Ron Smyth
University of Toronto
According to a view first advanced by Hankamer and Sag (1979), surface anaphors access representations which encode syntactic distinctions (subject/object; word order) while deep anaphors access representations which encode at most propositional semantic information.
I will present an experiment which addresses this issue with respect to participial adjunct control (Goodluck 1978). Previous research has focused on cases where PRO is coreferential with a c-commanding subject antecedent:
(1) Holly slapped Oksana while PRO leaving the party. (PRO = Holly)
However, other prepositions license thematic control:
(2) Holly slapped Oksana for PRO leaving the party. (PRO = Oksana)
I will show that cases like (2) are handled by an analysis under which "for" imposes semantic ("deep") Recipient Control, regardless of the Recipient's syntactic position (cf. 4 below).
In previous studies I have tried to test the deep/surface hypothesis by comparing adjunct reading times for structural vs. thematic control. Passive initial clauses were used to avoid confounds with syntactic position of the antecedent:
(3) Oksana was slapped by Holly while PRO leaving the party.
(4) Oksana was slapped by Holly for PRO leaving the party.
However, the findings were contaminated by differences between temporal and causal semantics, and complicated by the fact that 'before', 'after' and 'while' caused significant variation in reading times, preventing direct comparisons between deep and surface processes.
For this study I have devised a method for removing the confound. Reading times for items like (3) and (4) were first divided by reading time for the corresponding pronoun conditions:
(5) Oksana was slapped by Harry / [...] while / she left the party.
(6) Oksana was slapped by Harry / [...] because / she left the party.
Subject interpretation was guaranteed by gender agreement. To avoid reading time differences for 'for' vs. 'because', the prepositions were presented on a separate line in the phrase-by-phrase reading task.
Each item was either Short or Long, depending on whether an additional phrase/line (e.g., "one fateful evening") was inserted between PRO and its antecedent. Under the deep/surface hypothesis, reading times for surface control should be more affected by length as the surface representation of the initial clause is lost from working memory.
The results will be discussed within the context of previous work on verb phrase ellipsis, which have suggested a more complicated picture of the deep/surface distinction (e.g., Murphy, 1985; Belanger & Smyth, 2001).