The Role of Language Similarity in the Development of the Infant Bilingual Lexicon

在双语家庭中长大的婴儿是如何学习单词的，这些单词在婴儿双语头脑中是如何表现的？我们认为，这些问题的答案在很大程度上受到所涉及语言的相似性的影响。例如，荷兰语和英语是亲密的语言，而汉语和英语是遥远的。在单词层面上，语言相似性反映在两种语言共享的同源词的百分比上。同源词是具有相同词源和发音相似的单词翻译(如英语和荷兰语中的父亲和维达)。双语者同源词重要性的证据来自对成年人的命名研究。例如，成年双语者在对应于同源词(父亲/维德)时，比起对应于非同源词(自行车/汽车)的图片时，命名速度更快。成人双语的一个重要特征是非选择性地接触两种语言的单词。双语者激活来自任何一种语言的所有可能的候选单词，即使在完全只有一种语言的情况下，也是如此。这些发现表明，在双语思维中，两种语言之间存在着重要的互动。在双语环境下长大的婴儿中，这种互动可能更加重要。这些婴儿通常只知道他们的一种语言中的一个单词来表示一个概念。在这种情况下，跨语言词汇相似性可能有助于单词识别，但可能会阻碍特定语言词汇的建立。我们打算进行一系列的实验研究，以揭示双语婴儿的心理词汇结构，并构建脑启发模型来探索他们词汇结构的理论解释(S)。我们将利用牛津和巴塞罗那的双语学习环境。在巴塞罗那，双语婴儿可以随时学习西班牙语和加泰罗尼亚语(两种罗曼语有许多同源词)。在牛津，有足够多的西班牙语和英语双语家庭。西班牙语和英语的同源词较少，与巴塞罗那的语境形成了鲜明的对比。实验：我们将使用新的实验范式，该范式是牛津婴儿实验室在单语婴儿身上首创的，使用了众所周知的眼球跟踪和神经成像技术。用单语婴儿开发的方法非常适合探索我们的研究问题。这些任务要求婴儿在听讲话的同时看图片。通过监测他们的眼球运动和记录他们的大脑活动，我们希望发现关于他们的两种语言如何影响他们对所看到的图片的反应的重要线索。该项目的一个重要方法优势是，它将以纵向方式跟踪婴儿在18、24和30个月大时的表现。纵向设计使我们能够考虑婴儿之间在词汇发展方面观察到的显著个体差异，并对项目收集的数据进行复杂而有力的统计分析。我们还将在与我们的双语人群完全相同的一套研究中测试单语婴儿，从而提供适当的基线来评估双语婴儿。建模：构建神经模型的目的是了解作为观察到的行为的基础的机制。目前的双语计算模型只考虑成人(大而稳定)的词汇量。我们将开发第一个早期双语学习的计算模型。最近，我们展示了单词学习的神经网络模型中自组织过程的发展变化如何捕捉到婴儿行为的重要方面。我们将把这些建模见解从单语婴儿的单词学习和识别转化为婴儿的双语发展，以一种尊重我们实验结果的方式。

How do infants growing up in bilingual homes learn words and how are these words represented in the infant bilingual mind? We propose that answers to these questions are heavily influenced by the similarity of the languages involved. For instance, Dutch and English are close languages, while Chinese and English are distant. At the word level, language similarity is reflected in the percentage of 'cognate' words that two languages share. Cognates are word translations that share the same etymological origin and sound similar (such as father and vader in English and Dutch). Evidence for the importance of cognates in bilinguals comes from naming studies with adults. For example, adult bilinguals can name pictures faster when they correspond to cognates (father/vader) than pictures corresponding to non-cognates (bicycle/fiets). A critical property of adult bilingualism is that of non-selective access to the words in both languages. Bilinguals activate all potential word candidates from either language consistent with what they hear or want to say, even in fully monolingual situations. Such findings point to important interactions between the two languages in the bilingual mind. Such interactions are likely to be even more important in the case of infants raised bilingually. Such infants often know only one word in one of their languages to refer to a concept. In this situation, cross-language lexical similarity may facilitate word recognition, but it may hinder the establishment of language-specific lexicons. We propose to conduct a series of experimental studies to uncover the structure of the mental lexicon in bilingual infants, and to construct brain-inspired models to explore theoretical accounts of the structure of their lexicon(s). We will exploit the availability of bilingual learning environments in Oxford and Barcelona. In Barcelona, there is ready access to bilingual infants learning both Spanish and Catalan (two Romance languages sharing many cognates). In Oxford, there is a sufficiently large number of bilingual Spanish and English families. Spanish and English have fewer cognates and offers a critical contrast to the Barcelona context.Experiments: We will use new experimental paradigms, pioneered on monolingual infants in the Oxford BabyLab, that employ well-understood eye-tracking and neuro-imaging techniques. The methods developed with monolingual infants are ideally suited to exploring our research questions. The tasks involve infants looking at pictures at the same time as they hear speech. By monitoring their eye-movements and recording their brain activity, we hope to discover important clues as to how their two languages influence their reaction to the pictures they see. An important methodological advantage of this project is that it will follow infants' performance in a longitudinal manner - at 18, 24 and 30 months of age. A longitudinal design allows us to consider the substantial individual differences observed between infants in their vocabulary development and to apply sophisticated and powerful statistical analyses to the data collected on the project. We will also test monolingual infants on exactly the same set of studies as our bilingual population, thereby providing the appropriate baseline against which to assess the bilingual infants.Modelling: The purpose in constructing a neural model is to understand the mechanism that underlies observed behaviour. Current bilingual computational models only consider adult (large and stable) vocabularies. We will develop the first computational model of early bilingual learning. Recently, we have shown how developmental changes in the self-organising processes in neural network models of word learning can capture important aspects of infant behaviour. We will translate these modelling insights from monolingual infant word learning and recognition to infant bilingual development, in a manner that honours the results of our experiments.