Learnability arguments, i.e., claims that rules of language cannot derived
from experience, play an important role for the motivation of an innate
universal grammar. However, so far learnability arguments are drawn from
stipulations, not research, on what children do or do not hear (cf. Pullum
1996). A new generation of "high density developmental corpora" will allow
us to address this question with naturalistic data. I will present such
data from a corpus of a monolingual German child who was recorded for 320
hours over a two-year period, and address the issue of dual versus single
mechanism models of inflection with respect to his acquisition of the
German plural. The Dual Mechanism Model (DMM; Marcus et al. 1992; Clahsen
1999) distinguishes symbolic rules which are applied to a lexical roots,
from irregular forms which are stored holistically and subject to analogy
and frequency effects. The Single Mechanism Model (e.g., Nakisa & Hahn
2000, Goebel & Indefrey 2000) holds that both regular and irregular forms
are processed and acquired by associations and analogy. The German plural
served as a showcase example in this debate, because a low-frequent
allomorph - the -s plural - shows default (productive) characteristics.
Quantitive analyses of 140,000 utterances with 76000 nouns (10600 plurals
with 3.4% overgeneralization rate) show that contrary to claims by the DMM,
the -s plural is not overgeneralized overproportionally when computing the
absolute overgeneralization rate (Marcus et al. 1992), the proportion of
errors given the chance for error. By this measure, -en or -e are
overgeneralized slightly more frequently than -s. Absolute frequencies of
occurence reveal that the child has ample opportunity for generalizing the
conditions under which -s applies: it is a minority plural (6%), but plural
is a very high frequent phenomenon: The child produces a legal -s plural
every half hour, and of course hears more in the meantime.
Qualitative analyses do not confirm the default status of -s either. Plural
errors are mainly determined by the final sound (not rhyme) of the noun
root. Consequenty, plural marking in default conditions (foreign words,
mass nouns, names) follows analogy, not default. -s-errors are subject to
stricter phonological and metric constraints than -e and -en-errors.
Item-analyses show that they conform to the analogy space provided by the
default conditions, that is the dominant phonotactic patterns of legal -s
plurals. In sum, there is no evidence or necessity for special and
privileged knowlegde about default conditions. Rather, the (sub)rules of
the German plural system can be generalized on the basis of analogy, as
proposed by the SMM.