|
Abstract:
Adjacent sentences in discourse are usually understood as
being temporally connected.
(1) Sheila had a party last Friday. Sam got drunk. (Partee,
1984)
The most salient interpretation of (1) is that Sam's getting
drunk happened during Sheila's party. This connection is a case of
temporal anaphora. This paper presents two self-paced reading
studies examining when temporal anaphora connections are made
on-line. The results suggest that rather than making such
connections immediately, the parser delays until it has encountered
a minimal event: the verb and the arguments obligatorily selected
by it.
Study 1 looked at whether temporal connections are calculated
before the end of the sentence. Subjects read brief paragraphs,
whose final sentence contained an adverbial disambiguating the
sentence's connection to preceding context. (See sample item (2).)
The adverb either connected the sentence to the preceding one
(2a,c) or indicated that it was not connected to it (2b,d). It
appeared in either preposed (2c,d) or sentence-final (2a,b)
position. Corrected reading times for the segment containing the
adverbial were higher for the disconnected adverbial (2b) than for
the connected one (2a) in sentence-final position, 152 ms vs. 107
ms. This pattern was reversed in preposed position: the
disconnected adverbial (2d) was faster than the connected one (2c),
-226 ms vs. -161 ms. This pattern resulted in a near-significant
interaction, F
1
(1,43) = 3.3, p<.077; F
2
(1,15) = 2.733, p<.122. This result suggests that subjects
connected the critical sentence to the preceding one early, before
encountering the disambiguating sentence-final adverbial.
Study 2 looked at exactly when in the sentence the parser
consulted preceding context. Subjects read a simple past-tense
sentence preceded by either a past-tense context or an incompatible
future-tense one. (See sample item (3).) Corrected reading times
for the whole critical sentence were higher in the future context
than in the past, 80 ms vs. -81 ms (F
1
(1,43) = 6.73, p<.012; F
2
(1,11) = 9.24, p<.013). However, this difference only became
reliable at the second and third segments of the sentence, the
point at which the parser had encountered the verb and the
arguments obligatorily selected by it. (Reading times for segments
2 and 3 combined, -102 ms vs. -26 ms, F
1
(1,43) = 4.04, p<.051; F
2
(1,11) = 4.51, p<.059). The difference was not reliable at the
first segment, the position of the past-tense marker (F
1
,F
2
< 1), or even at the first and second segments combined (F
1
,F
2
< 2).
Together, these results suggest that the parser is cued to
calculate temporal anaphora relations by the events being
described, not by tense markers themselves (Trueswell &
Tanenhaus, 1991) or by rhetorical relations between whole sentences
(Lascarides & Asher, 1993).
Sample Items
(2) Last week the smartest girl in school competed in the state
spelling bee.
a. She drew | a word | with sixteen letters | in the first round.
b. She drew | a word | with sixteen letters | in the last spelling
bee.
c. In the first round | she drew | a word | with sixteen letters.
d. In the last spelling bee | she drew | a word | with sixteen
letters.
(3) a. Last weekend a notorious jewel thief broke into the
Smithsonian Natural History Museum after hours.
He opened several cases in the moon rock collection.
b. Next weekend a notorious jewel thief will break into the
Smithsonian Natural History Museum after hours.
He will open several cases in the moon rock collection.
He took | the Hope Diamond | and an emerald | during his
escape.
|
