| |
Abstract:
When evaluating empirical evidence on RC attachment ambiguity
resolution it is not unusual to oppose "Referentially based models"
(i.e. Construal (Frazier and Clifton, 1995): referential factors,
specifically relativized relevance, predict N1 attachment, but in
some languages, such as English, the existence of an alternative
structure, the Anglosaxon genitive, together with Gricean factors
("be unambiguous") overrides this tendency, resulting in N2
attachment) against "Statistical based models" (i.e the Tuning
Hypothesis: RC attachment preferences are determined by previous
linguistic experience, instances of successfully disambiguated
structures encountered in the past).
Dutch poses a serious problem for both of these theories: The
Gricean Theory predicts a tendency to attach low (since Dutch has
alternative genitive structures) but native Dutch speakers prefer
to attach high (Brysbaert and Mitchell, 1996). The Tuning
hypothesis, on the other hand, predicts that high attachment should
predominate in Dutch corpora (since that would be the linguistic
experience that "creates" the high attachment tendency in this
language) but Dutch corpora (Mitchell et al., in press) studied so
far have failed to show such a tendency.
However, if we take into account frequency and naturalness of
alternative structures, this finding would not be such a problem
for the Gricean Hypothesis: only when an alternative structure is
equally natural - or unnatural - in different languages should we
expect it to affect those languages evenly. To test this
possibility, we run a production questionnaire in English (two
possible genitives, low attachment), Spanish (one genitive, high
attachment) and German (which has, like Dutch, two possible
genitives, but attaches high). We used descriptions based on
sentences used in Cuetos and Mitchell 1988 that could be compatible
with either a high or a low attachment interpretation, and asked
participants to complete a sentence. For instance:
(a) There is an actress, and she has a servant. The servant was
on the balcony, and somebody shot him.
(b) There is an actress, and she has a servant. Somebody shot the
servant, and the actress was on the balcony.
Somebody
shot..................................................that was on
the balcony
This procedure has two obvious advantages over corpus studies:
first, it guaranties that participants had the same demand on
producing "high" and "low" attachment sentences - and no other
unexpected discourse factors, or just chance, were responsible for
the pattern found in corpus studies. Second, it allowed us to study
the effective use of possible alternative structures in the three
languages.
The pattern we obtained resembled that found in corpus studies
when attending just to the distribution of Norman genitives: a very
slight tendency towards high attachment in Spanish, and low
attachment for German and English.
However, when the data was considered as a whole, German and
Spanish behaved very similarly, and very differently than English:
both groups of participants preferred Norman genitives to express
high and low attachment, whereas English speakers used Anglosaxon
genitives for high attachment situations and Norman genitives for
low attachment situations. We claim that our data not only gives
support to a referentially based explanation of crosslinguistic
differences in RC attachment, but suggests that Gricean factors can
explain at least partly the frequency distribution on the basis of
the Tuning Hypothesis as well as the mismatch between Dutch corpus
and comprehension data.
Brysbaert, M., & Mitchell, D.C. (1996). Modifier attachment
in sentence parsing: Evidence from Dutch. Quarterly Journal of
Experimental Psychology, 49A, 664-695.
|
