lexicondriven

Lexicondriven is a term used in discussions of language technology to describe approaches in which a system's behavior is guided by a lexicon—the structured collection of words, lemmas, and their properties. A lexicon typically records parts of speech, sense inventories, subcategorization frames, morphological information, and sometimes multiword expressions and semantic relations. In a lexicondriven design, lexical knowledge constrains or informs tasks such as parsing, interpretation, disambiguation, and translation, often in combination with other resources such as grammars or statistical models.

In natural language processing, lexicon-driven methods contrast with purely data-driven or end-to-end approaches. They are common in lexicalized parsing, rule-based or hybrid machine translation, and word sense disambiguation where sense inventories (e.g., WordNet, FrameNet) guide interpretation. They can improve transparency, controllability, and performance on rare or domain-specific terms, but suffer from limited coverage, ongoing maintenance, and the cost of building comprehensive lexicons.

In practice, lexicondriven systems are often hybrid, using lexicon data to augment statistical models, or to initialize parameters and provide constraints that steer learning. The term is more often encountered in academic discussions of linguistic annotation, computational lexicography, or rule-based NLP than in mainstream commercial systems, where statistical methods predominate but lexical resources remain important for lexical disambiguation and interpretability.