Coreference resolution using parallel corpora

Speaker:
Michal Novák
Abstract:
The talk will focus on two approaches to employing parallel corpora in automatic coreference resolution. First, projection-based coreference resolution uses links projected from the source language to train a resolver in the target language, for which no manual annotation of coreference exists. Second, bilingually-informed coreference resolution is able to exploit features from the source language as additional information to find links in the target language. Unlike the first approach, manual annotation in the target language is available here. The approaches will be demonstrated on projection from English to German and Russian, and resolution on English-Czech texts. The work to be presented is in progress.
Length:
01:24:25
Date:
27/02/2017
views: 1108

Images:
Attachments: (video, slides, etc.)
116 MB
755 downloads
1.1 GB
1109 downloads
144 MB
807 downloads
201 MB
755 downloads
410 MB
700 downloads