|
Linguistic Typology and Language Technology: A Match Made in Statistical Parsing| old_uid | 7396 |
|---|
| title | Linguistic Typology and Language Technology: A Match Made in Statistical Parsing |
|---|
| start_date | 2009/10/05 |
|---|
| schedule | 14h30-16h |
|---|
| online | no |
|---|
| summary | Treebank-induced statistical parsers demonstrate excellent performance for parsing English, but it is has been repeatedly shown that adapting them to parsing other languages
often fails to yield comparable results. In this talk I consider the parsing problem from a typological perspective, arguing that the choice and arrangement of model parameters should
reflect dimensions of typological variation. I fi rstly demonstrate that head-centered parsing methods are inherently suboptimal for parsing languages that are less-con gurational
than English as they assume simple, transparent, mapping between grammatical functions and con gurational positions.
I then show that the Relational-Realizational approach, a proposal which explicitly parametrizes two dimensions of typological variation; basic word-order and morphological synthesis, more easily adapts to non-con gurational structures in the language.
I finally present performance gains of the Relational-Realizational model over Head-Driven varieties on parsing the Semitic language Modern Hebrew, and suggest the resulting empirical distributions as means to quantify typological classi cation. |
|---|
| responsibles | Information non disponible, Crabbé |
|---|
| |
|