Combinatory Categorial Grammar Goal Mildly-context sensitive grammar framework, with provable polynomial parseability “To present the intuitions behind Combinatory Categorial Grammar as a framework for relating the structure of a linguistic expression to the Capable of handling a large range of linguistic phenomena, across a broad meaning it realizes; and to illustrate a development platform for CCG.” spectrum of language types Overview Perspicuous interface between structure ( � syntax � ) and meaning ( � semantics 1. CCG & OpenCCG in the larger context of grammar engineering OpenCCG 2. CCG: Analysing the structure of expressions Platform for efficient parsing and realization with CCG grammars 3. CCG: Adding (linguistic) meaning to structure Computational lexicon facilitating inheritance over syntactic and semantic common structure, enabling reuse within a single grammar and across 4. OpenCCG: Organization and implementation of grammars Statistics-based realization; development to include StatCCG for parsing Grammar engineering Combinatory Categorial Grammar “The effort to develop natural language grammars and computational platforms for broad-coverage, robust, and efficient interpretation and & the OpenCCG platform production of linguistic utterances” Resource reuse in real-life grammar engineering Large-scale grammars, moving towards multi-lingual resource reuse Geert-Jan M. Kruijff Statistical NLP for acquiring, and processing with, large grammars But : between broad-coverage syntax and semantics; domain- Language Technology Lab specific interpretation in and beyond NL grammar; complexity DFKI GmbH Various available platforms gj@dfki.de / http://www.dfki.de/~gj
Formal systems In the lexicon, words are assigned categories ... usually talk about structured objects, the resources A category specifies the grammatical use ( syntactic behavior ) of the word ... to collections of which they apply operations A category can be atomic , or a function using “slashes” \ (left) and / (right) ... in order to determine some global properties of these collections Functional categories are written “result first”: RES | ARG 1 ... |ARG n Resource sensitivity governs the way in which a system Examples can use its resources: Nouns: N How often resources can be used, Adjectives: N / N When they can be assembled into more complex resources, How they can be modified Determiners: NP / N Transitive verbs: (S \ NP) / NP, also written as S \ NP / NP Part 1 Structured objects are linguistic signs Monostratal, multi-level representation A combinatory categorial approach to Grammatical composition is resource sensitive analysing the structure of expressions Signs can usually be used only once (no arbitrary copying, wasting) Signs cannot be arbitrarily combined or arranged ( linearization, dependency ) Control over when operations defining grammatical composition can be applied makes it possible to avoid a collapse to “anything goes”
Permutation -inducing rules based on the composition combinator B CCG adds further rules, based on the combinators of Curry & Feys � combinatory logic Forward crossed composition (> B x ): X/Y Y\Z ⇒ X\Z Forward type raising (> T ) : X ⇒ Y/(Y\X) Backward crossed composition (< B x ): Y/Z X\Y ⇒ X/Z Forward harmonic composition (> B ): X/Y Y/Z ⇒ X/Z < B x needed for e.g. heavy-NP shift: These rules induce associativity : Ed saw briefly his old friend from Skye Ed saw Ann (s\np)/np (s\np)/np np (s\np)\(s\np) (s\np)\(s\np) np < B x np (s\np)/np (s\np)/np np (s\np)/np > T > s/(s\np) s/(s\np) s\np > B < s/np s/np s > s Associativity in the structure of language Basic rules of combination Associativity introduces “flexible constituency”, non-associativity Forward application (>): X/Y Y ⇒ X strict domains Backward application (<): Y X\Y ⇒ X Evidence of associativity can be found in extraction, coordination, the interaction of intonational structure and syntactic structure, ... Example derivation Example derivation Ed saw Ann team that defeated Brazil n n\n/(s/np) (s\np)/np np np (s\np)/np np >T > s/(s\np) s\np >B s/np < s > n\n < n
★ ★ Modal control over access to combinators � Controlled through ★ � Modal marking m on connectives: \ m and / m All modes inherit from ★, i.e. all have access to Combinators are defined for particular modalized connectives only application Hierarchy over modals to inherit combinator accessibility (behavior) (>) X / ★ Y Y ⇒ X (<) Y X \ ★ Y ⇒ X Lexically specified derivational control The lexicon completely determines structure building the red car np/ ★ n n/ ★ n n If a modalized connective, controlling the applicability of a rule, is absent from the lexicon, then the rule can never be applied > n < No more need for ad hoc constraints on rule applicability np Universal set of combinator rules: Combinators as formal universals Application (non-associative) Crossed composition allows for ungrammatical permutations: � � Harm. composition (associative) Cross. composition (assoc a pub nice in Edinburgh a nice in Edinburgh pub np/n n n 1 /n 2 n 3 \n 4 np/n n 1 /n 2 n 3 \n 4 n >B <B � � n 1 \n 4 n 3 /n 2 < > n 1 n 3 > > np np All behaviors English-specific instantiation of < B � : English-specific ban on > B � : The grammar English backward crossing composition < B � : of English does not contain forward crossed � Y/Z X\Y � B X/Z where X=Y=s\$ composition. � The modalities restrict the upper limit, i.e. the strongest rule(s) that can be applied. If a category contains modalities that are stronger than required for a rule, these modalities can be carried over in the result (”no power loss”).
★ ★ ★ a pub nice in Edinburgh np/n n n 1 /n 2 n 3 \n 4 >B � n 1 \n 4 < n 1 Anna might marry Manny > np np (s\ � np)/ � np np (s\ � np)/ � (s\ � np) >B English-specific ban on > B � : The grammar (s\ � np)/ � np a of English does not contain forward crossed pub nice in Edinburgh > composition. � � s\ � np np/n n n 1 / � n 2 n 3 \ ★ n 4 >B � < � X s Observe: ● ⊆ � , so we can keep ● No need to ban > B � *man that John knew [that, s/ � s] [saw Ann, s\ � np] Complementizer “that” gets B mode � , which makes B � inapplicable � a nice in Edinburgh pub � np/n n 1 /n 2 n 3 \n 4 n <B � n 3 /n 2 Harmonic composition > n 3 > Controlled through � , i.e. accessible to � but not to ★ np (> B ) X/ � Y Y/ � Z ⇒ B X/ � Z ( < B ) Y\ � Z X\ � Y ⇒ B X\ � Z English-specific instantiation of < B � : English backward crossing composition < B � : � Y/Z X\Y � B X/Z where X=Y=s\$ � Examples a nice in Edinburgh pub A/ � B B/ � C ⇒ B A/ � C ● ⊆ � , observe that we remain with ● (not A/ � � � np/n n 1 / � n 2 n 3 \ ★ n 4 n <B � � X / � , hence the rule is not applicable A/ � B B/ ★ C ⇒ / B A/ ? C ★ ⊆
★ Modeling the complexity of meaning Combinatory analysis Relational structure: relations indicate how parts contribute to the overall Governed by Principle of Adjacency: String-based, not logic-based meaning of the construction ( interpretative import ) Building structures using combinatory rules Ontological richness: differentiation through ontological sorting of propositions Multi-modal combinatory analysis Linguistic meaning Applicability of combinatory rules is controlled through modalized slashes Meaning as far as linguistically realized (no inference, limited resolution) Control is thus fully lexicalized, the rule component is universal Information structure: “Pragmatization of semantics” man that sleeps and he talks s\np (s\s)/s np s\np < s Part 2 > s\s <B s\np Relating the structure of an expression Simplified coordination rule (for “conjoining like categories”) man that sleeps and he talks X CONJ X’ � X” (< � >) to the meaning(s) it realizes s\ � np (s\ ★ s)/ ★ s np s\ � np < s > � � s\ ★ s <B � X “The ability to speak does not make you intelligent.” Category (not a rule) for coordination Qui-Gon Jinn to Obi-Wan Kenobi, talking about Jar Jar Binks Star Wars Episode 1: The Phantom Menace [and, (X\ ★ X)/ ★ X]: * man that [sleeps, s \ � np ] [and he talks, s\ ★ s ] Coordinator “and” gets most restrictive modes, to make B inapplicable
Recommend
More recommend
