Word embedding-based task adaptation from English to Hungarian
In commercial Natural Language Processing (NLP) solutions, we frequently face the problem, that a particular NLP application has to work on several languages. Usually the solution is first developed on a single language – the source language – then it is adapted to the other languages – the target l...
Elmentve itt :
Szerzők: | |
---|---|
Testületi szerző: | |
Dokumentumtípus: | Könyv része |
Megjelent: |
2017
|
Sorozat: | Magyar Számítógépes Nyelvészeti Konferencia
13 |
Kulcsszavak: | Nyelvészet - számítógép alkalmazása |
Online Access: | http://acta.bibl.u-szeged.hu/59017 |
Tartalmi kivonat: | In commercial Natural Language Processing (NLP) solutions, we frequently face the problem, that a particular NLP application has to work on several languages. Usually the solution is first developed on a single language – the source language – then it is adapted to the other languages – the target languages . In this paper, we introduce experimental results on English to Hungarian adaptation of document classification tasks. In our setting, only an English training dataset is available and our aim is to get a classifier which works on Hungarian documents. We experimented comparatively with two different approaches for word embedding-based language adaptation methods and evaluated them along with monolingual methods in a sentiment classification and a topic classification dataset. |
---|---|
Terjedelem/Fizikai jellemzők: | 287-295 |
ISBN: | 978-963-306-518-1 |