Compositional Semantic Vector Models with a Simplified Quantum Logic

Abstract : We view the vectors of a semantic distributional model as vectors of a semi-module over the semi-ring of a real interval with its canonical lattice structure. In this approach, the quantum logic of subspaces is distributive and Boolean on the sublattice of subspaces generated by sets of basis vectors. If the basis vectors correspond to strings of words, the semi-module classifies words like a thesaurus. Its Boolean sublattice includes the concepts corresponding to single words. If the basis vectors identify with single words the connectives of quantum logic lift from the basis vectors to the whole space as linear maps. Concept vectors are linked to the vectors assigned by a pregroup grammar through the classes they determine. The functional composition of pregroup semantics corresponds to a composition of concepts by logical connectives. Examples including compound noun-phrases like \pi{no bank} and \pi{red herring} illustrate the equivalence of the quantum logic of subspaces to the compositional logic of pregroup grammars.
Liste complète des métadonnées
Contributeur : Anne Preller <>
Soumis le : samedi 1 mai 2010 - 10:06:43
Dernière modification le : jeudi 11 janvier 2018 - 06:26:53


  • HAL Id : lirmm-00479633, version 1



Anne Preller, Violaine Prince. Compositional Semantic Vector Models with a Simplified Quantum Logic. RR-10014, 2010, pp.0010026. 〈lirmm-00479633〉



Consultations de la notice