Putting Lexical Cues into Discourse: Corpus Study of Relative Clauses in Child-Directed and Child Speech

• In Event: Integrating Morphosyntactic and Discourse Cues to Process Sentences: Insights from Turkish, Cantonese, and English

Abstract

In many languages (e.g., English and German), object relative clauses (ORCs: the cat that the dog chased) are typically more difficult to process than subject relative clauses (SRCs: the dog that chased the cat). Yet, it has been found that if the noun phrases (NPs) in an ORC meet certain expectations (e.g., if the head NP is inanimate) or when an ORC is preceded by appropriate discourse, adults do not find ORCs harder to process than SRCs (e.g., Gennari & MacDonald, 2008; Roland et al., 2012). Children also have morphosyntactic expectations for the NPs in different types of relative clauses (RCs; e.g., Diessel & Tomasello, 2000) and are highly sensitive to aspects of discourse (e.g., Song & Fisher, 2007). However, experiments with children traditionally only focus on morphosyntactic factors and present RCs without any discourse context. Here, we examined the relationship between the form (morphosyntax) and function (discourse-related) of NPs in SRCs and ORCs in child-directed and child speech through densely-collected English-speaking developmental corpus data (range = 2-5 years) from three caregiver-child dyads.
We analysed 1126 and 495 RCs from the caregivers and children respectively. We coded type of RC (SRC vs. ORC) and animacy, NP type (pronoun vs. proper noun vs. lexical NP), givenness, and topichood of the head and embedded NPs, and definiteness of lexical NPs.
Separate sets of linear mixed-effects models were fitted on the caregiver and child data. For both the caregiver and child data, we found that, in line with previous findings (e.g., Macdonald et al., 2020), the head NP in ORCs and the embedded NP in SRCs tend to be inanimate.
For the caregiver data only, our model suggests that a RC is more likely to be a SRC if the embedded NP is a proper noun (compared to pronoun). However, inconsistent with previous findings (e.g., Roland et al., 2002), a RC is more likely to be an ORC if the embedded NP is not the topic of the on-going discourse. We will discuss this unexpected finding in relation to the operationalisation of topichood in different studies.
Our model on the child data suggests that, as expected, a RC is more likely to be a SRC if the head NP is indefinite and/or if the embedded NP is a lexical NP (compared to pronoun). However, dissimilar to findings from previous adult studies (e.g., Roland et al., 2002), a RC is more likely to be an ORC if the embedded NP is new. This finding suggests a developmental change on the role of embedded NP givenness on the production of ORCs.
Taken together, both children and adults draw on discourse-level cues when producing RCs. Adults’ production of ORCs is more contingent on the topichood of the embedded NP, whereas children’s production of ORCs is more contingent on the givenness of the embedded NP. Our findings suggest that the form-function expectations for NPs in RCs change across development, and experimental studies examining the processing of RCs should take both morphosyntactic and discourse-level factors into account.

Authors

• Kin Chung Jacky Chan, Lancaster University
  Presenting Author

• Anna Theakston, University of Manchester

• Silke Brandt