Bitext Mining
A classic evaluation task for cross-lingual embeddings. Given sentence sets in each of two languages, a system must find, for every sentence, the semantically corresponding translation sentence in the other language. Nearest-neighbor retrieval based on embedding cosine similarity is typically used, with F1 or accuracy as the metric.
Timeline (last 90 days)
The researchers conducted bitext mining tasks between English and more than 200 other languages in cross-lingual scenarios.
Bitext mining is a classic evaluation task for cross-lingual embeddings: given sentence sets in each of two languages, the system must find, for each sentence, the semantically corresponding translated sentence in the other language.
The bitext mining task exposes cross-lingual misalignment problems better than monolingual sentence similarity tasks.
All Facts (3)
The researchers conducted bitext mining tasks between English and more than 200 other languages in cross-lingual scenarios.
50%UnverifiedBitext mining is a classic evaluation task for cross-lingual embeddings: given sentence sets in each of two languages, the system must find, for each sentence, the semantically corresponding translated sentence in the other language.
50%UnverifiedThe bitext mining task exposes cross-lingual misalignment problems better than monolingual sentence similarity tasks.
50%