Experience and generalization in a connectionist model of Mandarin Chinese relative clause processing

Abstract

Sentences containing relative clauses are well known to be difficult to comprehend, and they have long been an arena in which to investigate the role of working memory in language comprehension. However, recent work has suggested that relative clause processing is better described by ambiguity resolution processes than by limits on extrinsic working memory. We investigated these alternative views with a Simple Recurrent Network (SRN) model of relative clause processing in Mandarin Chinese, which has a unique pattern of word order across main and relative clauses and which has yielded mixed results in human comprehension studies. To assess the model's ability to generalize from similar sentence structures, and to observe effects of ambiguity through the sentence, we trained the model on several different sentence types, based on a detailed corpus analysis of Mandarin relative clauses and simple sentences, coded to include patterns of noun animacy in the various structures. The model was evaluated on 16 different relative clause subtypes. Its performance corresponded well to human reading times, including effects previously attributed to working memory overflow. The model's performance across a wide variety of sentence types suggested that the seemingly inconsistent results in some prior empirical studies stemmed from failures to consider the full range of sentence types in empirical studies. Crucially, sentence difficulty for the model was not simply a reflection of sentence frequency in the training set; the model generalized from similar sentences and showed high error rates at points of ambiguity. The results suggest that SRNs are a powerful tool to examine the complicated constraint-satisfaction process of sentence comprehension, and that understanding comprehension of specific structures must include consideration of experiences with other similar structures in the language.

Effects of Early Cues on the Processing of Chinese Relative Clauses: Evidence for Experience‐Based Theories

“…An open question for future work has to do with animacy. The animacy configuration used in this study is uncommon because (i) the heads of our object‐extracted RCs are animate, whereas corpus findings show object‐extracted RCs typically have inanimate heads (Hsiao & MacDonald, ; Wu et al., ), and (ii) our passivized subject‐extracted RCs contain an inanimate instrument following BEI, whereas Chinese passive construction typically requires an animate agent. Thus, future work might employ more common animacy configurations to see whether the conclusion still holds.…”

The subject-relative advantage in Chinese: Evidence for expectation-based processing

“…In the materials used in previous experiments, several local ambiguities might have confounded the results (for a discussion of local ambiguities in Chinese RCs also see Lin & Bever, 2006;Lin & Bever, 2011;Qiao, Shen, & Forster, 2012;Vasishth et al, 2013). Indeed, Hsiao and MacDonald (2013) and Hsiao, Li, and MacDonald (2014) have argued that the results of previous studies on Chinese RCs can be largely explained by the local ambiguities in the stimuli. Thus, although the theoretically interesting aspect of Chinese relative clauses lies in the diametrically opposed predictions of the expectation-based account vs the memory-based accounts, it is vital to bring the local ambiguities under experimental control before we can investigate these opposing predictions.…”

When Structure Competes with Semantics: Reading Chinese Relative Clauses

“…Specifically, we expect a significantly shorter processing time for ORC at the second noun, relativizer de, and main clause verb if working memory capacity affects the retrieval of the subject/object in relative clauses (Gibson, 1998;Hsiao & Gibson, 2003). On the other hand, we expect faster processing of SRCs if frequency and experience with relative clauses dominate readers' processing (Hsiao & MacDonald, 2013;Jäger et al, 2015;Lin & Bever, 2006;Vasishth et al, 2013). We will examine these effects at the relativizer de, second noun, and main clause verb regions.…”

“…Researchers have found that Chinese SRCs are easier to process and comprehend than ORCs in a series of self-paced reading and eye-tracking experiments (e.g., Jäger, Chen, Li, Lin, & Vasishth, 2015;Lin & Bever, 2006;Vasishth, Chen, Li, & Guo, 2013). They argue that the findings support experience-based or frequentist theories: because SRCs are more frequent than ORCs in Chinese, readers have been exposed to SRCs more and thus process and comprehend Chinese SRCs more quickly and easily (Gennari & MacDonald, 2008;Hsiao & MacDonald, 2013;Levy, 2008).…”

“…Researchers have found that Chinese SRCs are easier to process and comprehend than ORCs in a series of self-paced reading and eye-tracking experiments (e.g., Jäger, Chen, Li, Lin, & Vasishth, 2015;Lin & Bever, 2006;Vasishth, Chen, Li, & Guo, 2013). They argue that the findings support experience-based or frequentist theories: because SRCs are more frequent than ORCs in Chinese, readers have been exposed to SRCs more and thus process and comprehend Chinese SRCs more quickly and easily (Gennari & MacDonald, 2008;Hsiao & MacDonald, 2013;Levy, 2008).Conversely, other researchers have observed that SRCs are processed more slowly than ORCs, as evidenced by inflated reading times on the first noun, relative clause verb, relativizer de, and second noun in several self-paced reading studies (Chen, Ning, & Dunlap, 2008; Gibson & Wu, 2011;Hsiao & Gibson, 2003). Based on these findings, these authors propose working-memory-based theories (e.g., Gibson, 1998) account for the difference: when reaching the head noun in a Chinese relative clause, readers have more difficulty retrieving the relative clause verb from their working memory in SRCs than ORCs, because the distance between the relative clause verb and the head noun is greater in SRCs than ORCs.…”