Introducing aWord Alignment Dissimilarity Indicator: Alignment Links as Conceptualizations of a Focused Bilingual Lexicon

Devin Gilbert

Michael Carl

Ladda ner artikel

Ingår i: Proceedings of the First Workshop on Modelling Translation - Translatology in the Digital Age

Linköping Electronic Conference Proceedings 179:8, s. 74-81

NEALT Proceedings Series 46:8, p. 74-81

Visa mer +

Publicerad: 2021-05-24

ISBN: 978-91-7929-613-1

ISSN: 1650-3686 (tryckt), 1650-3740 (online)


Starting from the assumption that different word alignments of translations represent differing conceptualizations of crosslingual equivalence, we assess the variation of six different alignment methods for English-to-Spanish translated and postedited texts. We develop a word alignment dissimilarity indicator (WADI) and compare it to traditional segment-based alignment error rate (AER). We average the WADI scores over the possible 15 different pairings of the six alignment methods for each source token and correlate the averaged WADI scores with translation process and product measures, including production duration, number of insertions, and word translation entropy. Results reveal modest correlations between WADI and production duration and insertions, as well as a moderate correlation between WADI and word translation entropy. This shows that differences in alignment decisions reflect on variation in translation decisions and demonstrates that aggregate WADI score could be used as a word-level feature to estimate post-editing difficulty.


Inga nyckelord är tillgängliga


Inga referenser tillgängliga

Citeringar i Crossref