On 20.4.2007 Dirk Elzinga wrote:
 > Two things occurred to me in watching this exchange.
 > First, stevo seems to be assuming a "dual-route" model of
 > morphological processing, and second, by his criterion of
 > complete predictability, all of Henrik's examples count as
 > processes which create new words.
 >
 > First, since the term 'dual-route' may not be familiar to
 > all, let me explain briefly. One definition of the lexicon
 > assumes that it contains only the information that is
 > unpredictable (this isn't the only definition, but I'll
 > use it since it seems to advantage stevo's claims);
 > anything that is predictable is provided by rule. Regular
 > word formation, by definition, is predictable, so words
 > formed in this way will not be included in the lexicon.
 > However, if the word formation process is unpredictable,
 > the result of applying it to a base must be listed
 > separately in the lexicon, thus creating a new word.
 > Hence, 'dual-route': one route is the regular application
 > of a rule of grammar creating a word "on the fly", the
 > other route is accessing a word already existing in
 > lexical memory.
 >
 > The dual-route model is not universally accepted, however.
 > There are many linguists (including me) who believe that
 > all morphological processing (and probably all
 > phonological processing) is done on the basis of whole
 > words used as analogical models or exemplars for novel
 > forms (thus the "single-route" of lexical access). That
 > is, we know that '-er' is a suffix because we recognize it
 > on a large number of words--not because it is provided by
 > a rule of grammar.
 >
 > Second, I can find examples of irregular or unpredictable
 > usages of each of the suffixes Henrik mentions. Just to
 > give one example: the suffix '-er' is used to form
 > agentive nouns from verbs; thus 'runner' is "one who
 > runs". But it can also be used to form nouns denoting
 > instruments; thus 'blender' is "an instrument with which
 > one blends". 65 years ago the word 'computer' was
 > understood as an agentive noun "one who computes", and
 > large companies which depended on numerical analysis hired
 > many people to perform numerical calculations. Now we
 > understand the word 'computer' as an instrumental noun. So
 > the result of adding this suffix to a stem is not entirely
 > predictable, and thus words formed by it must be listed
 > separately in the lexicon.
 >
 > It won't do to say "Yeah, but I mean the 'productive' use
 > of the suffix" without defining your criteria for
 > productivity. Even then, it's not an "either-or"
 > proposition; you will find word formation processes
 > arrayed on a continuum from highly productive to highly
 > unproductive. Dividing the continuum at *any* point will
 > be arbitrary.
 >
 > Dirk
 >

A quick googling showed that the term "dual-route" model of
morphological processing is mostly used in connexion with
learning to read. Can you point me to some sources on this
controversy. I must say that IME I use the single-route
model as a native speaker, but the dual-route model for
foreign languages :-) -- probably an artifact of having seen
derivational morphemes listed in grammars and the like.
Experience from trying to discuss morphology with my son
(now 9 y.o.) makes me come down on the side of the single-
route model as the natural one.  I wonder if and how
speakers of polysynthetical languages may differ, though.

/BP