Word recognition is a fundamental aspect of language processing, linking auditory input to meaning. Previous research has described word recognition as a competitive process, but hasn't identified the underlying dimensions of this competition across individuals. This study addresses this gap by investigating these dimensions in a diverse population of CI users with heterogeneous backgrounds and audiological profiles, and in a lifespan sample of individuals without hearing loss. Given the increasing prevalence of hearing loss and its association with cognitive decline, understanding the cognitive mechanisms of speech understanding is crucial. CI users, who receive profoundly altered auditory input, provide a unique population for examining the variability and underlying dimensions of word recognition mechanisms. This study uses the Visual World Paradigm (VWP), an eye-tracking technique, to analyze real-time lexical competition during isolated word recognition. The VWP allows for a precise characterization of the process of lexical competition, isolating lexical processing from broader sentence or discourse processes. Understanding the dimensions of lexical competition is essential for creating more inclusive theories of language processing, informing clinical assessments and treatment of speech and language disorders, and clarifying the relationship between hearing loss and dementia. The current study aims to identify these underlying dimensions, investigate their predictors (deafness onset, age, CI experience, auditory skills), and determine their relationship to speech perception outcomes.

Decades of cognitive science research have focused on the mechanisms of word recognition, particularly the process of lexical competition where listeners select the target word from a set of similar-sounding competitors. Studies have shown that this process involves a dynamic competition between words matching the input, with the competition continuously winnowed until a single word is selected. However, previous studies using the VWP to investigate these mechanisms were limited by small sample sizes, which constrained the analysis of individual differences and the ability to link these differences to outcomes. Small-scale studies involving CI users had suggested two processing profiles: Sustained Activation, where competition doesn’t fully resolve, and Wait-and-See, where lexical access is delayed. These profiles have also been observed in other hearing-impaired populations and in normal-hearing listeners in challenging conditions, suggesting generalized adaptation mechanisms. This paper aimed to test the underlying nature of these dimensions with a larger and more diverse sample, investigating their relationships to various factors and their impact on overall speech perception performance.

This study involved 101 CI users with varying demographics and device configurations (Table 1), including pre-lingually and post-lingually deaf individuals. A lifespan sample of 107 normal-hearing (NH) listeners from a prior study was also included for comparison. Peripheral auditory factors were characterized using pure-tone audiometry, spectral ripple tasks, and envelope tasks. Speech perception outcomes were assessed using CNC words in quiet, AzBio sentences in noise, and the Speech subscale of the SSQ questionnaire. The VWP was employed, participants listened to spoken words while viewing pictures of the target word and various competitors (onset competitors, rhymes, unrelated words). Eye movements were recorded using an eye tracker (SR Research Eyelink 1000) to track fixation patterns. The data from the VWP, which capture the real-time dynamics of lexical competition were characterized by fitting non-linear functions to the time course of fixations to different word classes (target, cohort, unrelated). Principal Component Analysis (PCA) was used on thirteen parameters from these curves to identify underlying dimensions of processing. Regressions were performed to predict these dimensions from demographic factors (age, CI experience, sex, deafness onset), and peripheral auditory variables (CI use, FAH, PTA, spectral and temporal fidelity). Additional regressions analyzed the relationship between the VWP indices and speech perception outcomes, controlling for demographic and auditory factors. A non-linguistic VWP (nlVWP) was used to assess general visual cognitive processes and rule out non-linguistic contributions to the VWP results. Imputation was used to handle missing data in the auditory fidelity measures. Model selection techniques using the Akaike Information Criteria and commonality analyses were used to prevent overfitting.

PCA of the VWP data revealed three primary dimensions of real-time lexical competition: Wait-and-See, Sustained-Activation, and Slow-Activation-Rate, which accounted for 62.1% of the variance. Wait-and-See, characterized by delayed target fixations and reduced cohort competition, was strongly associated with deafness onset and interacted with CI experience. Sustained-Activation, with slower activation rates and persistent competitor fixations, was primarily driven by age. Slow-Activation-Rate, reflecting slower overall activation, showed an interaction between CI experience and deafness onset and a quadratic effect of age. These dimensions were largely independent of each other. The NH listeners showed lower values on Wait-and-See and Sustained-Activation, but similar variation in Slow-Activation-Rate. Regressions revealed that distinct combinations of demographic and audiological factors predicted each dimension. The three VWP dimensions predicted speech perception outcomes (CNC words, AzBio sentences, SSQ), even after controlling for peripheral auditory factors. Higher Wait-and-See and Sustained-Activation scores consistently correlated with poorer outcomes. nlVWP results indicated that the first three PCs from the VWP were not strongly related to general visual-cognitive factors.

This study provides evidence for three independent dimensions underlying real-time lexical competition in both CI and NH listeners. Wait-and-See and Sustained-Activation are not merely ends of a single continuum; they are distinct processes with different predictors and relationships to outcomes. The finding that these dimensions are present in NH listeners as well suggests that hearing loss may exacerbate pre-existing individual differences rather than creating entirely new processing profiles. The strong relationship between these dimensions and speech perception outcomes highlights the importance of understanding cognitive processing mechanisms beyond peripheral auditory factors. While prior research has emphasized the importance of peripheral auditory factors, this study reveals that cognitive mechanisms related to lexical competition play a significant role, explaining considerable variance in outcomes not accounted for by auditory fidelity alone. The negative correlation between the lexical competition indices and outcomes suggests that these profiles are not adaptive strategies, but may reflect challenges in processing. Further research is needed to investigate the factors that drive these processing profiles. This might include exploring pre-implantation language abilities, cortical structure, and listening experience.

This study identified three dimensions of real-time lexical competition—Wait-and-See, Sustained-Activation, and Slow-Activation-Rate—that significantly contribute to explaining individual differences in speech perception outcomes in CI users and NH listeners. These dimensions are independent and are predicted by distinct sets of factors. The findings emphasize the importance of considering cognitive mechanisms in addition to peripheral auditory factors when assessing and treating hearing impairment. Future research should explore the factors driving these individual differences in lexical competition and investigate potential interventions to improve language processing efficiency.

This study focused solely on isolated word recognition, which may not fully capture the complexities of real-world speech perception in sentence or discourse contexts. While the sample of CI users was large and diverse, it was not fully representative of the broader CI user population in terms of racial diversity. The inclusion of a separate NH sample, while informative, presents a challenge in direct comparisons due to differences in participant selection and recruitment. The use of imputation to handle missing data might introduce some bias. Finally, this study only explored one aspect of word recognition.