Slot Grammar - A specific type of Dependency...

Slot Grammar

Slot GrammarA specific type of Dependency Grammar

Kurt Eberle

October 26 2017

1 / 50

Slot Grammar

Outline

Slot GrammarBackgroundDesignLiteratureDependency TreesTwo Level AnalysisRealization

Data Structures and ParsingLexiconGrammar

2 / 50

Slot Grammar

Outline

Slot GrammarBackgroundDesignLiteratureDependency TreesTwo Level AnalysisRealization

Data Structures and ParsingLexiconGrammar

3 / 50

Slot Grammar

Background

HistoryI A version of Dependency GrammarI Michael McCordI from the eighties onI IBM Thomas J. Watson Research CenterI IBM research centers EuropeI Spinoff companies: Synthema (IT), linguatec, lingenio (DE)I projects:

Logic based Machine Translation (LMT),Deep analysis in Watson question answering system

I Grammars for many languages. . . ESG, GSG, FSG, . . . Italian,Spanish, Danish, Hebrew, Arabian, .̇.

4 / 50

Slot Grammar

Design

Design

I Design closely related to implementation issuesI Implementation in PrologI . . . and later in C

5 / 50

Slot Grammar

Design

Design

I Own set of fine-grained dependency relationsclassified into

I subcategorized argument slotssubj, obj, auxcomp, nobj, . . .

I free modifier slotsvadv, vsubconj, ndet, nadj, . . .

I lexicalized theoryI passivization rule, . . .

I Exptraposition handlingI . . .

6 / 50

Slot Grammar

Literature

LiteratureMcCord 1980 Slot Grammars

Computational Linguistics, vol. 6, pp. 31-43.McCord 1982 Using Slots and Modifiers in Logic Grammars for Natural Language

Artificial Intelligence, vol. 18, pp. 327-367.McCord 1989 Design of LMT: A Prolog-based Machine Translation System

Computational Linguistics, vol. 15, pp. 33-52.McCord 1989 A New Version of Slot Grammar

Research Report RC 14506, IBM Research Division, Yorktown HeightsMcCord 1990 Slot Grammar: A System for Simpler Construction of Practical Natural

Language Grammarsin: Proceedings of the International Symposium on Natural Language andLogic, Springer, London

McCord 2006 The Slot Grammar Lexical FormalismResearch Report RC 23977, IBM Research Division, Yorktown Heights

McCord, Cavalli-Sforza 2007 An Arabic Slot Grammar ParserProceedings of the 5th Workshop on Important Unresolved Matters,pp. 81-88,Prague, Czech Republic

McCord 2010 Using Slot GrammarResearch Report RC 23978, IBM Research Division, Yorktown Heights

McCord, Murdock, Boguraev 2012 Deep Parsing in WatsonIBM J. RES. & DEV. VOL. 56 NO. 3/4 7 / 50

Slot Grammar

Dependency Trees

Dependency TreesExample (’Watson Tree’)

8 / 50

Slot Grammar

Dependency Trees

Dependency Trees

Tree visualization:I 90 ◦ rotation, left-to-right = top-to-bottomI 4 ’columns’:

I dependency relationsI labels (slots)I node descriptions (predicates),I node descriptions (morphosyntactic and semantic features)

9 / 50

Slot Grammar

Dependency Trees

Dependency Trees’Columns’:

I slotsI slot names

arguments (1-pl): subj, obj, comp, auxcomp, pred, objprep, . . .modifiers (0-pl): lconj, rconj, vadv, ndet, nrel, . . .

I slot fillersn (nominal), a (adjectival), en (participle),binf (bare infinitive), p (prepositional), . . .

I word ’predicates’I word reading

(optionally with link to the corresponding word sense in thelexicon)

I ’functional’/semantic arguments(positions of the corresponding (head) words)

I node descriptions10 / 50

Slot Grammar

Dependency Trees

Dependency Trees

’Columns’:I slotsI word ’predicates’I node descriptions

I morphosyntactic features(PoS + subclassifyication + agreement) noun cn pl, verb vfinpres pl, . . .

I semantic types physobj artf, sta, neg, . . .

11 / 50

Slot Grammar

Dependency Trees

Dependency Trees

Example (LMT Tree’)

12 / 50

Slot Grammar

Two Level Analysis

Two Level AnalysisI syntactic levelI ’functional’ semantic level

13 / 50

Slot Grammar

Two Level Analysis

Two Level AnalysisExample: Passive Voice’Deep’ and ’Surface’ Structure

14 / 50

Slot Grammar

Two Level Analysis

Two Level Analysis

used for . . .control phenomena

15 / 50

Slot Grammar

Two Level Analysis

Two Level Analysisused for . . .control phenomena: subject control vs object control

16 / 50

Slot Grammar

Two Level Analysis

Two Level Analysisused for . . .control phenomena in extended auxiliary structures (with/withoutpassive)

17 / 50

Slot Grammar

Two Level Analysis

Two Level Analysis

used for . . .(local and long-distance) extraposition

18 / 50

Slot Grammar

Two Level Analysis

Two Level Analysisused for . . .combining extraposition and control

19 / 50

Slot Grammar

Two Level Analysis

Two Level Analysis

used for . . .extraposition over several levels

20 / 50

Slot Grammar

Two Level Analysis

Two Level Analysis

used for . . .extraposition over several levels + passivization . . .

21 / 50

Slot Grammar

Two Level Analysis

Two Level Analysis

used for . . .coordination (scope and re-entrancy) . . .

22 / 50

Slot Grammar

Two Level Analysis

Two Level Analysis

used for . . . extraposition and coordination . . .

23 / 50

Slot Grammar

Realization

Realization

I lexicon and grammarI data structuresI parsing

24 / 50

Slot Grammar

Realization

Data Structures and ParsingI words (’wframe’)I phrasesI parser turns lexical words into phrases and combines these

phrases to phrases representing substrings of the sentenceI only one type of phrase:

’categorial’ differences expressed by PoS-type and bydifferent/differently instantiated lists of argumentssaturated phrases: list of subcategorized arguments completelyinstantiated

I modular composition (using filler rules, extraposition rules,coordination rules, ordering constraints . . . )

I extraposition done via a second list consisting of locally notrealized and (presumably) extraposed elements

25 / 50

Slot Grammar

Realization

Words

Example

26 / 50

Slot Grammar

Realization

Phrases

27 / 50

Slot Grammar

Realization

PhrasesExample

28 / 50

Slot Grammar

Realization

Parse Results

29 / 50

Slot Grammar

Realization

Lexicon

I lexical entriesI lexical rules

30 / 50

Slot Grammar

Realization

Words

I Lexicondifferent formats

I (Prolog) data base with word records (using defaults, seebelow: ’Watson’-version)

I Relational data base with tables for different types of lexicalinformation

→ morphological analysis→ wframe: interface to grammar

31 / 50

Slot Grammar

Realization

Slot Grammar-words (’Watson’-version)Word < PoS SlotFrame SynDescrs SemDescrs

32 / 50

Slot Grammar

Realization

Slot Grammar-words

ambiguities represented by alternative descriptions

33 / 50

Slot Grammar

Realization

Slot Grammar-wordsI Word < PoS SlotFrame SynDescrs SemDescrsI PoS : v, n, adj, adv, . . .I SlotFrame: [SlotNameOb(Opts) : SLConds, . . . ]I Ob = obligatory (=1) / optional (∅): ex: (obj1(n) vs obj(n))I Opts = n | v | p(. . . ) . . . (PoS of filler)I SynDescrs = Boolean combination of syntactic types (ss, so,

attr, . . . )I SemDescr = conjunction of semantic types (sent obj &

artefact)I SLConds = Boolean combination of semantic types,

SynDescrs, WordDescrs

34 / 50

Slot Grammar

Realization

Syntactic Types

35 / 50

Slot Grammar

Realization

Semantic Types

Watson

36 / 50

Slot Grammar

Realization

Relational Database

LMT/Lingenio version

37 / 50

Slot Grammar

Realization

Relational Database’Secondary’ Table (built from ’primary tables, used duringmorphological analysis)

38 / 50

Slot Grammar

Realization

Relational Database

’Secondary’ Table (built from ’primary tables, used duringmorphological analysis)

39 / 50

Slot Grammar

Realization

Morphological Analysis

Result (= input to syntactic analysis)I type informationI wframe representation:

wframe(Position,Sense,FeatureDescription,SlotFrame)

40 / 50

Slot Grammar

Realization

Morphological Analysis

Result (= input to syntactic analysis)

41 / 50

Slot Grammar

Realization

Lexical rules→ applied in morphological analysis

42 / 50

Slot Grammar

Realization

Slot Grammar phrase compositionDifferent types of rules (stored in modules)

I declaration of adjunctsI Slot filler rulesI or-slot rulesI slot ordering rulesI declaration of obligatory slotsI extraposition rulesI coordination rulesI punctuation rulesI evaluation rules

43 / 50

Slot Grammar

Realization

Slot Grammar-composition

44 / 50

Slot Grammar

Realization

Slot Grammar-composition

45 / 50

Slot Grammar

Realization

Slot Grammar-composition

46 / 50

Slot Grammar

Realization

Slot Grammar-extraposition

47 / 50

Slot Grammar

Realization

Slot Grammar-extraposition

I Declaration of extraposer slots(extraposer(obj) :- Conds.)

I Extraposition rules

48 / 50

Slot Grammar

Realization

Slot Grammar-extraposition

Example

49 / 50

Slot Grammar

Realization

Weighting of rules

Evaluation RulesSlot +++ Val (:- Conds)

obj( ) +++ 3.

50 / 50