Chapter 10
Impact of word alignment on word translation entropy and other
metrics
A comparison of translation process research findings derived from
different word alignment methods
Many of the findings from studies using the Center for
Research and Innovation in Translation and Translation Technology (CRITT)
Translation Process Research Database (TPR-DB) framework rely on
word(s)-to-word(s) alignments of the source text and target text. However,
little research has been done on the impacts different alignment methods
have on these findings. This study compares two different manual word
alignment methods and four automatic word alignment methods on the basis of
one English-Spanish TPR-DB study that has been used extensively (the BML12
dataset). We replicate past findings from the BML12 dataset using these
different alignments in order to determine the impact of alignment, and we
present qualitative/quantitative analyses of the different word-alignment
methods.
Article outline
- 1.Introduction
- 2.Summary of past research
- 3.Procedure
- 3.1Manual alignment
- 3.2Automatic alignment
- 3.3Post-processing
- 4.Research questions 1 and 2: Qualitative analysis and replicating past studies
- 4.1Carl and Schaeffer
(2017)
- 4.2Toledo-Báez and Carl
(2020)
- 4.3Ogawa et al.
(2021)
- 5.Research question 3: Comparing measures across the alignment methods
- 6.Conclusions
- 6.1RQ1: Is new manual alignment more consistent?
- 6.2RQ2: Will alignment change results?
- 6.3RQ3: Alignment consistency
- 6.4Final remarks
-
Notes
-
References
-
Appendix