Towards a Rewriting Framework for Textual Entailment
0 likes118 views
1) The document proposes using KIML representations of natural language paired with inference rules as a version of Textual Inference Logic (TIL) to determine textual entailment, translating TIL formulas to Maude rewriting logic for automated reasoning. 2) An example shows KIML representations of concepts, roles, contexts for the sentence "A crow slept" and how inference could derive facts like the crow's instantiability in context t. 3) Experimental results demonstrated proving 7 simple textual entailment relationships using the proposed rewriting framework in Maude. However, many challenges remain around producing unambiguous KIML representations and modeling temporal information.
![Textual Entailment
KIML
Inference Rules
Maude
An example: a crow slept
Conceptual Structure:
role(cardinality restriction,crow-1,sg)
role(sb,sleep-4,crow-1)
subconcept(crow-1,[crow#n#1,crow#n#2,brag#n#1])
subconcept(sleep-4,[sleep#v#1,sleep#v#2])
Contextual Structure:
instantiable(crow-1,t)
instantiable(sleep-4,t)
top context(t)
Temporal Structure:
trole(when,sleep-4,interval(before,Now))
Valeria de Paiva LSFA 2014](https://image.slidesharecdn.com/lsfa2014-171214020812/85/Towards-a-Rewriting-Framework-for-Textual-Entailment-4-320.jpg)
![Textual Entailment
KIML
Inference Rules
Maude
KIML versus FOL
In FOL could write ∃Crow∃Sleep.Sleep(crow)
Instead we will use basic concepts from a parameter ontology
O (could be Cyc, SUMO, UL, KM, etc...)
Instead of FOL have Skolem constant crow-1 a subconcept of
an ambiguous list of concepts:
subconcept(crow-1,[crow#n#1,crow#n#2,brag#n#1])
Same for sleep-2 and have roles relating concepts
role(sb,sleep-4,crow-1)
meaning that the sb=subject of the sleeping event is a crow
concept
Valeria de Paiva LSFA 2014](https://image.slidesharecdn.com/lsfa2014-171214020812/85/Towards-a-Rewriting-Framework-for-Textual-Entailment-6-320.jpg)
![Textual Entailment
KIML
Inference Rules
Maude
Ed knows that the crow slept
alias(Ed-0,[Ed])
role(prop,know-1,ctx(sleep-8))
role(sb,know-1,Ed-0)
role(sb,sleep-8,crow-6)
subconcept(Ed-0,[male#n#2])
subconcept(crow-6,[crow#n#1,crow#n#2,brag#n#1])
subconcept(know-1,[know#v#1,...,sleep-together#v#1])
subconcept(sleep-8,[sleep#v#1,sleep#v#2])
context(ctx(sleep-8)), context(t)
context-lifting-relation(veridical,t,ctx(sleep-8))
context-relation(t,ctx(sleep-8),crel(prop,know-1))
instantiable(Ed-0,t)
instantiable(crow-6,ctx(sleep-8))
instantiable(sleep-8,ctx(sleep-8))
Valeria de Paiva LSFA 2014](https://image.slidesharecdn.com/lsfa2014-171214020812/85/Towards-a-Rewriting-Framework-for-Textual-Entailment-9-320.jpg)
Towards a Rewriting Framework for Textual Entailment
- 1. Textual Entailment KIML Inference Rules Maude Towards a Rewriting Framework for Textual Entailment Valeria de Paiva Vivek Nigam Nuance Communications, CA, US Universidade Federal da Para´ıba, PA, Brazil LSFA 2014, September, 2014 Valeria de Paiva LSFA 2014
- 2. Textual Entailment KIML Inference Rules Maude Textual Entailment Textual entailment methods recognize, generate, and extract pairs T, H of natural language expressions, such that a human who reads (and trusts) T would infer that H is most likely also true (Dagan, Glickman & Magnini, 2006) Example: (T) The drugs that slow down Alzheimer’s disease work best the earlier you administer them. (H) Alzheimer’s disease can be slowed down using drugs. T ⇒ H A series of competitions since 2004, ACL “Textual Entailment Portal”, many different systems... Valeria de Paiva LSFA 2014
- 3. Textual Entailment KIML Inference Rules Maude Basic Idea of Work Use Xerox’s PARC Bridge system as a black box to produce NL representations of sentences in KIML (Knowledge Inference Management Language). KIML + inference rules = TIL (version of) Textual Inference Logic Translate TIL formulas to a theory in Maude, the SRI rewriting system. Use Maude rewriting to prove Textual Entailment “theorems”. NB: Bridge has its own reasoning module called ECD (for entailment and contradiction detection) which does packed rewriting... Valeria de Paiva LSFA 2014
- 4. Textual Entailment KIML Inference Rules Maude An example: a crow slept Conceptual Structure: role(cardinality restriction,crow-1,sg) role(sb,sleep-4,crow-1) subconcept(crow-1,[crow#n#1,crow#n#2,brag#n#1]) subconcept(sleep-4,[sleep#v#1,sleep#v#2]) Contextual Structure: instantiable(crow-1,t) instantiable(sleep-4,t) top context(t) Temporal Structure: trole(when,sleep-4,interval(before,Now)) Valeria de Paiva LSFA 2014
- 5. Textual Entailment KIML Inference Rules Maude KIML A representation language based on events (neo-Davidsonian), concepts, roles and contexts, McCarthy-style Using events, concepts and roles is traditional in NL semantics (Lasersohn) Usually equivalent to FOL (first-order logic), ours a small extension, contexts are like modalities Language based on linguists’ intuitions Exact formulation still being decided: e.g. not considering temporal assertions, yet... Valeria de Paiva LSFA 2014
- 6. Textual Entailment KIML Inference Rules Maude KIML versus FOL In FOL could write ∃Crow∃Sleep.Sleep(crow) Instead we will use basic concepts from a parameter ontology O (could be Cyc, SUMO, UL, KM, etc...) Instead of FOL have Skolem constant crow-1 a subconcept of an ambiguous list of concepts: subconcept(crow-1,[crow#n#1,crow#n#2,brag#n#1]) Same for sleep-2 and have roles relating concepts role(sb,sleep-4,crow-1) meaning that the sb=subject of the sleeping event is a crow concept Valeria de Paiva LSFA 2014
- 7. Textual Entailment KIML Inference Rules Maude What is Different? Corresponding to formulas in FOL, KIML has a collection of assertions that, read conjunctively, correspond to the semantics of a (fragment of a) sentence in English. Concepts in KIML – similar to Description Logic concepts primitive concepts from an idealized version of the chosen ontology constructed-on-the-fly concepts, always sub-concepts of some primitive concept. concepts are as fine or as coarse as needed by the application Roles connect concepts: deciding which roles with which concepts a big problem... for linguists Roles assigned in a consistent, coherent and maximally informative way by the NLP module Contexts as a quantification device Valeria de Paiva LSFA 2014
- 8. Textual Entailment KIML Inference Rules Maude Contexts for Quantification Contexts for modelling negation, implication, as well as propositional attitudes and other intensional phenomena. There is a first initial context (written as t), roughly what the author of the sentence takes the world to be. Contexts for existential statements about the existence and non-existence in specified possible worlds of entities that satisfy the intensional descriptions specified by our concepts. Propositional attitudes predicates (knowing, believing, saying,...) relate contexts and concepts in our logic. Concepts like knowing, believing, saying introduce context that represents the proposition that is known, believed or said. Valeria de Paiva LSFA 2014
- 9. Textual Entailment KIML Inference Rules Maude Ed knows that the crow slept alias(Ed-0,[Ed]) role(prop,know-1,ctx(sleep-8)) role(sb,know-1,Ed-0) role(sb,sleep-8,crow-6) subconcept(Ed-0,[male#n#2]) subconcept(crow-6,[crow#n#1,crow#n#2,brag#n#1]) subconcept(know-1,[know#v#1,...,sleep-together#v#1]) subconcept(sleep-8,[sleep#v#1,sleep#v#2]) context(ctx(sleep-8)), context(t) context-lifting-relation(veridical,t,ctx(sleep-8)) context-relation(t,ctx(sleep-8),crel(prop,know-1)) instantiable(Ed-0,t) instantiable(crow-6,ctx(sleep-8)) instantiable(sleep-8,ctx(sleep-8)) Valeria de Paiva LSFA 2014
- 10. Textual Entailment KIML Inference Rules Maude Inference to build reps and to reason with them In previous example can conclude: instantiable(sleep-8,t) if knowing X implies X is true. (Can conclude instantiable(crow-6,t) too, for definitiveness reasons..) Happening or not of events is dealt with by the instantiability/uninstantiability predicate that relates concepts and contexts e.g. Negotiations prevented a strike Contexts can be: veridical, antiveridical or averidical with respect to other contexts. Have ‘context lifting rules’ to move instantiability assertions between contexts. Valeria de Paiva LSFA 2014
- 11. Textual Entailment KIML Inference Rules Maude Inference Rules The totally obvious s → s s → t s → r r → t Valeria de Paiva LSFA 2014
- 12. Textual Entailment KIML Inference Rules Maude Inference Rules Inheritance rules Nina has a canary, canary bird Nina has a bird Ed kissed Nina, kiss touch Ed touched Nina Every carp is a fish, carp koi Every koi is a fish She didn’t give him a bird, bird canary She didn’t give him a canary Valeria de Paiva LSFA 2014
- 13. Textual Entailment KIML Inference Rules Maude Modifiers Inference Rules Ed arrived in the city by bus Ed arrived in the city Ed did not arrive in the city Ed did not arrive in the city by bus Ed arrived in the city, Ed person A person arrived in the city Ed arrived in Rome, Rome city Ed arrived in a city Note that Ed did not arrive in the city by bus does not entail that Ed did not arrive in the city. Valeria de Paiva LSFA 2014
- 14. Textual Entailment KIML Inference Rules Maude Implicative Commitments Rules Preserving polarity: “Ed managed to close the door” → “Ed closed the door” “Ed didn’t manage to close the door” → “Ed didn’t close the door”. The verb “forget (to)” inverts polarities: “Ed forgot to close the door” → “Ed didn’t close the door” “Ed didn’t forget to close the door” → “Ed closed the door”. There are six such classes, depending on whether positive environments are taken to positive or negative ones. Accommodating this fine-grained analysis into traditional logic description is further work. (Nairn et al 2006 presents an implemented recursive algorithm for composing these rules) Valeria de Paiva LSFA 2014
- 15. Textual Entailment KIML Inference Rules Maude Today: Towards a Rewriting Framework A implementation of TIL, using the traditional rewriting system Maude to reason about the logical representations produced by the NLP module we are considering. Hand-correct the representations given by the NLP module: the goal here is not to obtain correct representations, but to work logically with correct representations. Maude system is an implementation of rewriting logic developed at SRI International. Maude modules (rewrite theories) consist of a term-language plus sets of equations and rewrite-rules. Terms in rewrite theory are constructed using operators (functions taking 0 or more arguments of some sort, which return a term of a specific sort). Valeria de Paiva LSFA 2014
- 16. Textual Entailment KIML Inference Rules Maude A Rewriting Framework A rewrite theory is a triple (Σ, E, R), with (Σ, E) an equational theory with Σ a signature of operations and sorts, and E a set of (possibly conditional) equations, and with R a set of (possibly conditional) rewrite rules. A few logical predicates for our natural languages representations: (sub)concepts, roles, contexts and a few relations between these. But the concepts that the representations would use in a minimally working system in the order of 135 thousand, concepts in WordNet. Scaling issues! Valeria de Paiva LSFA 2014
- 17. Textual Entailment KIML Inference Rules Maude Maude Rewriting Framework Basic rewriting sorts: Relations, SBasic and UnifSet TIL basic assertions such as canary bird belong to Relations. Concept and contextual assertions, such as instantiable(drink-0,t) belong to the SBasic basic statements sort. The third basic sort, UnifSet, contains unification of skolem constants, such as crow-6 := bird-1. This last sort is necessary for for unifying skolem constants, engine of derivations here. Much more is needed... Valeria de Paiva LSFA 2014
- 18. Textual Entailment KIML Inference Rules Maude Experimental Results: a few theorems 1. a crow was thirsty a thirsty crow 2. a thirsty crow a crow 3. ed arrived and the crow flew away the crow flew away 4. ed knew that the crow slept the crow slept 5. ed did not forget to force the crow to fly the crow flew 6 the crow came out in search of water the crow came out 7. a crow was thirsty a bird was thirsty All easy, but the state-of-the-art winner does not do all of them... Valeria de Paiva LSFA 2014
- 19. Textual Entailment KIML Inference Rules Maude Conclusions Proof-of-concept framework Introduced a general rewriting framework, using KIML assertions and TIL inference system for textual entailment Demonstrated by example that framework can be implemented in Maude and used it to prove in an semi-automated fashion whether a sentence follows from another ‘shallow theorem proving’ for common sense applications? Many problems: black box, ambiguity, temporal information, etc.. Valeria de Paiva LSFA 2014
- 20. Textual Entailment KIML Inference Rules Maude Thanks! WOLLIC 2015 in Bloomington, Indiana, USA. Valeria de Paiva LSFA 2014