Using Proofs to Compute Implicatures [Abstract]

Bruno Woltzenlogel Paleo 1, * Ekaterina Lebedeva 2
* Corresponding author
1 VERIDIS - VERIfication pour les systèmes DIStribués
Inria Nancy - Grand Est, LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications : UMR7503
2 CALLIGRAMME - Linear logic, proof networks and categorial grammars
INRIA Lorraine, LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications
Abstract : Defining natural language discourse semantics compositionally has been one of the main challenges in the areas of natural language semantics and pragmatics. In [Mon70], Montague argued that there are no essential differences between natural and formal programming languages. He developed a semantics that assigns a lambda term (i.e. a functional program) to each word of a sentence. The meaning of the whole sentence is then obtained by composing the lambda terms via functional application [Mon73,DWP81]. However, Montague's semantics was originally limited to a single sen- tence, and extending it to discourses made of several sentences turned out to be non-trivial due to linguistic phenomena such as donkey anaphora. This problem was recently solved by de Groote [dG06], who extended Montague's semantics by using a continuation-passing-style technique [SW00]. In this approach the lambda terms take two additional argu- ments: the first argument stands for the left context or environment (i.e. a concise representation of what has already been processed); and the sec- ond argument stands for the continuation (i.e. everything that remains to be processed). Current work aims at using and improving this continuation-based ap- proach in order to handle other kinds of linguistic side effects, such as pressupositions [Kar74] and implicatures [Gri75], which are informally what is taken for granted and not explicitly expressed in a discourse. Un- derstanding how presuppositions and implicatures are triggered, com- puted and projected are among the hardest open problems in discourse semantics. Preliminary results [dGL10] showed that presuppositions stemming from proper names (and definite descriptions) can indeed be handled by de Groote's continuation-based approach extended with an exception raising and handling mechanism. The triggering of a presupposition requires a search for a referent marker for the proper name in the environment. If the search fails, an exception is raised. The code that handles the exception corresponds to accommodation of the presupposition. In this talk we show that a similar exception-based approach can, at least in principle, be used for implicatures. During the processing of the dis- course, a proof of the current sentence from the axioms contained in the environment is performed. If the search fails, an exception is raised. Han- dling the exception consists of abducing facts that are needed to complete the failed proofs. The abduced facts are the implicatures of the sentence, they are computed from the failed proofs and added to the environment.
Document type :
Conference papers
Liste complète des métadonnées
Contributor : Bruno Woltzenlogel Paleo <>
Submitted on : Friday, December 10, 2010 - 1:50:13 PM
Last modification on : Thursday, February 21, 2019 - 2:02:02 PM


  • HAL Id : hal-00545494, version 1



Bruno Woltzenlogel Paleo, Ekaterina Lebedeva. Using Proofs to Compute Implicatures [Abstract]. Computability in Europe, Jun 2010, Ponta Delgada, Portugal. ⟨hal-00545494⟩



Record views