ABSTRACT
In Chinese, the compound reflexive ta-ziji (“him/her-self”) has the gender marking pronoun ta, hence presenting a good test case for interference effects from structurally illicit antecedents predicted by cue-based retrieval models. Using reading eye-tracking, we manipulated the gender of ta-ziji that (mis)matches that of matrix- and local-subject. Results showed no interference whatsoever when ta-ziji matched local subjects. Only when ta-ziji mismatched local subjects did we find an inhibitory interference on first fixation duration and gaze duration at the verb immediately preceding ta-ziji, but a facilitatory interference on gaze duration at ta-ziji. Furthermore, at ta-ziji, total reading times were longer for gender-mismatching local subjects than for gender-matching ones. These findings are partially predicted by the standard cue-based retrieval model, but are mostly consistent with the structure-favoring cue-based retrieval model, suggesting that the structural cue plays a dominant role in the antecedent retrieval process, with interference occurring only in highly constrained situations.
Acknowledgements
This research was supported by Shanghai Social Science Foundation (2019BYY005), National Natural Science Foundation of China (31470976) and National Basic Research Program of China (973 Program: 2015CB856400). The data analyses were supported by High-performance Computing Platform of Peking University.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 Existing work has shown that the two eyes are well-aligned most of the time, and fixation durations recorded from two eyes are highly correlated (r = 0.98) (Kliegl et al., Citation2006; Nuthmann & Kliegl, Citation2009).
2 We thank Brian Dillon for suggesting this.
3 One thing worth noting is in all the four conditions, the probability of choosing structurally licit antecedents (i.e., local subjects) never reached 100%. In other words, there were retrieval errors (i.e., misretrievals of matrix subjects) across the board, even in the Local-mismatch, Matrix-mismatch condition where the rate of misretrieval was 33%. As we speculate, this might be in part due to the nature of the forced-choice questionnaire and the use of equi-biased verbs in our design.
4 Note that our 95% CrI estimation of the interference effect on total reading time is similar to the 95% CrI estimate in Jäger et al. (Citation2020, p. 13, Table 4): the present paper (back transform to ms): [–40, 27], mean = –17 ms; Jager et al.: [–43, 16], mean = –12 ms.
5 We thank Brian Dillon for suggesting this.