structure with a common noun as syntactic head element – NP

  • Prepositional phrases (with the telescope, into the

garden) – PP

  • Adjective phrases (friendly, very friendly, interested in

linguistics) - AP

slide-13
SLIDE 13

FLST 09-10 – Lecture 4 (28.10.09)

13

Categories and Functions

  • Syntactic categories denote classes of constituents

with similar internal structure, in particular, the category /part-of-speech of their lexical head.

  • Grammatical functions characterise the external

role of a constituent in its syntactic context, e.g.

  • Complements: Subject, (Direct, indirect, prepositional)

Object

  • Modifier / Adjunct
slide-14
SLIDE 14

FLST 09-10 – Lecture 4 (28.10.09)

14

Syntactic Description with CFGs

  • CFG is a formalism that allows to model the concept for

grammaticality for natural languages, by specifying the set

  • f grammatically correct sentences, and assigning them their

appropriate grammatical structures (in terms of their parse trees).

  • Is it a realistic and reasonable aim to describe the set of

grammatically correct sentences of a language?

  • What to do with ungrammatical input?
  • What does 'grammatical' mean after all? – Graded grammaticality!
  • Is a CFG the appropriate formalism to describe the grammar
  • f a language?
slide-15
SLIDE 15

FLST 09-10 – Lecture 4 (28.10.09)

15

Syntactic Processing with CFGs [1]

  • Morphological analysers are finite-state automata (or

transducers) working in linear time.

  • The syntax of programming languages is recursive, and

therefore described by CFGs. Because the languages typically are unambiguous, and described by deterministic CFGs, parsers for programming languages are also linear time.

  • Unfortunately, grammars of natural languages are

ambiguous and non-deterministic. The best algorithms (Earley Algorithm, Chart Parsing) take quadratic time to find one parse, and cubic time to find all parses.

slide-16
SLIDE 16

FLST 09-10 – Lecture 4 (28.10.09)

16

Syntactic Processing with CFGs [2]

  • Good news: There are techniques to compile CFGs down to

FSAs for many applications, without loosing much coverage (e.g., by constraining recursion depth; "finite-state technology")

  • Bad news: Constituent structure is only the tip of the

iceberg: More descriptive power is needed to describe syntactic structure of natural languages appropriately. Modern grammar formalisms like LFG or HPSG come in the format of typed feature structures with a context-free backbone.

slide-17
SLIDE 17

FLST 09-10 – Lecture 4 (28.10.09)

17

Variable Word-Order in German

Peter hat der Dozentin das Übungsblatt heute ins Büro gebracht. Peter has the lecturer the exercise-sheet today into-the office brought Das Übungsblatt hat Peter der Dozentin heute ins Büro gebracht. Der Dozentin hat Peter heute das Übungsblatt ins Büro gebracht. Ins Büro hat heute Peter der Dozentin das Übungsblatt gebracht. Heute hat Peter das Übungsblatt der Dozentin ins Büro gebracht. Ins Büro hat das Übungsblatt der Dozentin Peter heute gebracht. * Ins Büro heute Peter das Übungsblatt hat gebracht der Dozentin. * Ins heute Büro der Peter Dozentin das hat Übungsblatt gebracht.

slide-18
SLIDE 18

FLST 09-10 – Lecture 4 (28.10.09)

18

More syntactic phenomena

  • Agreement
  • Subcategorisation
  • Long-distance Dependencies
slide-19
SLIDE 19

FLST 09-10 – Lecture 4 (28.10.09)

19

Computational Grammar Formalisms

Computational Grammar formalisms share several properties:

  • Descriptive adequacy
  • Precise encodings (implementable)
  • Constrained mathematical formalism
  • Monostratalism
  • (Usually) high lexicalism
slide-20
SLIDE 20

FLST 09-10 – Lecture 4 (28.10.09)

20

Descriptive Adequacy

Some researchers try to explain the underlying mechanisms, but we are most concerned with being able to describe linguistic phenomena

  • Provide a structural description for every well-

formed sentence

  • Gives us an accurate encoding of a language
  • Gives us broad-coverage, i.e., can (try to) describe

all of a language No notion of core and periphery phenomena

slide-21
SLIDE 21

FLST 09-10 – Lecture 4 (28.10.09)

21

Precise Encodings

Mathematical Formalism: formal way to generate sets of strings Precisely define:

  • elementary structures
  • ways of combining those structures

=> Such an emphasis on mathematical precision makes these grammar formalisms more easily implementable

slide-22
SLIDE 22

FLST 09-10 – Lecture 4 (28.10.09)

22

Constrained Mathematical Formalism

A formalism must be constrained, i.e., it cannot be allowed to specify all strings

  • Linguistic motivation: limits the scope of the

theory of grammar

  • Computational motivation: allows us to define

efficient processing models

slide-23
SLIDE 23

FLST 09-10 – Lecture 4 (28.10.09)

23

Monostratal Frameworks

Only have one (surface) syntactic level

  • Make no recourse to movement
  • Augment your basic (phrase structure) tree with

information that can describe „movement“ phenomena => Without having to refer to movement, easier to process sentences on a computer

slide-24
SLIDE 24

FLST 09-10 – Lecture 4 (28.10.09)

24

This should be avoided!

Sue gave Paul an old penny

NP V VP NP S NP Aux NP-Q IP S

What did Sue give Paul ___

slide-25
SLIDE 25

FLST 09-10 – Lecture 4 (28.10.09)

25

Lexical

In the past, rules applied to broad classes and only some information was put in the lexicon, e.g., subcategorisation information

  • Linguistic motivation: lexicon is the best way to

specify some generalisations: He told/*divulged me the truth

  • Computational motivation: can derive lexical

information from corpora (large computer- readable texts) => Shift more of the information to the lexicon; each lexical item may be a complex object

slide-26
SLIDE 26

FLST 09-10 – Lecture 4 (28.10.09)

26

Context-Free Grammars (CFGs)

Context-Free Grammars (CFGs) are one kind of constrained mathematical formalism, a precise way of encoding syntactic rules:

  • elementary structures: rules composed of non-

terminal and terminal elements

  • combine rules by rewriting them
slide-27
SLIDE 27

FLST 09-10 – Lecture 4 (28.10.09)

27

Context-Free Rules

Example of a set of rules:

  • S NP VP
  • NP Det N
  • VP V NP
  • ...

But these rules are rather impoverished.

slide-28
SLIDE 28

FLST 09-10 – Lecture 4 (28.10.09)

28

Are CFGs good enough?

  • Data from various languages show that CFGs are

not powerful enough to handle all natural language constructions

  • CFGs are not easily lexicalised
  • CFGs become complicated once we start taking

into account agreement features, verb subcategorisations, unbounded dependency constructions, raising constructions, etc. We need more refined formalisms...

slide-29
SLIDE 29

FLST 09-10 – Lecture 4 (28.10.09)

29

Beyond CFGs

Move beyond CFGs, but stay „mathematical“:

  • Extend the basic model of CFGs with, for

instance, complex categories, functional structure, feature structures, ...

  • Eliminate CFG model (or derive it some other

way)

slide-30
SLIDE 30

FLST 09-10 – Lecture 4 (28.10.09)

30

Computational Grammar Frameworks

  • Dependency Grammar (DG)
  • Tree-Adjoining Grammar (TAG)
  • Combinatory Categorial Grammar (CCG)
  • Lexical Functional Grammar (LFG)
  • Head-Driven Phrase Structure Grammar (HPSG)
slide-31
SLIDE 31

FLST 09-10 – Lecture 4 (28.10.09)

31

Dependency Grammar (DG)

  • The way to analyse a sentence is by looking at the

relations between words

  • A verb and its valents/arguments drive an analysis,

which is closely related to the semantics of a sentence

  • No grouping, or constituency, is used
slide-32
SLIDE 32

FLST 09-10 – Lecture 4 (28.10.09)

32

Tree-Adjoining Grammar (TAG)

  • Elementary structures are trees of arbitrary height
  • Trees are rooted in lexical items, i.e., lexicalised
  • Put trees together by substituting and adjoining

them, resulting in a final tree which looks like a CFG-derived tree

slide-33
SLIDE 33

FLST 09-10 – Lecture 4 (28.10.09)

33

Combinatory Categorial Grammar (CCG)

  • Categorial Grammar derives sentences in a proof-

solving manner, maintaining a close link with a semantic representation

  • Lexical categories specify how to combine words

into sentences

  • CCG has sophisticated mechanisms that deal

nicely with coordination, extraction, and other constructions

slide-34
SLIDE 34

FLST 09-10 – Lecture 4 (28.10.09)

34

Lexical Functional Grammar (LFG)

  • Functional structure (subject, object, etc.) divided

from constituent structure (tree structure)

– kind of like combining dependency structure with phrase structure

  • Can express some generalisations in f-structure;

some in c-structure; i.e., not restricted to saying everything in terms of trees

slide-35
SLIDE 35

FLST 09-10 – Lecture 4 (28.10.09)

35

Head-driven Phrase Structure Grammar (HPSG)

  • Sentences, phrases, and words all uniformly

treated as linguistic signs, i.e., complex objects of features

  • Similar to LFG in its use of feature architecture
  • Uses an inheritance hierarchy to relate different –

types of objects (e.g., nouns and determiners are both types of nominal)