Jacolien van Rij

Cognitive architectures and language acquisition: A case study in pronoun comprehension

In this paper we discuss a computational cognitive model of childrens poor performance on pronoun interpretation (the so-called Delay of Principle B Effect, or DPBE). This cognitive model is based on a theoretical account that attributes the DPBE to childrens inability as hearers to also take into account the speakers perspective. The cognitive model predicts that child hearers are unable to do so because their speed of linguistic processing is too limited to perform this second step in interpretation. We tested this hypothesis empirically in a psycholinguistic study, in which we slowed down the speech rate to give children more time for interpretation, and in a computational simulation study. The results of the two studies confirm the predictions of our model. Moreover, these studies show that embedding a theory of linguistic competence in a cognitive architecture allows for the generation of detailed and testable predictions with respect to linguistic performance.

Autocorrelated Errors in Experimental Data in the Language Sciences: Some Solutions Offered by Generalized Additive Mixed Models

A problem that tends to be ignored in the statistical analysis of experimental data in the language sciences is that responses often constitute time series, which raises the problem of autocorrelated errors. If the errors indeed show autocorrelational structure, evaluation of the significance of predictors in the model becomes problematic due to potential anti-conservatism of p-values.

Toward cognitively constrained models of language processing: A review

Language processing is not an isolated capacity, but is embedded in other aspects of our cognition. However, it is still largely unexplored to what extent and how language processing interacts with general cognitive resources. This question can be investigated with cognitively constrained computational models, which simulate the cognitive processes involved in language processing. The theoretical claims implemented in cognitive models interact with general architectural constraints such as memory limitations. This way, it generates new predictions that can be tested in experiments, thus generating new data that can give rise to new theoretical insights. This theory-model-experiment cycle is a promising method for investigating aspects of language processing that are difficult to investigate with more traditional experimental techniques. This review specifically examines the language processing models of Lewis and Vasishth (2005), Reitter, Keller, and Moore (2011), and Van Rij, Van Rijn, and Hendriks (2010), all implemented in the cognitive architecture ACT-R (Anderson, Bothell, Byrne, Douglass, Lebiere, & Qin, 2004). These models are all limited by the assumptions about cognitive capacities provided by the cognitive architecture, but use different linguistic approaches. Because of this, their comparison provides insight into the extent to which assumptions about general cognitive resources influence concretely implemented models of linguistic competence. For example, the sheer speed and accuracy of human language processing is a current challenge in the field of cognitive modeling, as it does not seem to adhere to the same memory and processing capacities that have been found in other cognitive processes. Architecture-based cognitive models of language processing may be able to make explicit which language-specific resources are needed to acquire and process natural language. The review sheds light on cognitively constrained models of language processing from two angles: we discuss 1) whether currently adopted cognitive assumptions meet the requirements for language processing, and 2) how validated cognitive architectures can constrain linguistically motivated models, which, all other things being equal, will increase the cognitive plausibility of these models. Overall, the evaluation of cognitively constrained models of language processing will allow for a better understanding of the relation between data, linguistic theory, cognitive assumptions, and explanation.

Perception of spectrally degraded reflexives and pronouns by children

Speech perception skills in cochlear-implant users are often measured with simple speech materials. In children, it is crucial to fully characterize linguistic development, and this requires linguistically more meaningful materials. The authors propose using the comprehension of reflexives and pronouns, as these specific skills are acquired at different ages. According to the literature, normal-hearing children show adult-like comprehension of reflexives at age 5, while their comprehension of pronouns only reaches adult-like levels around age 10. To provide normative data, a group of younger children (5 to 8 yrs old), older children (10 and 11 yrs old), and adults were tested under conditions without or with spectral degradation, which simulated cochlear-implant speech transmission with four and eight channels. The results without degradation confirmed the different ages of acquisition of reflexives and pronouns. Adding spectral degradation reduced overall performance; however, it did not change the general pattern observed with non-degraded speech. This finding confirms that these linguistic milestones can also be measured with cochlear-implanted children, despite the reduced quality of sound transmission. Thus, the results of the study have implications for clinical practice, as they could contribute to setting realistic expectations and therapeutic goals for children who receive a cochlear implant.

Short-term exposure enhances perception of both between- and within-category acoustic information

A critical question in speech research is how listeners use non-discrete acoustic cues for discrimination between discrete alternative messages (e.g. words). Previous studies have shown that distributional learning can improve listeners’ discrimination of non-native speech sounds. Less is known about effects of training on perception of within-category acoustic detail. The present research investigates adult listeners’ perception of and discrimination between lexical tones without training or after a brief training exposure. Native speakers of German (a language without lexical tone) heard a 13-step pitch continuum of the syllable /li:/. Two different tasks were used to assess sensitivity to acoustic differences on this continuum: a) pitch height estimation and b) AX discrimination. Participants performed these tasks either without exposure or after exposure to a bimodal distribution of the pitch continuum. The AX discrimination results show that exposure to a bimodal distribution enhanced discrimination at the category boundary (i.e. categorical perception) of high vs. low tones. Interestingly, the pitch estimation task results followed a categorisation (sigmoid) function without exposure, but a linear function after exposure, suggesting estimates became less categorical in this task. The results suggest that training exposure may enhance not only discrimination between contrastive speech sounds (consistent with previous studies), but also perception of within-category acoustic differences. Different tasks may reveal different skills.

Visual World Paradigm Data: From Preprocessing to Nonlinear Time-Course Analysis

The Visual World Paradigm (VWP) is used to study online spoken language processing and produces time-series data. The data present challenges for analysis and they require significant preprocessing and are by nature nonlinear. Here, we discuss VWPre, a new tool for data preprocessing, and generalized additive mixed modeling (GAMM), a relatively new approach for nonlinear time-series analysis (using mgcv and itsadug), which are all available in R. An example application of GAMM using preprocessed data is provided to illustrate its advantages in addressing the issues inherent to other methods, allowing researchers to more fully understand and interpret VWP data.