|
Integrating domains and modalities: a construction-based architecture
for language processing| old_uid | 15741 |
|---|
| title | Integrating domains and modalities: a construction-based architecture
for language processing |
|---|
| start_date | 2015/06/05 |
|---|
| schedule | 11h15-11h45 |
|---|
| online | no |
|---|
| location_info | salle B011 |
|---|
| details | Cycle Cognition |
|---|
| summary | Modularity is not anymore an option for language processing. However,
it remains difficult to explain how the different linguistic domains
interact (not to say the different modalities when taking into account
natural contexts). The problem is twofold: we need first a framework
making it possible to describe, formalize and implement such an
interaction and second, we have to explain and (when possible)
evaluate the relative contribution of the different sources of
linguistic information. This question is a new challenge addressed by
modern linguistic theories, typically those starting from data (usage-
based theories, such as Construction Grammars). These theories explain
and represent precisely the domain interaction phenomenon (in
particular the syntax/semantic one). However, such mechanism only
works for local linguistic phenomena (called constructions) and seeing
how a description based on set of constructions can form an entire
system is still problematic.
I will present in this talk a new architecture for language
processing, taking into account the fact that sources of information
are multiple, not necessarily coherent and well synchronized. This
approach relies on a two-level organization in which we first identify
the local phenomena (constructions), defined as a "sites" with a
certain density of information, resulting from the convergence/
interaction of different sources of information. We will show how it
is possible to measure the convergence level, thanks to an activation
function. The second stage of this architecture consists in
identifying the relations between constructions. In a cognitive
perspective, this approach comes to a chunk-based model in which
chunks are defined in terms of constructions. Several examples will be
presented (among which the case of idiom processing), showing the
effect of constructions in the global processing mechanism. |
|---|
| responsibles | Bigi |
|---|
| |
|