Constructional patterns in child bilingual code-mixing: A usage-based corpus approach

本项目探讨了在基于用法的结构语法意义上，如何用结构模式来解释双语语言习得中的语码混合话语。这种方法在单语习得研究中已被证明是非常成功的，但直到最近才被用于多语习得研究。在拟议的项目中，语码混合被定义为在一个话语中使用多个问题，例如der moon kann fly“月亮可以飞”。我们认为，基于使用的方法可以为我们理解多语言语言习得，特别是儿童的语码混合话语提供重要的见解。为此，我们采用定量方法系统分析了德英双语儿童及其照顾者的数据，以识别语言模式。我们的主要假设是，儿童的话语可以在结构模式的帮助下进行解释——固定的多词块，比如What 's this？以及像[What 's X]这样的帧槽模式——从（单语）输入中抽象出来，孩子们的代码混合也可以用这样的模式来解释。我们的数据库是在不同输入情境下成长的2 - 4岁德英双语儿童的三个语料库。我们使用的语料库在大小和密度方面都是独一无二的：它们涵盖了超过一年半的时间，平均每周记录两个小时的孩子。该项目的一个主要目标是为这些语料库添加语言注释，并使其对公众可用。为了分析数据，我们使用了最近文献中的两种互补方法。使用McCauley & Christiansen的（例如2017,2019）基于块的学习者，我们基于纵向语料库数据对习得过程中结构模式的发展进行了建模。为了做到这一点，简单的频率测量与心理启发的学习机制相结合，以发展语言学习者的（预测）语法。该方法由Ibbotson等人（2019）提出的动态网络方法进行补充，该方法基于共现频率和过渡概率对语言学习者语言知识的发展进行建模。总而言之，我们的项目可以为更好地理解多语言习得以及儿童接受的输入与他们使用代码混合之间的复杂关系做出重大贡献。

This project addresses the question of how code-mixing utterances in bilingual language acquisition can be accounted for in terms of constructional patterns in the sense of usage-based construction grammar. This approach has proven highly successful in research on monolingual language acquisition but has only recently been adopted in research on multilingualism as well. In the context of the proposed project, code-mixing is defined as the use of multiple question in one utterance, e.g. der moon kann fly ‘the moon can fly’. We argue that a usage-based approach can add important insights to our understanding of multilingual language acquisition in general and children’s code-mixing utterances in particular. To this end, we systematically analyze data of German-English bilingual children and their caregivers using quantitative methods to identify linguistic patterns. Our main hypothesis is that the children’s utterances can be accounted for with the help of constructional patterns – fixed multi-word chunks such as What’s this? and frame-and-slot patterns like [What’s X] – that are abstracted away from the (monolingual) input, and that the children’s code-mixing can be accounted for using such patterns as well. Our database are three corpora of German-English bilingual children aged between 2 and 4 years growing up in different input situations. The corpora we use are unique in terms of size and density: They cover a period of more than one and a half years, with the children being recorded on average two hours per week. A major goal of the project is to add linguistic annotations to these corpora and to make them available to the public. To analyze the data, we use two complementary methods from the recent literature. Using McCauley & Christiansen’s (e.g. 2017, 2019) Chunk-Based Learner, we model the development of the constructional patterns over the course of acquisition on the basis of our longitudinal corpus data. To do so, simple frequency measures are combined with psychologically inspired learning mechanisms in order to develop a (predictive) grammar of a language learner. This method is complemented by a dynamic network approach as proposed by Ibbotson et al. (2019), which models the development of the language learners’ linguistic knowledge on the basis of co-occurrence frequencies and transitional probabilities. All in all, our project can thus contribute significantly to a better understanding of multilingual language acquisition in general and the complex relationship between the input that children receive and their use of code-mixing in particular.