Periodic Reporting for period 2 - ModelingCommonGround (Modeling Common Ground)
Berichtszeitraum: 2019-06-11 bis 2020-06-10
Language is one of the most important building blocks of human social life. We use it to coordinate our social activities, to transmit knowledge from one individual to another and to express our thoughts and feelings. One of the most astonishing features of language is that it allows us to communicate very precise meanings despite the fact that each utterance is inherently ambiguous. The meaning of words and sentences depends on the identity of the communicative partners and the nature of the context. In simple behavioral experiments children and adults can use a wide variety of social-contextual cues (jointly known as “common ground”) to interpret ambiguous utterances. But this limited empirical evidence – especially in the developmental context – does not live up to the theoretical importance of common ground: In theory, common ground is not only involved in online language use but it is also a necessary prerequisite to learn language in the first place. Studying the development of children’s ability to form and use common ground is therefore crucial to understand the psychological foundation of language. It is still unknown how both adults and children integrate different social-contextual cues in complex, naturalistic interactions. Bayesian modeling provides a mathematical framework for formalizing theoretical assumptions about this interaction and deriving quantitative predictions about new experimental situations.
This project will unite developmental and computational approaches. The key objective is to find out what constitutes common ground at different ages and how it informs language learning across development. I will develop mathematical models and behavioral experiments in parallel to obtain quantitative predictions for different forms of interactions between social-contextual cues. By comparing these predictions to data from early children’s word learning at different stages of development, I will be able to empirically evaluate the theoretical importance of the different components of common ground. The interdisciplinary focus of the project at the intersection of psychology, linguistics and computer science will open up new avenues for the empirical study of language use and language learning.
This project will unite developmental and computational approaches. The key objective is to find out what constitutes common ground at different ages and how it informs language learning across development. I will develop mathematical models and behavioral experiments in parallel to obtain quantitative predictions for different forms of interactions between social-contextual cues. By comparing these predictions to data from early children’s word learning at different stages of development, I will be able to empirically evaluate the theoretical importance of the different components of common ground. The interdisciplinary focus of the project at the intersection of psychology, linguistics and computer science will open up new avenues for the empirical study of language use and language learning.
Since the beginning of the project I reached the training objectives described in the proposal: I acquired skills in programming web-based experiments in javascript/html and programmed experiments for data collection with adults and children. In addition, I attended a course on probabilistic modeling and implemented the methods taught in this course for the modeling objectives of WP3. By the end of the outgoing phase in June 2019, all data collection and modeling objectives have been reached. We collected data from adults and children for WP1 and WP2. We used this data to inform model parameters and generate model predictions as specified in WP3. We also collected data from adults and children for WP4, in which we test the model predictions from WP3 against empirical data. The results show that the hypothesized model, in which information sources are flexibly traded off with one another, accurately predicts the empirical data. Furthermore, this model explains the data better compared to alternative models that consider only one type of information. Parts of these results were presented at the Boston University Conference on Language Development in November 2018. The empirical study was presented at the Cognitive Science Society Conference in 2019, the Boston University Conference on Language Development in 2019, and at the Budapest CEU Conference on Cognitive Development in 2020. In sum, all objectives that have been proposed for the outgoing phase have been reached. The results have been written up (adults and children combined to increase the impact of the paper) and submitted for publication at the journal Cognition (current status: revise and resubmit). In addition, we completed a second study within the project in which we asked how common ground information is integrated with linguistic information. This study is structured in the same way as the study described in the proposal: We measured children's developing linguistic knowledge, expectations about speaker inforamtiveness, and sensitivity to common ground (analogous to WP 1 and 2). We then formalized the process by which these types of information should be integrated (WP4). Finally, we collected data from children in a task in which the three information sources were manipulated at a time (WP3). When we compared the data from this experiment to the predictions by the model, we found a close alignment between the model predictions and the data. Furthermore, we formulated a number of competitor models that explored different ways in which information sources could be integrated and how this process develops. The model that best predicted the data was one in which all information sources are part of an integrated inference process which is stable over time. The main locus of development is an increased sensitivity to the individual information sources. These results partially replicate, but also extend, the outcome of the first study. We are in the process of writing this second study up and will submit it for publication within the next month. As announced in the proposal, I also worked on a theoretical paper, which was published in December 2019 in the inaugural issue of Annual Review of Developmental Psychology. It reviews the literature on pragmatic reasoning in language learning and describes a new theoretical framework to synthesize different strands of the literature.
All objectives described in the proposal were reached. In addition, we were able to conduct a second empirical study that extended the study described in the proposal. The manuscript reporting the first study has been submitted to one of the top journals in the field of cognitive science: Cognition (current status: revise and resubmit). The publication that will cover the second study is even more innovative and we are thus planning to submit it to a high ranking generalist journal (Nature Human Behaviour or PNAS). Our theoretical review paper in which we, among other things, embed the findings from the project in the wider literature, has been published in the inaugural issue of "Annual Reviews in Developmental Psychology", a new journal from the Annual Reviews family (which are usually benchmark journals in the respective field). Taken together, the three publications are likely to have a substantial impact on the field as they provide fellow researchers with a theoretical perspective as well as a methodological blueprint for how to study information integration in pragmatic language learning. Furthermore, all of our experiments, data, and model code is available in public repositories. Several researchers have already contacted me about re-using some of the experiments or stimuli. Thus, the project is already beginning to have an impact on the field.