Subject Preference in the Processing of Relative Clauses in Chinese

Relative clauses are processed differently depending on whether they involve subject or object extractions.
In head-initial languages such as English, German, and Dutch, where relative clauses follow the
antecedent head nouns, it was found that subject relative clauses such as (1) are easier to comprehend than
object relative clauses such as (2). This preference was confirmed using various tasks, including
self-paced reading tasks (King & Just 1991; Gibson et al. 2005) and eye-movement monitoring reading
tasks (Traxler et al. 2002).
     (1) The girl who [GAP] bumped into me yesterday.
     (2) The girl who I bumped into [GAP] yesterday.
Such an effect could be explained both by a locality-based account (Crain & Fodor 1985; Frazier &
Clifton 1989; Gibson, 1998) and a structure-based account (Lin et al. 2005; Miyamoto & Nakamura 2003)
since the gaps in subject relative clauses are located both closer to the head nouns and higher up in the
structure for probing. It is, however, crucial to distinguish a locality-based and a structure-based theory
because the former relies on functional factors such as cognitive resources, integration costs, and working
memory, while the latter emphasizes the importance of structural positions and implies a theory of
goal-probing based on syntactic knowledge. Head-final languages, where the relative clauses precede the
head nouns, provide a chance to tease the two theories apart, since they have distinct predictions, as
summarized in (3).
     (3) Filler-gap distances in prenominal relative clauses according to the 2 theories:
                             Locality-based theory        Structure-based theory
           Subject RC        Farther                      Closer (thus easier)
           Object RC         Closer (thus easier)         Farther

The subject preference effect found in English, however, remains controversial in head-final languages
(such as Japanese and Chinese). Miyamoto and Nakamura (2003) reported that subject relative clauses are
preferred in Japanese, consistent with the subject preference in head-initial languages and thus supporting
a structure-based theory. However, Hsiao and Gibson (2003) reported faster reading times for Chinese
object relative clauses, supporting a locality-based theory.
      In this study, we revisit the processing of Chinese relative clauses by carefully re-examining Hsiao
and Gibson's experimental materials and conducting a self-paced reading experiment of our own. We
found the selection of verbs in their experiment to be problematic in that more than half of their verbs can
take sentential and verbal complements, which increased the possibility of garden pathing. In addition, all
their relative clauses appeared only at the subject position. Their results showed no significant difference
at the head noun section, only at the relative clause section. By selecting verbs that take only nominal
objects and contrasting between relative clauses in both subject and object positions, we found subject
relative clauses to be read significantly faster than the object counterparts in terms of overall reading
latencies when the relative clauses are in the subject position (F1(1, 47) = 4.89, p < .05; F2(1, 143) = 2.07,
p = .09) and when they are in the object position (F1(1, 47) = 11.89, p < .01; F2(1, 143) = 10.15, p < .01).
The effect at the object position was more robust. By-region analyses also showed subject preference in
the head noun and matrix verb regions, and a stronger effect (i.e. larger latency differences) for relative
clauses at the object than the subject position (which may have been why Hsiao and Gibson did not get the
effect in their study). Our experimental results thus suggest that there is a universal preference for gaps at
the subject position of relative clauses, which supports the structure-based theory rather than the
locality-based theory. The results will be discussed in terms of recent minimalist probe-goal model and
implications for goal-probing in sentence processing.