Linguistic prediction can be defined as pre-activation of upcoming linguistic information based on certain cues in the language (e.g., Huettig, 2015). Studies with monolingual speakers show that individuals predict upcoming linguistic elements during sentence processing (e.g., Altmann & Kamide, 1999; Kuperberg & Jaeger, 2016; Yoshida et al., 2013). In a seminal eye-tracking study, Altmann and Kamide (1999) discovered that monolingual individuals moved their eye gaze towards specific objects (e.g., ‘cake’) before they were mentioned when listening to constraining verbs (e.g., ‘eat’), but not when listening to neutral verbs (e.g., ‘move’). Linguistic prediction behavior has been found to be less consistent in studies with bilingual individuals performing in their non-native language, with studies showing that bilingual people have limited ability to predict in their second language (e.g., Ito et al., 2018). Moreover, mixed results have also been reported for neurotypical older monolingual adults (e.g., Huettig & Janse, 2016; Maquate & Knoeferle, 2021). In the current study we utilized an eye-tracking paradigm to investigate whether bilingual younger and older neurotypical individuals predict upcoming nouns in sentences that include constraining verbs (e.g., an edible object such as ‘pizza’ after a verb such as ‘eat’) in a similar manner as monolingual individuals do, and if they do so both in their first language (L1) and in their second language (L2).
1.1 Prediction studies with bilingual people
The emerging literature on prediction abilities in bilingual individuals – those who use more than one language in their lives – reveals mixed results. For instance, English-speaking L2 learners of Spanish did not predict upcoming nouns based on gender-marked articles in the same way as L1 speakers of Spanish did (Lew-Williams & Fernald, 2010). In contrast, in another study, experienced L2 learners of Spanish predicted nouns similar to, albeit slightly slower than, L1 speakers (Grüter et al., 2012). Additional studies showed that Chinese learners of English processed prosodic cues (contrastive pitch accent) but did not then use them to predict upcoming referents in their L2 (Perdomo & Kaan, 2021), whereas native speakers of English did. Similarly, English speaking L2 learners of German did not predict upcoming information based on case markings in German, while L1 German speakers did. In turn, the L2 learners predicted upcoming information based on semantic cues, indicating that linguistic prediction may be different for L1 and L2 speakers (Hopp, 2015), especially for people who are not highly proficient in their L2.
Indeed, other studies found that proficient bilingual speakers’ prediction abilities were similar to those of their monolingual counterparts. For example, Dutch-English proficient bilingual participants predicted upcoming nouns based on constraining verbs in both their languages, but were slower in both languages compared to English monolingual participants (Dijkgraaf et al., 2017). Moreover, another group of Dutch-English proficient bilinguals predicted semantically related nouns when hearing constraining verbs in both languages, but the effect size was larger in Dutch (L1) than in English (L2) (Dijkgraaf et al., 2019). English L2 speakers of various L1 backgrounds performed similarly to English L1 speakers in an eye-tracking experiment using orally presented sentences with constraining verbs, but were slower when cognitive load was manipulated (Ito et al., 2018, see below).
Bilingual speakers were found to also predict upcoming information on the basis of implicit causality, for instance predicting pronoun referents (Contemori & Dussias, 2019; Kim & Grüter, 2021). Although both studies found that bilingual participants predicted referents, they found differences in the speed of prediction. Kim and Grüter (2021) found that Korean learners of English were slower at predicting in their L2, whereas Contemori and Dussias (2019), who studied early Spanish-English bilinguals, found no difference in speed of prediction compared to monolingual English speakers. They argued that the participants in their study were able to predict referents similarly to monolingual English speakers, since the bilingual participants in their sample were early bilinguals (Contemori & Dussias, 2019).
Prediction ability has been associated with lexical knowledge (e.g., Borovsky et al., 2012; Federmeier et al., 2002; Rommers et al., 2015). That is, a stable vocabulary representation may be needed for listeners to be able to predict, which may explain why less proficient L2 users do not predict upcoming information (e.g., Pickering & Gambi, 2018). For example, Rommers, Meyer and Huettig (2015) found that high receptive vocabulary and high category fluency scores were associated with higher prediction ability in their Dutch speaking participants. Nevertheless, Corps et al. (2023) concluded from their eye-tracking experiment with advanced English learners (of various L1 backgrounds) that L2 proficiency did not mediate prediction.
A recent review article (Schlenter, 2023) focused on a number of studies that found later prediction onset times and weaker effects for processing in L2 compared to L1. Schlenter (2023) brings up the possibility that differential cue weighting (e.g., prosodic, morphological) across languages can explain different prediction patterns. A potential way to explore language specific effects may be to compare words that are cognates (share meaning and form) and non-cognate translation equivalents. Few prediction studies thus far examined this aspect with eye-tracking experiments, and findings are mixed (Martinez-Garcia, 2019). Most of such investigations involve nouns only; one exception is Van Assche et al. (2013), who found a smaller and later cognate effect in sentence processing using verb stimuli than generally found with noun stimuli. As well, the consistency and reliability of the cues in the specific language or for the specific group of speakers may account for the prediction pattern observed (Schlenter, 2023), which would suggest a role of L2 proficiency and exposure.
Other, non-linguistic variables that have been associated with prediction ability include speed of processing and working memory (e.g., Huettig & Janse, 2016; Ito et al., 2018; Li & Qu, 2024). Huettig and Janse (2016) collected eye-tracking data from 105 native speakers of Dutch between the ages of 32 and 77. In their regression analyses they demonstrated that working memory (WM), as measured by non-word repetition, digit span backwards, and Corsi block tasks, as well as processing speed, as measured by digit-symbol substitution and letter comparison tasks, predicted anticipatory eye movements based on article gender cues. Ito et al. (2018) examined eye-tracking patterns of L1 and L2 users when they listened to sentences and viewed target and non-target images compared to when they performed a memory task in addition to the listening task. The authors found that whereas the participants who performed the task in their L2 had similar prediction behavior to the L1 participants in the listening only condition, where their eye movements were delayed in the cognitively taxing condition (Ito et al., 2018). Consistently, Li and Qu (2024) found that Chinese speaking participants with higher verbal WM spans showed earlier prediction times than participants with lower WM spans. In addition, age, which correlated with both speed and WM, emerged as a significant predictor, as will be discussed next.
1.2 Prediction studies with older adults
Prediction ability has not been studied extensively in older age and results to date are mixed (Maquate & Knoeferle, 2021). Older age has been associated with compromised cognitive abilities such as processing speed and WM (e.g., Salthouse, 1991), the same abilities that have been found to modulate prediction behavior. However, research has demonstrated that older adults may use top-down processing more than do younger adults, perhaps to compensate for reduced sensory processing, and thus demonstrate better prediction ability (Federmeier et al., 2010). Indeed, Huettig and Janse (2016) demonstrated that older age, when dissociated from cognitive abilities, was associated with more anticipatory eye movements. Age-related cognitive changes were hypothesized to explain the differential context effects found for older compared to younger adults in an eye-tracking study that manipulated the speaker's emotional facial expression (Maquate & Knoeferle, 2021). In another study, a self-paced reading paradigm with younger and older monolingual adults revealed that older age correlated with faster responses related to bigrams with higher transition probability (i.e., the probability of occurrence in a given context) (McConnell & Blumenthal-Dramé, 2021).
The association of better lexical abilities and older age fits the approach promoted by Ramscar, Baayen and colleagues (Ramscar et al., 2014; Ramscar & Baayen, 2015). They suggest that given that learning continues throughout the lifespan, the increasing amount of information needs to be considered when accounting for observed slower or less efficient processing. Evidence for less efficient processing in older age can be seen in the findings of reduced prediction ability associated with older age (e.g., Federmeier et al., 2002).
1.3 The current study
As the brief review above suggests, mixed results have been reported for the prediction behavior of older adults and of bilingual adults. Larger vocabulary and richer language experience, associated with high proficiency and with older age, can contribute to better prediction ability. In contrast, reduced processing speed and memory capacity – also associated with older age and with processing in a non-native language – may contribute to reduced prediction ability. Consistently, language proficiency emerged as a critical variable for prediction ability in L2 users. It is likely that some of the mixed results reported for prediction behavior in L2 may be explained by considering L2 proficiency of the participants enrolled. Few studies compared directly prediction performance in the two languages of bilingual individuals, rather, in most cases, processing in L2 of bilingual groups has been compared to processing of monolingual speakers of that language. Furthermore, the role of lexical similarity between the two languages has not been addressed much in the literature, and no studies have examined prediction performance in older bilingual adults.
The aim of the present study was therefore to examine the prediction ability of younger and older sequential bilingual individuals during processing in their L1 and in their L2, by means of the visual world paradigm, which is particularly well suited to study online language processing (Huettig, 2015). We asked the following research questions:
1) Do people predict upcoming language (specifically, nouns following constraining and neutral, non-constraining verbs) as they hear sentences in their L1 and in their L2?
2) How does aging affect the prediction processing in L1 and in L2?
3) What is the influence of language proficiency, cognitive abilities, and word characteristics (i.e., cognate status) on the prediction process?
On the basis of previous research, we expected that bilingual people who were highly proficient in both languages would predict nouns following constraining verbs in both their L1 and L2, and that prediction abilities will be evident in younger and older participants. Furthermore, we predicted that L2 proficiency will be associated with better prediction performance, as will better cognitive abilities, such as working memory.