wordtoword

wordtoword is a modular software framework for word-to-word translation and lexical mapping between languages. It provides a centralized lexical database, a lightweight API, and tooling for researchers and developers. The project emphasizes word-level translation rather than sentence-level mechanics.

It maintains a bilingual lexicon with entries sized with senses, part-of-speech tags, pronunciation, and usage examples. Each entry can be linked to multiple translations with context tags and frequency data. The API offers endpoints such as lookup, translate (word to word), and batch translate; data can be exchanged in JSON or TSV; supports numerous languages via plug-ins. The project is open-source under a permissive license.

The architecture consists of core components including a lexical database, a disambiguation module to select senses based on minimal context, a back-end translator interface to plug into external MT engines, and an API layer. Ingest pipelines import data from public bilingual dictionaries and parallel corpora, with quality-control checks.

Use cases: language learning apps for vocabulary drills, translation memory tools for translators, linguistic research into lexical semantics and polysemy. It is designed to complement statistical and neural MT systems by providing high-quality word-level mappings.

Limitations and reception: word-to-word translation often fails to capture idioms or syntactic structure; wordtoword is not intended to replace full-sentence translation. Critics note data coverage and sense disambiguation challenges. Current development is community-driven with active maintenance in recent years.

See also: machine translation, bilingual lexicon, word alignment, lexical database, language learning tools.