mappingtranslation

Mapping translation is a method for rendering text from one language to another by constructing and applying mappings between linguistic units. Core mappings link individual words, phrases, or semantic roles across languages, often stored in bilingual lexicons, phrase tables, or semantic frames. The approach contrasts with purely generative or statistical methods by emphasizing explicit correspondences that guide or constrain the translation output.

In practice, mapping translation underlies several MT paradigms. Dictionary-based translation relies on word-to-word mappings and glossaries to generate translations, optionally augmented with rules to handle morphology. Transfer-based and rule-based systems use linguistic structures to map syntactic and semantic representations from source to target language. Hybrid systems combine mapping dictionaries with statistical signals to improve fluency and accuracy. In contemporary settings, neural methods may incorporate explicit mappings as features or constraints, or learn implicit mappings from data.

Applications include localization, terminology management, and translation memory workflows used in publishing, software, and regulatory domains. Mapping translation also informs cross-lingual information retrieval, where queries in one language are mapped to indexed terms in another.

Challenges include polysemy, context-dependence, idioms and culture-specific expressions, and data sparsity for less-commonly paired languages. Mapping quality depends on the granularity of units (word, phrase, or concept), alignment quality, and consistency of the mapping resources. Evaluation typically uses standard MT metrics or human judgments to assess adequacy and fluency.

See also: bilingual lexicon, word alignment, translation memory, cross-lingual transfer, phrase-based translation.