Statistical Parsing October 27, 2016 Dependency grammars Grammar - - PDF document

SLIDE 1

Statistical Parsing

Grammars and grammar formalisms Çağrı Çöltekin

University of Tübingen Seminar für Sprachwissenschaft

October 27, 2016

Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

This course is about …

NP NP JJ statistical NN constituency CC and NN dependency NN parsing PP IN

• f

NP NN natural NN languages nmod amod case nmod conj cc amod

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 1 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Why do we need syntactic parsing?

• Often, syntactic analysis is an intermediate step helping

(semantic) interpretation of sentences hence it is useful for applications like question answering, information extraction

• (Statistical) parsers are also used as language models for

applications like speech recognition and machine translation

• It can be used for grammar checking, and can be a useful tool

for linguistic research

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 2 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Ingredients of a parser

• A grammar
• An algorithm for parsing
• A method for ambiguity resolution

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 3 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Grammars

The term grammar is used for,

• a description of the whole system/structure of a

language—as in a ‘grammar (book) of English’

• a grammar formalism, that are often developed as theory
• f language—as in HPSG, LFG, CCG
• A formal (fjnite) specifjcation of a language as a possibly

infjnite set of strings (not necessarily a natural language)

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 4 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Plan of the lecture

• Constituency grammars
• Dependency grammars
• Brief notes on some major grammar formalisms

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 5 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Constituency grammars

• Constituency grammars are

probably the most studied grammars both in linguistics, and computer science

• The main idea is that a group of

words form natural groups, or ‘constituents’, like no noun phrases

• r word phrases
• phrase structure grammars or

context-free grammars are often used as synonyms S NP John VP V saw NP Marry

Note: many grammar formalisms use constituency grammars in some way, we will not focus on a particular grammar formalism here.

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 6 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

What is a constituency

Linguists ofger a number of tests for constituency, such as

• They can answer questions:

Q: ‘What did John do? →A: ‘saw Marry’

but, presumably, no question with answer ‘John saw’

• Substitution with a pronoun forms:

Q: ‘John [read the book] last week? →A: ‘John [did that] last week.’

• Fronting, topicalization:

‘John likes [reading books]’ →‘[Reading books], John likes’

• Coordination:

John [saw Marry] and [said ‘hi’]

• …

Note, however, these tests are leaky, e.g., ‘[John saw] and [Peter greated] Marry’ (see Müller 2016, for more examples).

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 7 / 31

SLIDE 2

Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Formal defjnition

A phrase structure grammar is a tuple (Σ, N, S, R) Σ is a set of terminal symbols N is a set of non-terminal symbols S ∈ N is a distinguished start symbol R is a set of rules of the form

αAβ → γ for A ∈ N α, β, γ ∈ Σ ∪ N

• The grammar accepts a sentence if it

can be derived from S with the rewrite rules R S NP John VP V saw NP Marry S → NP VP VP → V NP NP → John | Marry V → saw

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 8 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Example derivation

The example grammar: S → NP VP VP → V NP NP → John | Marry V → saw

• Phrase structure grammars derive a sentence with

successive application of rewrite rules.

S ⇒NP VP ⇒John VP ⇒John V NP ⇒John saw NP ⇒John saw Marry

• r, S

∗

⇒John saw Marry

• The intermediate forms that contain non-terminals are

called sentential forms

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 9 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Chomsky hierarchy of grammars

type 0 Recursively enumerable, recognized by Turing machines (HPSG, LFG) αAβ → γ type 1 Context sensitive, recognized by linear-bound automaton αAβ → αγβ, γ ̸= ϵ type 2.1 Mildly context sensitive (TAG, CCG) type 2 Context free, recognized by push-down automata A → α type 3 Regular, recognized by fjnite-state automata A → aB

• r

A → Ba

In all of the above A and B are non-terminals, a is a terminal symbol, α, β, γ are sequences of terminals and non-terminals, and ϵ is the empty string.

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 10 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Some examples

• Regular grammars (fjnite-state automata) do not have any

memory can represent a∗b∗, but not anbn

• Finite-state automata are used in many tasks in CL,

including morphological analysis, partial parsing.

• Context free grammars (push-down automata) uses a stack

can represent anbn, anbmcmdn, but not anbmcndm

• Context-free grammars form the basis of most natural

language parsers

• Context-sensitive languages can do all of the above but

they are too powerful, hence too expensive

• Some level of context sensitiveness seems to be necessary

for some syntactic phenomena.

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 11 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Chomsky hierarchy: the picture

Regular Context Free Context Sensitive Recursively Enumerable

• Chomsky hierarchy of languages form a hierarchy (with some

care about empty language)

• It is often claimed that mildly context sensitive grammars

(dashed ellipse) are adequate for representing natural languages

• Note, however, not even every regular language is a potential

natural language (e.g., a∗bbc∗). The possible natural languages probably cross-cut this hierarchy (shaded region)

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 12 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Expressiveness of grammar classes

• The class of grammars adequate for formally describing

natural languages has been an important question for (computational) linguistics

• For the most part, context-free grammars are enough, but

there are some examples, e.g., from Swiss German (Shieber 1985) Jan säit das… …mer em Hans es huss hälfed aastriiche …we Hans (dat) house (acc) helped paint Note that this resembles anbmcndm.

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 13 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Constituency grammars and parsing

• Context-free grammars are often parseable with

complexity O(n3) using dynamic programming algorithms

• Mildly context-sensitive grammars can also be parsed in

polynomial time (O(n6))

• Often greedy search algorithms are used (even for CFG or

equivalent classes of grammars)

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 14 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Constituency grammars summary

• Constituency, or phrase structure, grammars builds on the

idea that some words form constituents (non-terminals in a formal grammar)

• They are well studied, both in linguistics and computer

science

• Context free grammars are the most common class of

phrase structure grammars used in parsing natural or programming languages (maybe with some extensions)

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 15 / 31

SLIDE 3

Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Dependency grammars

• Dependency grammars gained popularity in (particularly

in computational) linguistics rather recently, but their roots can be traced back to a few thousand years (modern dependency grammars are attributed to Tesnière 1959)

• The main idea is capturing the relation between the words,

rather than (abstract) phrases John saw Marry

subject

• bject

root

Note: like constituency grammars, we will not focus on a particular dependency formalism, but discuss it in general in relation to parsing.

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 16 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Properties of dependency grammars

John saw Marry

subject

• bject

root

• The structure of the sentence is represented by asymmetric

binary links between lexical items

• Each relation defjnes one of the words as the head and the
• ther as dependent
• The links (relations) have labels (dependency types)
• Most dependency grammar require each word to have only

a single head

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 17 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

A more realistic example

From the AP comes this story :

root nmod det case nsubj det punct

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 18 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

How to determine heads

• 1. Head (H) determines the syntactic category of the

construction (C) and can often replace C

• 2. H determines the semantic category of C; the dependent (D)

gives semantic specifjcation

• 3. H is obligatory, D may be optional
• 4. H selects D and determines whether D is obligatory or
• ptional
• 5. The form and/or position of dependent is determined by

the head

• 6. The form of D depends on H
• 7. The linear position of D is specifjed with reference to H

(from Kübler, McDonald, and Nivre 2009, p.3–4) Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 19 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Issues with head assignment and dependency labels

• Like the tests for constituency, determining heads are not

always straightforward

• A construction is called endocentric if the head can replace

the whole construction, exocentric otherwise

syntactic parsing

amod

saw Marry

• bj
• It is often unclear whether dependency labels encode

syntactic or semantic functions

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 20 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Some tricky constructions

• Coordination

John and Marry work

subj cc conj

John and Marry work

subj cc conj

John and Marry work

subj conj conj

• Prepositional phrases

…works from home

vcompl pcompl

…works from home

nmod case

• Subordinate clauses

think that they can…

• bj

sbar subj

think that they can…

• bj

mark subj

• Auxiliaries vs. main verbs

…will work

root aux

…will work

root aux

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 21 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Projective vs. non-projective dependencies

• If a dependency graph has no crossing edges, it is said to

be projective, otherwise non-projective

• Non-projectivity stem from long-distance dependencies

and free word order A non-projective tree example:

A hearing is scheduled

• n

the issue today .

ROOT VC PUNC SBJ NMOD PP TMP NP NMOD (tree reproduced from McDonald and Satta 2007) Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 22 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Parsing with dependency grammars

• Projective dependency parsing can be done in polynomial

time

• Non-projective parsing is NP-hard (without restrictions)
• For both, it is a common practice to use greedy (e.g., linear

time) algorithms

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 23 / 31

SLIDE 4

Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Dependency vs. constituency

• Constituency grammars are based on units formed by a

group of lexical items (constituents or phrases)

• Dependency grammars model binary head–dependent

relations between words

• Most of the theory of parsing is developed with

constituency grammars

• Dependency grammars has recently become more popular

in CL

• Note that many formalisms and treebanks follow a hybrid

approach, using ideas from both

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 24 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Conversion between constituencies and dependencies

• Although non-trivial conversion between dependency and

consitituency annotation is possible

• On can take the path between two words as a dependency

relation

S NP John VP V saw NP Marry John saw Marry

subject

• bject

root

• The conversion from constituencies to dependencies is a

common practice in the fjeld

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 25 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Chomskian tradition

1950’s Phrase structure/context sensitive grammars - more emphasis on precise/computational descriptions 1960’s Transformational grammars - not as precise defjnition, diffjcult for computational approaches 1980’s Government and binding theory (GB) 1990’s Minimalist program

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 26 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Some computationally oriented grammar formalisms

Some of the grammars that we will encounter in the articles we will read are.

• Generalized phrase structured grammars (GPSG)
• Head-driven phrase structure grammar (HPSG)
• Lexical functional grammar (LFG)
• Tree adjoining grammars (TAG)
• Combinatory categorial grammar (CCG)

Common themes:

• they are lexicalized (we’ll see later what that means)
• most combine features of both constituency and

dependency grammars

• most of them are more expressive/complex than CFGs

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 27 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Summary

• A grammar a formal device for specifying a language
• Grammars are one of the important components of a

parser, they can be hand-crafted or extracted from a treebank

• Most of the parsing theory and practice is based on

constituency, particularly context-free, grammars

• Dependency grammars have become more popular

recently, and often easier to use in NLP applications

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 28 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Where to go from here?

• Müller (2016) is a new open-source text book on Grammar

formalisms.

• Aho and Ullman (1972) is the classical reference (available
• nline) for parsing (programming languages) and also

includes discussion of grammar classes in the Chomsky

• hierarchy. A more up-to-date alternative is Aho, Lam, et al.

(2007).

• There is a brief introductory section on dependency

grammars in Kübler, McDonald, and Nivre (2009), for a classical reference see Tesnière (2015), English translation

• f the original version (Tesnière 1959).

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 29 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Pointers to some treebanks

Treebanks are the main resource for statistical parsing. A few treebank-related resources to have a look at until next time:

• Tübingen treebanks:

TüBa-D/Z written German TüBa-D/S spoken German TüBa-E/S spoken English TüBa-J/S spoken Japanese available from http:

//www.sfs.uni-tuebingen.de/en/ascl/resources/corpora.html

• Universal dependencies project, documentation, treebanks:

http://universaldependencies.org/

• TüNDRA - a treebank search and visualization application

with the above treebanks and few more

– Main version:

https://weblicht.sfs.uni-tuebingen.de/Tundra/

– New version (beta):

https://weblicht.sfs.uni-tuebingen.de/tundra-beta/

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 30 / 31 Recap Introduction Constituency grammars Dependency grammars Grammar formalisms Finale

Next week: your second assignment

• We will got through your example sentences and try to

analyze them with constituency and dependency annotations

• Before next Thursday:

– Annotate the sentences using UD annotation scheme – Annotate the sentences using constituency annotations (you can freely choose the annotation scheme, if you need inspiration check out the Penn treebank annotation guidelines)

Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 31 / 31

SLIDE 5

Bibliography

Aho, Alfred V., Monica S. Lam, Ravi Sethi, and Jefgrey D. Ullman (2007). Compilers: Principles, Techniques, and Tools.

• 2nd. Pearson Education. isbn: 0-321-48681-1.

Aho, Alfred V. and Jefgrey D. Ullman (1972). The Theory of Parsing, Translation, and Compiling. Volume I: Parsing. Vol. I. Upper Saddle River, NJ, USA: Prentice-Hall. isbn: 0-13-914556-7. url: http://dl.acm.org/citation.cfm?id=578789. Kübler, Sandra, Ryan McDonald, and Joakim Nivre (2009). Dependency Parsing. Synthesis lectures on human language technologies. Morgan & Claypool. isbn: 9781598295962. McDonald, Ryan and Giorgio Satta (2007). “On the complexity of non-projective data-driven dependency parsing”. In: Proceedings of the 10th International Conference on Parsing Technologies. Association for Computational Linguistics, pp. 121–132. Müller, Stefan (2016). Grammatical theory: From transformational grammar to constraint-based approaches. Vol. 1. Textbooks in Language Sciences. Language Science Press. isbn: 9783944675213. doi: 10.17169/langsci.b25.168. url: http://langsci-press.org//catalog/book/25. Shieber, Stuart M. (1985). “Evidence against the context-freeness of natural language”. In: Linguistics and Philosophy 8.3, pp. 333–343. doi: 10.1007/BF00630917. Tesnière, Lucien (1959). Éléments de syntaxe structurale. Paris: Éditions Klinksieck. — (2015). Elements of Structural Syntax. Trans. by Timothy John Osborne and Sylvain Kahane. Amsterdam: John Benjamins Publishing Company. isbn: 9789027212122. Ç. Çöltekin, SfS / University of Tübingen October 27, 2016 A.1