Séminaire de Recherche en Linguistique
Ce séminaire reçoit des conférenciers invités spécialisés dans différents domaines de la linguistique. Les membres du Département, les étudiants et les personnes externes intéressées sont tous cordialement invités. Description du séminaire ![Imprimer Print](design/print.png)
Titre | Experimental evidence for gradient effects in islands, and their consequences for theories of islands |
Conférencier | Sandra Villata (University of Connecticut) |
Date | mardi 17 décembre 2019 |
Heure | 12h15 |
Salle | L208 (Bâtiment Candolle) |
Description | It has long been recognized that acceptability judgments are gradient: sentence acceptability spans from full acceptability to full unacceptability passing through a wide range of intermediate values. In this talk, I offer evidence that gradience extends well beyond acceptability judgments – it also manifests in speakers’ ability to parse, and ultimately comprehend, a sentence. As a test case, I consider islands. There is converging empirical evidence coming from acceptability judgment studies showing that the island effect is reduced, but not eliminated, in weak islands, thus suggesting that the strong/weak island distinction might be gradient, rather than categorical. I show that the gradient patterns observed in acceptability judgments also manifest at the level of the mechanisms that are claimed to underlie the processing of islands. I do so by presenting a series of experiments employing a variation of the self-paced reading task, the Maze task. Results show that the parser does establish a dependency inside of islands in a significant proportion of the cases, contrary to the received wisdom beginning with Stowe 1986 that dependency formation is suspended inside of islands. These results raise significant challenges for any theory of islands and/or dependency processing that assumes islands are categorically impenetrable. I conclude by discussing the theoretical implications of these findings for theories of islands, and I provide some insights on how gradience can be generated in a computational model. |
Document(s) joint(s) |
- |