Challenges in the semi-automatic reversion of a Latvian-English dictionary

The electronic version of the Latvian-English dictionary has been significantly supplemented over the last year with new linguistic material from corpora, databases and other sources. In contrast, the English-Latvian dictionary can be considered outdated as its electronic version was updated 10 years ago. This motivated us to create a semi-automatic process for reversion of the Latvian-English dictionary in order to supplement the English-Latvian dictionary with missing entries. Some of the major challenges for automatic reversion were as follows: grouping translations by part of speech, deciding to which entry the example should be attached, and ordering translations with similar meaning. By using automatic scripts it was possible to create reversed entries of quite good quality within a short time. Three groups of entries were prepared for manual post-editing: new entries with a single translation, new entries with a more complex structure, and existing entries with additional new content. The tasks for post-editing are: to check the suitability of the chosen headword, part of speech and translation order, to group the translations having the same meaning, and to move examples after appropriate translations.