Modeling Human Reasoning: A Computational Comparison of Single- vs. Dual-Process Theories

This project investigates the long-standing debate between Dual Process Theory (DPT) and Single Process Theory (SPT) in human reasoning through the use of computational cognitive modeling. While traditional dual-process accounts argue that intuitive and deliberative reasoning arise from qualitatively distinct cognitive systems, single-process approaches propose that both forms of reasoning emerge from a unified cognitive mechanism. Despite decades of research, these competing frameworks have largely remained verbal theories, making it difficult to distinguish between them empirically. To address this limitation, the present project applies Evidence Accumulation Models (EAMs), particularly the Diffusion Decision Model (DDM), to formally represent and test the assumptions underlying these theories.

Eye-Movement Indicators of Reading Comprehension Outcomes: A Meta-analysis

Recent work on the relationship between eye movements and reading comprehension suggests that while eye movements can predict comprehension outcomes, this relationship is influenced by aspects of the reader, text, and task. To date, eye-movement indicators of reading comprehension outcomes have not been clearly identified. We explored this issue by running a meta-analysis on the relationship between eye-movement and reading comprehension. We ran a search across major databases with keywords limiting the search to eye-tracking studies (e.g., eye-tracking, eye movements) and reading comprehension (reading, learning, text comprehension). The initial search yielded 15022 results after duplicate removal, and a total of 135 articles were included after abstract and full text screening. Correlations between eye-movement measures and comprehension outcomes were extracted directly from published reports, calculated from open-source data, or provided directly by the authors upon request. As many of the studies identified did not directly test or report the relationship between eye movements and comprehension outcomes, data was only available from a limited number of studies. From the original 135 studies, 104 datasets from 37 studies/corpus were available for analysis. However, due to limited data availability across samples, text types, and tasks formats, we focused the analysis on data from adults reading expository texts followed by multiple choice questions (~54% and ~78% of the data for L1 and L2 readers respectively). Random-effects meta-analysis models were conducted separately for eye movement measures for L1 and L2 speakers for individual eye-movement measures, and examined effects of language characteristics (e.g., transparency) on the relationship between eye movements and comprehension. For L1 readers, results suggest that making shorter fixations, particularly during first-pass, and making more fixations is associated with better comprehension outcomes. For L2 readers the pattern was similar with shorter fixations across both early and late measures associated with better comprehension outcomes, although fixation count was not associated with comprehension. These results will be further examined with regards to language characteristics. Limitations due to the lack of data from children and data using varied types of text, comprehension tasks, as well as the possible mediating role of these characteristics will also be discussed.

Language, Perception, and Attention: Shared Representations & Cross-Domain Interplay

The world that we perceive and describe changes constantly. If we believe our descriptions of the world to be accurate and consistent, we must assume that the content and the structure of our speech accurately and consistently reflect the world’s constantly changing nature. This requires a cognitive system that maps dynamic perceptual input onto language in a systematic way—one where attentional salience, conceptual organization, and grammatical structure interact seamlessly. At the heart of this process is the Cognition-Language Interface: a mechanism that filters, selects, and encodes visual and conceptual information into linguistic output, reflecting both the event’s structure and the speaker’s attentional state. Critically, this interface operates via regular mappings between the speaker’s or comprehender’s attentional state (e.g., noticeability, relevance, accessibility) and the language they produce or process, blending bottom-up and top-down cognitive processes. This interface also illuminates a broader question about the interplay between domain-general (e.g., attention, memory) and domain-specific (e.g., language, arithmetic) cognitive systems, which can be further subdivided into two more specific questions:

1. Do mental representations across different knowledge domains share organizational principles, either due to representational overlap or higher-level structures that bridge these domains?

2. How do domain-specific processes—including language—interact with general cognitive systems such as attention and memory?

In my talk, I will address these questions through two lines of research. First, I will present evidence from domain-specific and cross-domain priming studies, highlighting common organizational principles underlying abstract-knowledge systems (e.g., mental arithmetic and syntactic processing). Second, I will discuss studies investigating the visuospatial and attentional mechanisms involved in accessing both concrete and abstract lexical-semantic representations. Together, this research reveals how the Cognition-Language Interface bridges perception and language and exemplifies the integration of specialized and general cognitive systems.

Investigating Perceptual Span Across Writing Systems

In the process of text reading, individuals extract visual and linguistic information from words located beyond the point of direct fixation. The spatial limit of visual processing in reading, referred to as the perceptual span, constitutes a fundamental construct in reading and eye-movement studies. Perceptual span is commonly investigated through the gaze-contingent moving window paradigm. In this paradigm, a segment of text is displayed around the reader’s gaze position, which dynamically shifts in accordance with eye movements. Thus, the text visible at any given moment is contingent upon where the reader is looking. By experimentally manipulating the size of this window, researchers can determine the minimal amount of information necessary to sustain typical reading performance. This presentation will review empirical findings on the perceptual span across diverse writing systems and populations, underscoring that the span is modulated by factors such as visual complexity, script orientation, and individual reading proficiency. Collectively, these studies demonstrate that the linguistic environment exerts a significant influence on the mechanisms of reading.

The role of probabilistic processes in deductive and inductive reasoning

Deductive and inductive reasoning are two fundamental forms of human inference. Deductive reasoning concerns whether a conclusion necessarily follows from its premises, whereas inductive reasoning concerns whether a conclusion is plausible given the available evidence. A key question in reasoning research is whether these two forms of reasoning rely on distinct cognitive processes or on a shared mechanism. In this talk, I present two experiments examining how logical structure and semantic cues influence reasoning judgments under deductive and inductive instructions. Participants evaluated arguments that varied in logical validity and in semantic properties of the premises, such as their diversity or specificity. The results show strong and consistent sensitivity to logical validity across both instruction types, suggesting that logical structure influences judgments even when people evaluate plausibility. In contrast, semantic cues had smaller and less consistent effects. These findings challenge strong dual-process accounts that assume a clear separation between deductive and inductive reasoning and instead support the view that both deduction and induction operate through the integration of multiple cues.

Co-registration of EEG and eye-tracking in reading research: Results from preview benefit and naturalistic reading studies

EEG and eye-tracking research on reading have long seemed methodologically incompatible because of the artefacts that eye movements cause in the EEG. This has resulted in the adoption of single word reading paradigms or unnatural word-by-word presentations of sentences to study processes underlying reading with EEG. Recent methodological advances allow for the co-registration of EEG and eye-tracking data and for artifact correction in the EEG. I will present a series of co-registration studies in Chinese that demonstrate the feasibility of the method and highlight its use to address questions beyond what has been possible with traditional EEG studies. We first looked at parafoveal preview effects in fixation-related potentials (FRP) in a series of studies and found these effects showed a consistent reduction of the N1/N250 negativity after preview of the same character compared to a different one across experiments. The preview effects in the FRPs were also similar to foveal repetition effects in a masked priming study suggesting facilitation of visual-orthographic processing. Moreover, the preview effects were modulated by the experience with the reading direction. In another study, we presented text in a more naturalistic sentence-by-sentence way and used a regression approach to analyze word characteristics. In particular, we were interested in the effects of the orthographic prediction error (a pixel-by-pixel deviance from the visual average across all words), word frequency, and context-based surprisal (derived from a Large Language Model). Results showed oPE effects at around 150 ms, word frequency effects at around 200 ms, and context effects at around 400 ms. Moreover, context-based characteristics constrained oPE effects, which is in agreement with a hierarchical predictive coding model of neural processing during naturalistic reading. Overall, the studies suggest that co-registration of EEG and eye-tracking is feasible to conduct experiments with time-sensitive boundary paradigms and naturalistic reading opening the way to a better integration of the fields of EEG and eye-tracking in research of reading.

Using eye-tracking to evaluate how individual difference affect the processing of lexically and syntactically ambiguous sentences

Despite decades of psycholinguistic research, we still do not understand how readers resolve and understand lexical and syntactic ambiguities. Current theories posit that either one (Rayner et al., 1983) or two (Duffy et al., 1988) processes are required to resolve the two types of ambiguity, and these theories also differ in other key assumptions (e.g., whether readers maintain one or multiple meanings at a time). In addition, many of the seminal experiments on ambiguity resolution used methods that fall short of today's standard (e.g., small samples, outdated statistical tests, unnatural presentation formats). Our eye-tracking experiment will redress these limitations by replicating the sentence reading experiments of both Rayner et al. and Duffy et al. using modern methods (e.g., a larger sample of ~120 participants). We will also investigate the influence of individual differences in ambiguity processing by replicating MacDonald et al.'s (1992) observation that working memory capacity is related to ambiguity resolution and by examining the possible role of reading ability in ambiguity resolution. Using this approach, we will attempt to determine whether single- or dual-process theories provide a better account of ambiguity resolution. Our findings will also provide a corpus for evaluating computer models of reading (see Reichle, 2021) and for guiding future neuroimaging studies (e.g., to understand the involvement of specific cortical regions in ambiguity resolution).

What is the shape of semantic predictions?

It is common for language researchers to claim that humans consistently make predictions about upcoming linguistic input, such as the meaning of an upcoming word.  While many researchers agree that semantic predictions are graded in nature, rather than all-or-none, there is little direct work exploring the contours of these predictions. In this talk, I attempt to answer several questions about semantic predictions, including: How do raters and readers respond as presented words become more and more semantically distant from an expected word? How semantically distant must a presented word be before it becomes anomalous and triggers a ‘prediction error’? How do semantic predictions change as contextual constraint increases?  Do modality (offline ratings vs. online reading) and time (early processing vs. late) influence these patterns? Is prediction an obligatory or optional process? To answer these questions, I rely on a variety of data, including corpus analyses, semantic acceptability ratings, and eye tracking data. 

Bridging the Vocabulary Gap: Using GenAI to Support EALD Students

This is a practice run of my upcoming presentation for the UNESCO event entitled 'Harnessing Emerging Technologies to Support Language Inclusion'.

EALD students face a persistent vocabulary gap that limits their access to curriculum content. I'll present a pilot study investigating whether a generative AI tool can support multilingual primary students' academic vocabulary development through personalised, conversational pre-teaching aligned with the NSW science curriculum. With data collection still underway (n = 24 so far), preliminary results showed significantly gains for trained vocabulary compared to untrained words.

The Curious Case of Rereading

Research on eye movements during reading shows that the eyes move backwards in the text approximately 10%-15% of the time. These “regressions” are not all the same, however, in terms of distance or subsequent reading behaviors. This talk considers processes that are proposed to occur during the rereading that typically follows long-distance regressions. It is argued that previous assumptions about these processes might be incorrect. Finally, one possible option for improving the efficiency of rereading will be suggested.

Morphosemantic/morphosyntactic parafoveal processing in skilled readers: evidence from English

Skilled reading is characterized by eye movements that allow the reader to gather the necessary word related information quickly and efficiently from the page by being able to better use information in the parafovea. The open question is what kind of visual and linguistic information skilled readers gather from the parafovea and use to plan their eye movements. In this talk I present results from several experiments with gaze-contingent boundary-change paradigm that manipulated the predictability of the semantic and syntactic context in English. The results reveal differentiated time-course and depth of processing for the semantic and morphosemantic/morphosyntactic information that was manipulated parafoveally prior to being directly fixated by the readers.

While target words were read faster in the predictable context regardless of the previews (main effect of predictability), semantically manipulated previews yielded consistent preview benefit over the nonword and preview cost over the identical (no-change) previews (main effect of the preview manipulation no interaction with the context predictability). The effect of morphosyntactic/morphosemantic information presented parafoveally depended on the context, specifically how predictable the previewed word would be given the prior context. Specifically, the effect of the previewed word, including whether it led to longer or shorter looks, varied based on how predictable or unpredictable that word would be given the previous context.

The present work contributes to growing evidence that points to the flexible, contextually sensitive capacity of the allocation of attention within the parafoveal portion of the visual field and proposes predictability of the syntactic context as an additional dimension that guides parafoveal perception in skilled readers.

Understanding Emotion Word Processing in Aphasia and Language Unimpaired Speakers: A Multiple Methods Approach

Words that describe emotions, such as happiness, surprise, and anxiety, play a crucial role in communication by helping individuals express opinions, build relationships, and construct a sense of self. This thesis investigates how emotion words are processed, which semantic dimensions underlie their processing, and whether they form a distinct subset within abstract words. It tests and extends previous research by addressing an apparent contradiction: studies with language-unimpaired speakers suggest that emotion words may have a processing advantage over other abstract words, whereas studies with people with aphasia suggest that emotion words may be selectively impaired in discourse production.

With this motivation, emotion word processing was explored using different single-word tasks (lexical decision, reading aloud, free recall, and item recognition) and participant groups (people with aphasia and language unimpaired speakers). Findings were interpreted with respect to the cognitive demands of each task and the role played by semantic processing. The findings have implications for theoretical accounts of emotion-label word processing and for clinical understanding of abstract word production and processing in aphasia.

So what you're really saying is...

Our ability to comprehend language relies on our pragmatic capacity to infer meaning that extends beyond the literal content of words, phrases, and sentences. In this study we used eye-tracking to investigate the processing costs associated with generating and cancelling one type of pragmatic inference, scalar implicatures, during reading. Our findings demonstrate that readers show increased reading times and patterns of re-reading when implicatures are cancelled, suggesting that readers do represent inferential meanings that are not overtly present in the text. Our results help to inform theoretical accounts of linguistic knowledge and how that knowledge is integrated into sentence processing.

Natural Word Segmentation Cue in Chinese Sentence Reading

The absence of clear word boundaries in Chinese raises questions about how readers segment continuous lines of characters into words for the purposes of their identification and saccade targeting.  Findings from other unspaced languages (e.g., Thai; Kasisopa et al., 2013) suggest that skilled readers can utilise orthographic cues to guide their eyes and segment words. This research investigates whether similar cues are used in Chinese by examining whether the frequently occurring, single-character function word ‘de’ (的) facilitates saccadic targeting and word segmentation. The eye movements of 41 native Chinese-speaking participants were recorded as they read sentences where the position of the character ‘de’ (i.e., the beginning vs. middle of a 5-character string, e.g., 的雪白天鹅 vs. 雪白的天鹅) and its presence (i.e., ‘de’ vs. a substitution character, e.g., 的雪白天鹅 vs. 只雪白天鹅) were orthogonally manipulated.  Our results showed that the character 'de' facilitated word segmentation: readers spent less time fixating five-character strings when ‘de’ was present than absent, and when it occurred in the middle of the five-character strings. However, our results also suggest that readers employ two distinct saccade-targeting processes: one that utilises cues such as ‘de’ to constrain saccade targets, and another that dynamically adjusts saccade lengths in response to processing difficulty.  These results are discussed in relation to models of Chinese reading (Li & Pollatek, 2020; Liu et al., 2024).

Individual differences in reading fluency in Schizophrenia: Evidence from eye movements and fixation-related potentials

Readers with Schizophrenia (Sz) show decreased reading fluency, which negatively impacts both social and occupational outcomes. This talk will present an ongoing line of work that integrates eye tracking, EEG, and computational modeling, to investigate individual differences in reading fluency in readers with and without Schizophrenia. Our findings suggest that both higher-level (e.g., lexical) and lower-level (e.g., oculomotor and visual) processes contribute to slower reading fluency in readers with Schizophrenia. Compared to a control group, readers with Sz show reduced fixation-P1 amplitudes, longer fixation durations, fewer skipped words, and large increases in refixation rates (i.e., the probability of fixating on a word more than once). Also, as evidence that readers with Sz are making more refixations to compensate for reductions in parafoveal processing, their refixations landed closer to the center of the word (i.e., the optimal viewing location) compared to the control group. Building on these findings, simulations using the E-Z Reader model of eye movement control during reading also indicated that reduced reading fluency in Schizophrenia reflects changes to both high- and low-level processing. This talk will discuss how these findings can provide empirical constraints for the development of eye movement control models that better account for individual differences during reading.

Conspiracy Theories, Misinformation, and Sincerity of Belief Reports

A key assumption of psychological research on belief in conspiracy theories and other forms of misinformation is that participants provide sincere belief reports. However, this talk presents new empirical research suggesting that insincerely claiming to believe in misinformation while participating in psychology studies might be quite common.

Rob will also facilitate some discussion about whether there might be scope for using eye-trackers to study whether or not people believe statements that they read during the discussion.

The role of semantic interpretability and syntactic legality in complex nonword recognition

This study explored the role of semantic interpretability and syntactic legality on complex nonword recognition. A rating experiment tested the correlation between these twolinguistic factors by asking participants to rate the interpretability of legal and illegal nonwords made up of stem-suffix combinations. Results showed that these two factors are highly correlated. In two further lexical decision experiments (unprimed lexical decision in Experiment 2 and masked primed lexical decision in Experiment 3), we carefully dissociated interpretability and legality by comparing four types of nonwords: high-interpretability syntactically legal, high-interpretability syntactically illegal, low-interpretability syntactically legal, and low-interpretability syntactically illegal nonwords. To test whether or not the activation of embedded stems was modulated by their morpho-semantic and morpho-syntactic context, all complex nonwords were compared against a matched non-stem control.  A significant effect of stem status was found in Experiments 2 and 3, providing evidence for the important role of embedded stems in complex nonword recognition. Moreover, a significant effect of interpretability was found only in the unprimed lexical decision (Experiment 2), but not in masked priming (Experiment 3), suggesting that semantics does not influence complex word recognition until participants have enough time to thoroughly process the nonword. No effect of syntactic legality was found in either experiment. These results highlight the independent roles of semantic interpretability and syntactic legality in visual nonword recognition, supporting an initial semantically blind stage in morphological parsing.

Causes and consequences of misinformation during reading: Insights from eye tracking

People frequently rely on information even after it has been retracted, a phenomenon known as the continued-influence effect (CIE) of misinformation. This research uses eye tracking to investigate the cognitive processes underlying the CIE during reading. The eye movements of 85 participants were recorded as they read pairs of short newspaper-style articles containing a critical piece of information that was either retracted or not. Participants subsequently completed a comprehension questionnaire that tested their memory for general and critical details as well as inferential judgements related to the retracted information. To determine whether the CIE is modulated by individual differences in cognition, participants also completed tests of their language proficiency and working memory ability. The results of the comprehension questionnaire replicated previous evidence that repetition of the original information improved recall memory of the event. Linear mixed-effects models assessed the impact of misinformation and individual differences on readers’ eye movements across early and late stages of processing. The results provide insights into the cognitive causes and consequences of the CIE during reading. Implications for the development of effective interventions to reduce effects of misinformation will be discussed.

What can emotion and abstract words tell us about subjective semantic ratings?

Semantic dimensions such as context availability, imageability and valence, form core components of many theoretical accounts of lexical processing. Typically, normative data for such semantic dimensions are drawn from subjective ratings, however, questions have been raised regarding the reliability and validity of these ratings. In this talk, I will discuss this issue with a focus on context availability norms. Using data collected for another study, we show that context availability ratings required significantly higher rates of data exclusions at the level of both participants and items compared to other variables. In addition, high standard deviations at the item level, indicated a substantial degree of disagreement between participants. Recommendations will be discussed  for norm collection procedures more broadly in order that the validity of such norms can be improved. In particular, clear guidelines are required for data cleaning in order that the reliability of such norms is maximised and to facilitate replication across studies.

Direct lexical control of eye movements in Chinese reading: Evidence from the co-registration of EEG and eye tracking

The direct-lexical-control hypothesis stipulates that some aspect of a word’s processing determines the duration of the fixation on that word and/or the next. Although direct lexical control is incorporated into most current models of eye-movement control in reading, conclusive empirical evidence supporting this hypothesis is lacking to date. In this article, we report the results of an eye-tracking experiment using the boundary paradigm in which native speakers of Chinese read sentences in which target words were either high- or low-frequency and preceded by a valid or invalid preview. Eye movements were co-registered with electroencephalography, allowing standard analyses of eye-movement measures, divergence point analyses of fixation-duration distributions, and fixated-related potentials on the target words. These analyses collectively provide strong behavioral and neural evidence of early lexical processing and thus strong support for the direct-lexical-control hypothesis.