[...] I proposed a mechanism, called L1, for the acquisition of lexical entries. [...] The child encodes the argument-taking predicate of the sentence, and cognitive abilities allow him or her to know how many arguments the predicate takes and which thematic role each argument bears. The child parses the incoming sentence using a combination of existing phrase structure rules and the tree-building procedures [...]. Once the sentence is parsed, the child applies the following procedure: L1. Add subcategorization information to the lexical entries of argument-taking predicates by examining the functions encoding its arguments in the tree and listing those functions in the PRED equation for that entry. If an argument is not encoded by any phrase in the tree, assign it the symbol ∅ in the PRED equation. [...] Thus once phrase structure rules are learned, mechanism L1, which I will also call "direct learning from positive evidence", allows the child to learn any association whatsoever between a verb's arguments and grammatical functions.
- Pinker (2004), a pag.243-245