Canada

Little research to date has examined how listeners integrate cues to non-native speaker identity in real time sentence processing. Here, we examine listeners’ interpretation of the semantic and socio-pragmatic content of utterances produced by either a foreign accented speaker or a native speaker. Overall, our findings suggest that processing speed was slower in the presence of foreign accents. However, the extra perceptual demands of processing unfamiliar accents did not translate into listeners’ accuracy rates, and in certain sentence contexts, non-native speakers were also more likely to elicit higher semantic or pragmatic interpretation accuracy. Our findings show that non-native speaker identity plays an important role in listeners’ sentence interpretations.

CogSci 2022

Integrating Non-Native Speaker Identity in Semantic and Pragmatic Processing

poster

Join us for CogSci2022! Our 44th Annual Conference will be centered around the theme of diversity, with an accompanying range of keynotes and invited speakers who themselves highlight the diversity of our field. Our three symposia are Cognition across Diverse Populations, Cognition within Diverse Societies, and Global Challenges for Cognitive Science. The conference is being offered in a fully hybrid environment for onsite and online attendees.
 
 [**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/07/Cogsci-2022-PAAG-Detailed-July-12final-1.pdf)
 [**The Detailed Program can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/08/DeArmond-CogSci2022-program-v113-final.pdf)

This conference requires a registration to attend. Please use the link below to secure your ticket. The website will be available to all registered attendees on July 25th. 

[Register to Attend](https://cognitivesciencesociety.org/registration/)

Registration Is Required

The 44th Annual Meeting of the Cognitive Science Society highlighted research on the theme Cognitive Diversity in addition to the full breadth and diversity of research topics offered by the society’s membership. 





The study reported here examined the role of expertise in the composite face effect which constitutes better recognition of the top half of a face when in composite with a congruent vs. an incongruent (in terms of response required) bottom half. Experiment 1a (n=96) used prototype-defined artificial stimuli (checkerboards) to investigate the composite effect. The advantage of using these stimuli is that expertise can be fully controlled. Experiment 1b (n=96) aimed to replicate the composite effect in face stimuli which served as a control and provided a direct comparison of the composite effect between face and checkerboard stimuli. A full experimental design including congruent/incongruent aligned and misaligned composites was used in both experiments to measure the composite effect. Experiment 1a revealed that the composite effect could not be obtained in checkerboard composites. Experiment 1b confirmed the robust composite face effect. We interpret our results as suggesting that expertise/perceptual learning does not contribute to the composite effect for faces.

Investigating the Composite Effect in Prototype-Defined Checkerboards vs. Faces

In recent evidence, classification training can elicit two qualitative patterns of generalization: one is exemplar-based such that close proximity to known members of a category best predicts membership in that category; the other involves inducing a global form of coherence in the mapping between input space and category membership. Such global coherence is an abstraction about category membership – not in the form of clusters or prototypes, but grounded in regularities like categories alternating in input space (Kurtz & Wetzel, 2021) or one category having correlated feature values while the other is anti-correlated (Conaway & Kurtz, 2017). We investigate the extent to which categorization is driven by local match to exemplars versus conforming to global structural regularities using generalization items as critical tests: proximal to members of one category but conforming to the global regularity underlying the other. Results are discussed in terms of implications for theoretical accounts of category learning.

Local versus global coherence in the generalization of category training

We investigated how the properties of lexical items, which label object features, affect concept tokening. We addressed this issue by modeling data from three sources: (1) norms obtained from a dataset of 78,000 features to a set of pictures representing living and nonliving objects; (2) accuracy data from a picture-word priming congruency task with stimuli presented for 50-60 milliseconds; and (3) corpus data on the lexical properties of four different social usage count measures. We conducted two sets of analyses: one relying on sample count-based measures (i.e., measures based on the norming study: sample frequency, cue validity, feature distinctiveness), and a second relying on the social usage count-based measures (i.e., word frequency (WF), contextual diversity (CD), discourse contextual diversity (DCD), and user contextual diversity (UCD). Contrasting count and social usage-based measures allowed us to gain insight into the contribution of diverse semantic and socially oriented contextual measures of lexical items, and how they may affect concept tokening. Our results show that cue validity and feature distinctiveness were negative predictors of participants’ accuracy to congruency decisions—an effect which was more pronounced for distinctive features of living things. There was also a noticeable advantage for the UCD and DCD variables, over CD and WF. Overall, our results suggest that the conceptual system may be organized as a function of both, intrinsic properties of object features and usage based contextual measures of lexical items that label these features.

Which semantic properties of a feature affect access to an object concept?

Rosch (1975) proposed that some exemplars are more typical of a category than others. Typicality gradients have historically been estimated from data collected from human participants. Here, we investigated whether they can be estimated from a computer vision model, VGG19, guided by cognitive science models of concepts and categorization. Following prototype models, we estimated the similarity of each exemplar to the concept prototype computed in two ways (average across all exemplars, most typical exemplar). Following exemplar models, we computed the pairwise similarities between pairs of exemplars and used MDS to order them on a continuum. The prototype-average model achieved the highest rank-order correlation to human typicality ratings of exemplars of the bird category. Thus, computer vision models may have some promise for generating human-like typicality gradients. We are extending this work to utilize newer computer vision models such as ResNet, and to encompass a broader range of categories.

Typicality Gradients in Computer Vision Models

We study the wisdom of the crowd in the context of spatial knowledge, asking participants to identify US states and African countries on unlabeled tile maps. We use two question frames, asking participants to select where the target is present or eliminate where it is absent. Participants generally display overconfidence, often selecting small regions that do not include the target. We find strong wisdom of the crowd effects by aggregating participants' responses, especially by weighting the individual responses according to the size of their selection. The weighted crowd outperforms all but a few participants for the US states and all participants for the African countries. We also find wisdom of the crowd within effects, by aggregating the present and absent frames for the same participant. We discuss the implications of our findings for understanding how people express uncertain spatial knowledge and the potential use of crowd aggregation in real-world applications.

The Wisdom of the Crowd and Framing Effects in Spatial Knowledge

Many non-geographic factors influence spatial judgments, which implies that spatial representations are not metrically veridical. We investigated the influence of social and political factors in the geopolitical context of Cyprus–an island divided since 1974 into the Turkish Cypriot and Greek Cypriot communities in the north and south, respectively. Participants (249 Greek Cypriots, 322 Turkish Cypriots) indicated their familiarity with 19 towns, estimated town locations, and the straight-line distance between those towns. They also rated their attitudes toward the other community. Cypriots underestimated distances and contracted the placement of towns within the other community more so than within their own community. Moreover, those more willing to live together with Cypriots from the other community underestimated distances between towns, whereas those less willing to live together overestimated distances. The results support the notion that representations of global-scale environments have multi-faceted origins, including sociopolitical factors not usually associated with spatial representations.

Beyond the physical divide of Greek Cypriots and Turkish Cypriots: Social and political variables shape geographical estimates

People often segment spaces into hierarchically structured subspaces. Judgments about inter-point distance and direction are more accurate within than between segments. However, especially in large-scale complex spaces, segmentation may be necessary for flexible navigation. In this study, we looked at spatial segmentation in a real-life city. We asked citizens of Istanbul, a transcontinental city spread over Europe and Asia with natural waterways that divide it into multiple neighborhoods, to indicate how they segment their city and to perform spatial judgments between well-known landmarks. We examined segmentation effects for divisions they endorsed, and for those others use but they do not report using. Additionally, we examined the impact of gender, age, time spent in the city, and frequency of using connecting routes and bridges. We replicated basic segmentation effects for the primary division, used by all, between the European and Asian sides. For the European side, which has a geographic boundary (The Golden Horn), segmentation impaired the accuracy of spatial representation of participants. For the Asian side, where there is a potential division that is more notional, we found different effects. Individual’s age, sex, time spent in the city, and frequency of using connecting routes also influenced spatial judgments. These results suggest that (i) spatial segmentation effects exist in the real-world, (ii) segmentation in a city-scale environment is differently affected by physical and conceptual boundaries, and (iii) sex, age, and navigation experiences are associated with the cognitive representation of a city.

Neighborhoods, Directions and Distances: Segmentation Effects in a Real-World City

There is growing evidence that spatial abilities can be improved through training, including participation in hobbies and everyday activities that involve spatial thinking. In order to better assess the contributions of everyday spatial activities to the development of spatial skills, we developed a new self-report questionnaire of spatial activities by adding updates and navigation activities to an existing questionnaire. A principal component analysis revealed five interpretable components which were compared to measures of perspective taking, mental rotation and two other self-report scales. Small but significant correlations were found between the ‘navigation’ component of the spatial activities questionnaire and a self-report measure of sense of direction, as well as self-reported childhood wayfinding experience. No sex difference was found on the ‘navigation’ component. This questionnaire is currently being used in a large study of spatial abilities.

Large-Scale vs Small-Scale Spatial Abilities: Development of a Broad Spatial Activities Questionnaire

Linguistic alignment---the contingent reuse of our interlocutors' language at all levels of linguistic structure---pervades human dialogue. Here, we design unique measures to capture the degree of linguistic alignment between interlocutors' linguistic representations at three levels of structure: lexical, syntactic, and semantic. We track these measures in a longitudinal dataset of early conversations between caregivers and children with and without perinatal brain injury. Specifically, we test the predictions of the well-known Interactive Alignment Model, taking advantage of the variability within our sample in terms of the strength of interlocutors' linguistic representations, whether owed to age or injury. 
Ultimately, we find inconsistent support for the (largely untested) predictions of the Interactive Alignment Model, pointing to a need for new quantitative accounts of the mechanisms underlying linguistic alignment. Our results regarding the trajectory of interactive alignment broadly replicate developmental trends documented by other researchers, though analyses linking concurrent vocabulary and child alignment, as well as caregiver alignment and later child vocabulary---defy predictions from previous work. Our goal with these analyses is to start a conversation regarding the mechanisms underlying linguistic alignment, and to inform theories of how interactive linguistic experience supports language development.

Getting to the root of linguistic alignment: Testing the predictions of Interactive Alignment across developmental and biological variation in language skill

Computational feasibility is a widespread concern that guides the framing and modeling of biological and artificial intelligence. The specification of cognitive system capacities is often shaped by unexamined intuitive assumptions about the search space and complexity of a subcomputation. However, a mistaken intuition might make such initial conceptualizations misleading for what empirical questions appear relevant later on. We undertake here computational-level modeling and complexity analyses of segmentation - a widely hypothesized subcomputation that plays a requisite role in explanations of capacities across domains - as a case study to show how crucial it is to formally assess these assumptions. We mathematically prove two sets of results regarding hardness and search space size that may run counter to intuition, and position their implications with respect to existing views on the subcapacity.

Premium content

Integrating Non-Native Speaker Identity in Semantic and Pragmatic Processing

Downloads

Next from CogSci 2022

Investigating the Composite Effect in Prototype-Defined Checkerboards vs. Faces

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES