Using Simple Word-Alignment Measures to Study Discourse Particles

Autor:
Spenader, Jennifer
Aufsatztitel:

Using Simple Word-Alignment Measures to Study Discourse Particles


Jahrgang:
28
Heft:
1 (2004)
Seiten:
9-19
Abstract:

This work demonstrates that simple word-alignment measures from statistical machine translation can help in studying the contribution of discourse particles even though these lexical items often do not have recognizable translation equivalents. The aligned Europarl parallel corpora in Swedish and English was mined to extract examples of the use of the Swedish discourse particle ju. χ2 tests were used to determine which unigrams and bigrams were signifi cantly more frequent in the English translations of aligned areas where the particle appeared than in the rest of the corpus. Further, φ2 tests were done on relevant signifi cant co-cccuring unigrams and bigrams to rank them in order of the strength of their relationship with the particle. These tests therefore offer a method by which intuitions obtained through introspective study of a small number of examples can be automatically corroborated with a large number of examples.

Zurück