A basic psychological question about the lexicon that has not received quite enough attention in the theoretical literature is: What words and uses of words must be stored in long-term memory, and which ones can be computed online in short-term memory? This question leads to a distinction among two kinds of regularities that have both been called 'lexical rules': productive rules, for which one always knows, given an appropriate input, what the output will be (e.g. regular inflection, diminutives, and expletive infixation); and semiproductive rules, for which one has to know whether the output is an actual word and what its peculiarities might be (e.g. much derivational morphology and irregular inflection). I will argue that the productive cases warrant setting up independent lexical entries for the productive affixes, within-word combination being accomplished online by rules of the same character as regular phrasal rules; while the instances of semiproductive regularities must be listed individually and related by inheritance hierarchies.

I then will look at idioms, and show that they form a continuum with constructional idioms such as the resultative, way-construction, and time-away construction. The latter have often been treated as argument-structure-changing lexical rules; but in fact they fall in with productive rather than semiproductive rules, so they should not be encoded as alternative lexical meanings for verbs. Rather they should be viewed as online construction of argument structure, with the verb's argument structure as one input -- essentially the Construction Grammar approach. The constructions turn out to be written in the same format as more stereotypical lexical items, and like lexical items they turn out to be related in terms of inheritance hierarchies. In turn this leads to questions about the status of ordinary rules of phrase structure: Are they lexical items too?