|
Abstract:
Smyth (1998) reports online anomaly detection experiments on
sentences with 3-argument verbs, which have two potential gap
locations:
1. This is the jar that is easy to put t in the can.
2. This is the jar that is easy to put the can in t.
3. This is the jar that the can is easy to put t in t.
Comparisons of Types 1 and 2 showed a filled gap effect:
although reading times did not differ, there were significantly
more comprehension errors on Type 2, which has a legal filler
("the can") in the first possible gap position. Type 3 are the
famous "sonata/violin" sentences with nested dependencies. Using
Gibson's (1997) Syntactic Prediction Locality Theory, I predicted
that they should be very complex, despite the unmarked nature of
nesting and the fact that they conform to the Most Recent Filler
strategy. These sentences took 50% longer to read/judge and
resulted in significantly more comprehension errors.
In the poster I explore the possibility that Type 3, with two
gaps both signalled at the final word "in," are near the limits
of human processing capacity. The research strategy is to present
participants with ungrammatical versions of Type 3 sentences for
timed "makes sense" judgments. Ungrammatical gap filling in the
postverbal position should trigger a filled gap effect. Since the
processor must undo its premature gap filling, I predict that
this makes it difficult to continue to keep the initial filler
active. Participants should fail to detect the ungrammaticality
because they will interpret the final phrase as "put coins in the
can":
4. This is the jar that the can is easy to put *COINS in
t.
However, ungrammatical gap filling should be easier to detect
when there is in fact a gap in the expected location after the
verb. That gap will be correctly filled, thereby discharging part
of the processing load, and the ungrammatically filled final gap
(alternatively, the fact that the first filler has not found a
gap) should be more readily detected:
5. This is the jar that the can is easy to put in *THE
BOX.
In accordance with these predictions, sentences like 4
resulted in far more false acceptances than sentences like 5,
whose ungrammaticality was nearly always detected. I will also
discuss the pattern of response times for both correct and
incorrect judgments, and the implications for models of the
complexity of gap filling.
|
