fasttext-subword-character-ngram-embeddings
IN premise — summaries/2026/08/24/wiki-Word_embedding-chunk-2.md
Created 2026-08-24T17:11:28+00:00
fastText (Bojanowski et al., 2017) represents words via character n-grams in addition to whole-word vectors, enabling subword-level compositionality and representations for unseen or rare words in morphologically rich languages
Summary
Instead of treating every word as an indivisible unit, this approach also encodes the smaller letter sequences that make up each word, so the system can still produce a meaningful representation for a word it has never encountered. This matters most for languages where words are built from many morphological pieces, because it lets the system generalize from known parts to unknown wholes rather than hitting a dead end at every rare or novel word.