Résumé publicCoLan
Quantification in naturallanguage: a formal and experimentalstudy
The goal of this project is to study how humans naturally compute Quantifier Scope Disambiguation (QSD) in universal-existential double-quantified sentences. More specifically, we focus on two factors influencing QSD: linear order (Johnson-Laird et al., 1989), and lexical bias (Ioup, 1975). The study consists in studying differences between French universal quantifiers "chaque" and "tous les", matched with the existential indefinite quantifier "un". We focus on a simple syntactico-semantic aspect, excluding explicit reflection of listeners/readers: the purpose is to study linguistic aspects of QSD, rather than logical abilities.
There are three different hypothesis: (1) generally, the wide scope is given to the first quantifier of the sentence, according to linear order factor (mentioned as natural interpretation, or natural scoping), (2) "chaque" and "tous les", being the strongest quantifiers, they should sometimes inverse natural scope when in second position in the sentence, according to lexical bias factor (mentioned as inverse interpretation, or inverse scoping), (3) "chaque" being stronger than "tous les", it takes wide scope more often than "tous les".
We also expect that "un" should be put between "chaque" and "tous les" on the hierarchy scale of quantifier’s strength. This result contradicts point (2) for "tous les" barely taking wide scope on "un".
Experiment : Auditory and verbal stimuli consisted of sentences combining circles with existential (un) quantifier and squares with universal (tous les/chaque) quantifiers, in any order. Eye movements were recorded with a Tobii T120 eye-tracker. 4 different drawings were presented with each stimulus and subjects had to point to the most corresponding picture to the verbal stimulus. The number and duration of the fixations are considered to reflect ambiguity.
The results confirmed several theoretical assumptions about general phenomena influencing QSD, such as linear order in sentence or lexical realization of quantifiers, confirming Ioup’s (1975) scale at its strong end. Moreover, they suggest, as we expected, that "un" should be placed between "tous les" and "chaque" with some precision about the distance with its neighbors.
P Blache, L Prevot (LPL)
Responsable Christian RETORE
Laboratoire d'accueil :
Laboratoire Bordelais de Recherche en Informatique (LaBRI), CNRS, IPB-ENSEIRB-MATMECA, Bordeaux I, Bordeaux Segalen, Talence (France)
Laboratoires associés :
Laboratoire Parole et Langage (LPL), CNRS : UMR 7309, Aix-Marseille Université, Aix-en-provence (France)
Laboratoire d'Informatique, de Robotique et de Microélectronique de Montpellier (LIRMM), CNRS, UM2, Montpellier (France)
Institut de Mathématiques de Luminy (IML), CNRS FRE3529, Aix-Marseille Université, Marseille (France)