This archive report was first published on 13 October 2019.
Published on October 13, 2019, researchers at Facebook are working on a new approach to machine translation that uses mathematical representations of words to improve translation accuracy.
Currently, machine translation relies heavily on dictionaries and large databases of identical texts in both languages. However, for many language pairs, there just aren't enough parallel texts available.
Facebook's researchers have developed a system that creates a mathematical representation for words, known as a vector, in a space of several hundred dimensions. Words that have close associations in the spoken language also find themselves close to each other in this vector space.
According to Guillaume Lample, one of the system's designers, "For example, if you take the words 'cat' and 'dog', semantically, they are words that describe a similar thing, so they will be extremely close together physically" in the vector space.
These language maps can then be linked to one another using algorithms, which can eventually become more refined, allowing entire phrases to be matched without too many errors.
Results are already promising, with the word vector system outperforming Facebook's traditional machine translation system for the language pair of English-Urdu.
However, experts at France's CNRS national scientific centre note that the approach may not result in perfect translations, but could still produce useful results.
