|
Harvesting Multilingual Information for Frame Semantics| old_uid | 7562 |
|---|
| title | Harvesting Multilingual Information for Frame Semantics |
|---|
| start_date | 2009/11/09 |
|---|
| schedule | 14h30-16h30 |
|---|
| online | no |
|---|
| summary | Frame semantics has been recently suggested as a relevant information in several language processing tasks, such as question answering or ontology learning. However, several limitations have been noticed when resources related to Framenet are applied to multilingual tasks. Most of them are strictly oriented to English, and, more importantly, even the current coverage over English is limited in view of its uses on real applications. Research carried out in the AI labs of the University of Roma, Tor Vergata has focused on the use of distributional models of frame semantics able to provide flexible solutions to the above problems.
The seminar will survey research on three different tasks: predicate induction, semantic transfer of role information across language pairs and automatic argument classification. In the first task, the role of vector space models for inducing predicates from large scale corpora will be discussed as a way to augment the coverage for English phenomena as well as to develop systems of lexical units in another language, e.g. Italian. In the second, methods devised to semi-automatically acquire large sets of annotated sentences in Italian through semantic transfer methods over a bilingual English-Italian corpus will be presented. Finally, more recent results on a weakly supervised distributional model for argument classification in semantic role labeling, whose effectiveness is close to fully supervised models will be presented. |
|---|
| responsibles | Information non disponible, Crabbé |
|---|
| |
|