Tips for working on bilingual dictionaries

This page is to catalogue tips for working on bilingual dictionaries for use in machine translation.

If you have two target translations to choose between (and don't have a lexical selection module), you should typically choose the one that fits the most circumstances, or will be correct in the most circumstances (letting the other one get a directional restriction).

If you can't translate it, don't.

Example:

<e><p><l>laufabrauð<s n="n"/><s n="nt"/></l><r>traditional<b/>Icelandic<b/>deep-fried<b/>patterned<b/>Christmas<b/>wafer<s n="n"/></r></p></e>

There is literally nowhere that that description would fit fluently into a sentence. First try and find a more literal translation in the target language, for example in this case "laufa" + "brauð" = "leaf" + "bread". Check in a search engine, or one of the image search engines to make sure it comes up with the same kinds of results, and if it does use that. Just because it doesn't appear in a dictionary does not make it a bad translation.

As a last resort, just use the source language word in the target language, consider the following translations:

• Börnin bökuðu laufabrauð
• The children baked leafbread.
• The children baked laufabrauð.
• The children baked traditional Icelandic deep-fried patterned Christmas wafer.

People who are going to be reading machine translation have the nouse to go and search for a word they are unfamiliar with. People who are post-editting machine translation are not going to want to remove six words, to re-add one.

See also

• Bilingual dictionary