A Data-driven Investigation of Corrective Feedback on Subject Omission Errors in First Language Acquisition

Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning (CoNLL), pages 105–114, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics A Data-driven Investigation of Corrective Feedback on Subject Omission Errors in First Language Acquisition Sarah Hiller and Raquel Fern´andez Institute for Logic, Language & Computation University of Amsterdam sarah.hiller@posteo.de | raquel.fernandez@uva.nl Abstract We investigate implicit corrections in the form of contrastive discourse in child- adult interaction, which have been argued to contribute to language learning. In con- trast to previous work in psycholinguis- tics, we adopt a data-driven methodol- ogy, using comparably large amounts of data and leveraging computational meth- ods. We conduct a corpus study on the use of parental corrective feedback and show that its presence in child directed speech is associated with a reduction of child sub- ject omission errors in English. 1 Introduction It is widely agreed that children learn how to use language through interaction with their caregivers and peers. There is, however, a long-standing dis- cussion concerning the exact nature of the learn- ing mechanism enabling this. For example, while there is no doubt that children are exposed to posi- tive input (i.e., grammatically correct utterances in context), it is an open question whether they also receive negative input—evidence about the inade- quacy of their erroneous utterances. What seems clear is that explicit disapprovals are very rarely used to correct grammatical mis- takes, as already shown by Brown and Hanlon (1970). However, certain contrastive constructions may provide negative input in an implicit way. For instance, in the following exchange between 2-year-old Lara and her father, from the corpus by Rowland and Fletcher (2006), the father picks up the child’s erroneous utterance in the follow- ing turn and presents a form with the appropriate preposition: (1) CHI: I climb up daddy . DAD: you did climb over daddy . The contrast between the two forms is particularly noticeable and could potentially lead the child to recognise and correct their own error. It has thus been argued that this type of construction presents children with negative evidence and, unlike ex- plicit corrections, it does so in the course of con- versation, without disrupting the dialogue flow. Researchers have used different terms to refer to this phenomenon, including recast (Brown, 1973), reformulation (Chouinard and Clark, 2003), em- bedded correction (Clark, 2003), and corrective input (Saxton, 2000). In this paper, we adopt the term corrective feedback (CF), which we will de- fine precisely in Section 3, and analyse its effect on first language learning—in particular on retreating from subject omission errors in English. In con- trast to previous work in psycholinguistics, we in- vestigate these questions in a data-driven manner, using comparably large amounts of data from the CHILDES Database (MacWhinney, 2000a) and developing computational methods to support lin- guistically motivated studies. More concretely, we make the following contributions: • We present a taxonomy of child error types that can receive CF and an annotation scheme for coding instances of CF. • We report …

Content truncated to 500 words