Any response to a dual-mechanism approach?

[Steve Pinker]

Liz Bates's examples lead me to think that I wasn't sufficiently clear
in the previous posting about the relationship between the lexicon and
grammar. There are *two* distinctions here: (a) memorized versus
composed (versus connectionist-like analogy), and (b) level of
representation: root versus complex word versus phrase. The
30-year-old idea that Bates refers to is that these used to be
collapsed into a single distinction: words are stored, phrases are
composed. Indeed, that must be rejected, because there are structures of all
kinds that must be stored. For example, the lexicon might contain
entries such as these (lots of notations possible):

die:   VP           angry:     VP        devour:   VP       duck:  N
      / \                     / \                  /\              |
     V  NP                   V   PP               V  NP           duck
     |   /\                  |   /\               |
  kick the bucket          mad at  NP          devour

This embraces the continuum of structures from roots to phrases (or
constructions) that Bates referred to. 	BUT: That is independent of
the second sense of the grammar-lexicon distinction, namely lookup
versus computation (composition, unification, etc.); that is, the
difference between the two psychological processes below (again,
simplified):

(1)

die:   VP        
      / \        
     V  NP           -->      "kick the bucket"
     |   /\      
  kick the bucket

(2)

kick:   VP           duck:  N
       / \      +           |       --> "kick the duck"
      V  NP                duck 
      | 
   kick 

This distinction is largely indpendent of the first one. Theoretical
linguists don't talk about the composition process, because they treat
it as a black box, to be studied by the computational linguists and
psycholinguists. But the theory of language as a whole still needs it.

--Steve Pinker