The mental lexicon formally stores semantic, phonological, syntactic, and pragmatic properties of words. Words are organized by morphological family (run, running, runner), semantic field (emotions: angry, sad, happy), and phonological similarity (rhyming words). Psycholinguistic evidence reveals that lexical entries contain subcategorization frames, argument structures, and selectional restrictions.
From your study of lexical semantics, you understand that word meaning is richly structured — words exist in semantic fields, relate through hyponymy and antonymy, and have componential meanings that distinguish, say, *kill* from *cause to die*. But a word is more than its meaning. The mental lexicon is not a dictionary of definitions; it is a structured database where each entry stores multiple types of information simultaneously: what the word means, how it sounds, how it is spelled, what grammatical category it belongs to, and what syntactic environments it fits into. The formal structure of this database is what we're examining here.
Consider what you know about the word *persuade*. You know its meaning — causing someone to believe something through argument. But your lexical entry also encodes its phonological form (/pɚˈsweɪd/), its grammatical category (verb), and crucially its subcategorization frame: *persuade* requires a direct object (a person) and typically a complement clause or infinitive — *I persuaded her that X* or *I persuaded her to do X*. This subcategorization frame is part of the lexical entry, not learned anew each time the word is encountered. It's why "She persuaded enthusiastically" sounds wrong even though *persuade* is a verb and *enthusiastically* is a legitimate adverb — the frame specifies what complements the verb requires.
Words are also cross-indexed by multiple organizing principles simultaneously. Morphological family links *run*, *running*, *runner*, *ran*, and *outrun* — accessing one can prime the others. Semantic field clusters words by conceptual domain: the emotion field contains *angry*, *sad*, *happy*, *fearful*, *disgusted*, each related to the others through shared features and contrasts you studied in lexical semantics. Phonological similarity creates a separate organizational layer: *cat*, *bat*, *hat*, *sat* are linked by rhyme, which is why tip-of-the-tongue states often produce phonologically similar words rather than semantically similar ones. Psycholinguistic experiments — particularly priming studies, where seeing one word speeds recognition of related words — reveal all three organizational dimensions operating in parallel.
The formal concepts of argument structure and selectional restrictions extend the subcategorization frame into semantic territory. A verb like *devour* takes an agent subject and a patient object; but selectional restrictions further specify that the patient must be something edible (*She devoured the meal*; *?She devoured the idea* is odd, though acceptable metaphorically). These restrictions are not syntactic rules — they are semantic properties encoded in the lexical entry. When they're violated, the result isn't ungrammatical but semantically anomalous, as in Chomsky's famous "colorless green ideas sleep furiously" — syntactically impeccable, semantically incoherent. Understanding the mental lexicon as a multi-dimensional formal structure helps explain why language processing is so fast and accurate: the parser doesn't just look up a word's meaning, it immediately retrieves the full package of phonological, syntactic, and semantic constraints that govern how the word will combine with everything around it.