Within- and cross-language syntactic priming in bilingual adults, and increasingly in children, offers compelling evidence for shared representations across languages. However, little is known about whether such shared representations extend beyond syntax to other structural levels, or how they develop during childhood. We addressed this limitation in three experiments with Polish-English bilingual children aged 7;11–12;4 (N = 96), and adults (N = 45). We investigated bidirectional priming both across languages (children and adults) and within languages (children only) for transitive structures that contrasted in whether they had shared syntactic structure (passive primes/passive targets) or shared thematic role orders (OVS primes/passive targets) in English and Polish (Nitems = 48/language). We found significant priming of syntactic structures (speakers produced more passives after passive than active primes) in both languages and significant priming of thematic role order (speakers produced more passives after OVS than active primes) from Polish to English. For children, these effects were stronger within languages than across languages, and cross-linguistic priming was only significant when English, the societal language, was the target language. For adults, these effects did not differ by target language, and priming was stronger than in children. These results provide an important picture of the developmental trajectory of shared representations at the level of syntactic structure and thematic roles: bilingual children acquire robust within-language representations for each language, while shared representations develop later and appear entrenched within adult bilingual speakers. The emergence of shared representations reflects asymmetrical language experience, manifesting in stronger priming for the children's societal language than for their heritage language.
The mechanisms and representations underlying word production are studied in two literatures. Psycholinguistic models describe the processes going from the activation of a concept to the selection of abstract phonetic units, while speech motor control models describe how these units are translated into articulatory movements. Bridging the gap between these two literatures, we examine whether and how the timing of higher-level processes impacts articulation. We quantify the evidence for and against the claims that (1) the ease with which the abstract word form is encoded influences the acoustic duration of the articulated word, and (2) the duration of planning processes prior to articulation onset (as reflected in response latencies) predicts acoustic duration. We analyze three datasets in which participants named pictures with distractor words unrelated to the target words and distractors overlapping phonologically with the target words. Bayesian analyses provide evidence that this manipulation impacts naming latencies but does not impact articulation durations. Moreover, latencies predict acoustic durations when participants are pressured to respond quickly in a subset of trials, such that for trials with longer latencies, words are articulated with shorter durations. These findings provide new constraints for the architecture of the language production system.
What makes some stimuli more memorable than others? Recent work has shown that image memorability is predicted by the magnitude of population responses in both monkey inferotemporal cortex and convolutional neural networks, suggesting that stimuli that more strongly activate distributed feature representations are more likely to be remembered. However, it remains unclear whether this effect is specific to the visual domain or reflects a more general property of distributed representations. Here, we show that representational magnitude predicts memorability beyond the visual domain: the effect not only replicates for images in an independent dataset, but extends to lexical memory. Across three large-scale word datasets, word embedding magnitude reliably predicts recognition memorability, independent of word frequency, valence, or length. However, the effect did not extend to a recently published voice memorability dataset, possibly reflecting distinct features driving auditory memory. Together, the cross-domain generalization suggests that the representational magnitude effect is a general property of distributed representations. Memorability, on this view, is inherent to encoding: stimuli that activate more features, and activate them more strongly – whether in brains or artificial neural networks – leave a larger representational footprint, and therefore a more lasting memory trace.
Skilled readers recognize words efficiently despite wide variation in their visual form (e.g., table, TABLE, table, table). The present experiments examined how case invariance emerges in the earliest moments of word recognition using the masked priming technique to isolate early stages of processing. Specifically, we tested whether case alternation disrupts access to abstract orthographic representations or merely modulates early perceptual encoding. In two masked priming lexical decision experiments, we compared matched-case and alternating-case identity primes (e.g., TABLE–TABLE, tAbLe–TABLE) with unrelated baselines, while manipulating prime–target visual overlap (equal vs. reduced prime size). In the two experiments, we found a sizable overall masked identity priming effect that was slightly larger for matched-case identity primes than for alternating-case primes. Diffusion-model analyses showed that priming effects were expressed mainly as reductions in nondecision time (Ter parameter), indicating facilitation at early encoding stages. These results indicate that early orthographic processing is largely, but not completely, case-invariant; alternating-case primes rapidly activate abstract letter identities, while a small residual benefit remains for matched-case primes, constraining models of visual word recognition.
In four experiments, we investigated the negation-induced forgetting effect. Participants first saw a video of an apartment showing several objects and their properties. In a subsequent verification task, participants verified true or false affirmative or negative statements by responding with yes or no. After a distractor task, participants completed a final recognition test for the objects. Forgetting rates were higher for objects that in the verification phase had been mentioned in statements requiring a no-response compared to objects that had been mentioned in statements requiring a yes-response (Experiment 1). A comparable effect was obtained for objects that in the verification phase had been mentioned in negated vs. affirmative verification statements (Experiment 2). To test whether this effect depends on the semantic scope of negation, Experiment 3 placed the critical object outside the scope of the negation operator by embedding it in a non-restrictive relative clause. Contrary to the prediction of a strictly scope-based account, this manipulation did not provide conclusive evidence that negation-induced forgetting disappears when the object falls outside the semantic scope of negation. Finally, the detrimental effects on memory of no-responses and sentence negation did not combine additively (Experiment 4). Together, these findings suggest that negation-induced forgetting is not specific to the response particle no but generalizes to sentence polarity more broadly. Furthermore, the absence of additive effects suggests that the effects of the response particle no and sentence negation are unlikely to reflect two independent inhibitory mechanisms. At the same time, our findings indicate that semantic scope alone does not fully account for negation-induced forgetting, leaving open whether the underlying mechanism reflects scope-insensitive conceptual inhibition, the influence of more surface-level linguistic representations, or a combination of both.
Words denoting animate beings are better recalled than words denoting inanimate objects. Here we examine the contributions of item-specific and relational processing to the animacy effect in free recall. According to the item-specific and relational processing framework, free recall particularly benefits when both item-specific processing, which supports the discrimination among retrieval candidates, and relational processing, which supports the generation of retrieval candidates, have occurred during encoding. Applied to the animacy effect, animate words may be better recalled than inanimate words because animate words receive both item-specific and relational processing automatically, whereas inanimate words receive both forms of processing to a lesser extent. It follows from this assumption that the animacy effect should be reduced to the extent that external factors such as the structure of the list of to-be-recalled words and the orienting task promote both forms of processing for inanimate words. We tested this assumption systematically across four experiments. In Experiments 1 and 2, in which a non-salient list structure that promotes item-specific processing was used, an orienting task that promotes relational processing significantly reduced the animacy effect. In Experiments 3 and 4, in which a salient list structure that promotes relational processing was used, an orienting task that promotes item-specific processing also reduced the animacy effect, but the reduction was very small. Promoting relational processing reduced the animacy effect more than promoting item-specific processing, consistent with the assumption of a comparatively stronger contribution of relational processing to the animacy effect in free recall.
The influence of a talker's first language (L1) on the production of their second (L2) can complicate the task of word recognition. Nevertheless, previous research has shown that listeners demonstrate remarkable speed and plasticity in adapting to L2-accented talkers. Listeners are also able to generalize this learning to novel L2-accented talkers. There are two competing explanations for these talker-independent adaptation effects: one argues that increasing variation during exposure facilitates generalization to a novel test talker (exposure-to-variability hypothesis), while the other argues that increasing acoustic-phonetic similarity between the exposure and test talkers facilitates generalization (similarity-based hypothesis). We conducted three experiments to compare the effects of exposure variability and exposure-test similarity on generalization. During exposure, monolingual English participants performed an auditory lexical decision task with multisyllabic English tokens spoken by three Spanish-accented talkers. After exposure, we tested performance on a novel Spanish-accented talker with a cross-modal matching task (Experiments 1 and 2) or a primed cross-modal lexical decision task (Experiment 3). Exposure variability increased both identity and competitor priming, consistent with a relaxation of phonetic categorization criteria. Exposure-test similarity decreased competitor priming without changing identity priming, while exposure uniformity increased identity priming without changing competitor priming, providing some evidence of phonetic recalibration in these conditions. Our results provide equivocal support for the similarity-based hypothesis and against the exposure-to-variability hypothesis.
Iconicity in spoken language refers to a perceived resemblance between the form of a word and its meaning. Here we examined emotional iconicity: the relationship between the sound of a word and its emotional arousal. We first examined which language sounds are associated with calmness vs. excitement by collecting emotional arousal ratings of 5,457 nonwords (Experiment 1). This revealed a pattern in which high pitch sounds (e.g., voiceless stops) were associated with excitement while low pitch sounds (e.g., sonorants) were associated with calmness. We used these ratings to identify words whose sounds were iconically congruent (e.g., "shock", containing exciting sounds and with an exciting meaning), incongruent (e.g., "alarm", containing calming sounds and with an exciting meaning) neutral. In two experiments (Experiment 2 and 3) we found that congruent words were responded to faster, and that incongruent words were responded to accurately, on a semantic decision task ("Is this word calming or exciting?"). Congruent words were also more likely to be recalled in Experiment 2, but not Experiment 3 after removing taboo items. However, we found that emotionally iconic words were more likely to be recalled in a mega study dataset of 1,551 items. We also examined whether individual differences in the experience of emotion moderated the effects we observed. Together these results suggest that iconicity is important factor in language processing.
Existing models of bilingualism, models that are primarily based on alphabetic-script bilinguals, widely assume that cross-language phonological similarity facilitates processing of cognates. However, it remains unclear to what degree this assumption generalizes to logographic-script bilinguals. While an effect of phonological similarity has been observed for alphabetic-script bilinguals in both word naming and lexical decision tasks, previous studies on Chinese-Japanese bilinguals have reported that cross-language phonological similarity facilitates cognate processing for L2 Japanese readers in word naming but not in lexical decision. The present research examined the mechanism underlying the effect of cross-language phonological similarity in logographic-script bilinguals' cognate processing. In Experiment 1, we affirmed that the phonological facilitation effect in naming is a reliable phenomenon. In Experiments 2 and 3, we demonstrated that no parallel facilitation occurs in lexical decision tasks. These contrasting findings suggest that the phonological facilitation does not arise during lexical access, which is engaged in both tasks, but rather arises during word production. Experiments 4 and 5 were designed to further identify the locus of this phonological facilitation, adopting the theoretical framework of the WEAVER++ model. Specifically, we contrasted two stages of word production that differ in their sensitivity to phonological similarity. If facilitation is sensitive to overall similarity, the locus would be in phonological retrieval; if it is sensitive only to word-beginning similarity, the locus would be in phonological encoding. Our findings supported the former locus. We propose an extended version of the WEAVER++ model to account for L2 word production in Chinese-Japanese bilinguals.
There is now robust evidence that individuals can direct their attention towards particularly valuable information in working memory. However, the cognitive mechanisms underlying such effects are still under debate. One possibility is that participants preferentially encode high value information, creating a stronger representation relative to other information presented. We tested this hypothesis in two experiments. Participants completed a visual working memory task, which involved remembering the color of shapes for a brief period of time. During the encoding phase participants’ eye movements were recorded to assess fixation behaviors. Participants were then presented with one shape and asked to report the color as accurately as possible on a color wheel. In Experiment 1, participants were either told that one particular item was more valuable (i.e., worth more notional points than the rest), or that all items were of equal value. In Experiment 2, each item differed in its value in a graded manner (i.e., 1-2-3-4 points). Across both experiments, participants exhibited the lowest recall error for the highest value information. Moreover, participants spent significantly longer fixating on high value items during the encoding phase relative to other information presented. Further, the extent to which participants fixated on an item during encoding partially mediated the effect of value on recall error. Taken together, these experiments demonstrate that participants vary their eye movements during encoding as a function of value, and suggest that value-driven prioritization effects in working memory are partially driven by such behaviors.
Whether the predictability effect in language comprehension reflects the pre-activation of specific lexical candidates or the probabilistic inference of text interpretation remains debated. These two accounts make testable predictions regarding the functional form of this effect, positing either a linear or a logarithmic relationship to the word probability. The present study asked whether the above two mechanisms can co-exist, which could be reflected in a composite functional form of the predictability effect. Based on six large-scale Chinese and English eye-tracking natural reading datasets and word surprisal estimates from 14 large language models for each language, we compared linear, logarithmic, additive, and weighted functional forms. Our results revealed that an additive functional form comprising both linear and logarithmic components provides a better fit. It provides behavioral evidence for the co-existence of pre-activation and probabilistic inference in natural reading. This co-existence was reliably detected in early eye-movement measures across languages, and the contribution of pre-activation was more dominant and temporally sustained in Chinese than in English. This cross-language difference likely reflects the demands of word segmentation in Chinese text. We proposed a dual-process framework to explain how the pre-activation and probabilistic inference operate compatibly in language comprehension.
Part of successful communication involves recognising the purpose of, or the intentions underlying, what speakers choose to say. Often, such pragmatic inferences are studied with an emphasis on informativity. The present work however moves beyond the types of inferences typically studied in prior work and instead investigates inferences from more naturalistic utterances, specifically those whose triviality may invite addressees to reason about why a speaker would have made such a discourse contribution. We present four studies (total N = 777) using offline and online methods to investigate how and when listeners derive relevance inferences from trivial utterances. We manipulate speaker knowledge, speaker style, and linguistic properties of the utterances to show that, even in the absence of explicit emphasis cues, trivial utterances such as “the library walls are blue” are likely to be understood as conveying more than what is stated explicitly (e.g. that the walls used to be a different colour), and that these inferences are more likely to arise when produced by a speaker who is knowledgeable about the situation and who does not typically talk a lot. Our results suggest that comprehenders have pervasive expectations of cooperativity which, when seemingly violated by a speaker’s trivial utterance, prompt reasoning about a speaker’s motivation for speaking to determine how the communicated content is relevant. We then turn to the processing costs of computing triviality-driven inferences and find evidence that there may be a cost to deriving relevance inferences. These findings extend previous work on inferencing, which typically targets specific classes of words that give rise to inferences and demonstrates that broader, systematic inferencing that can arise when addressees reason about speaker goals even in the absence of cues to pragmatic enrichment.
It is well established that semantics influence verbal working memory (WM). The present study addressed two unresolved questions: Could semantic strategies help compensate for the age-related deficit in WM? And how do they influence the maintenance of bindings? Across two experiments using a complex span task, opportunities to rely on semantic strategies were manipulated in two ways: passively, through the intrinsic features of the to-be-remembered stimuli (word concreteness), and actively, through the disruption of phonological strategies (articulatory suppression in Experiment 1 and a phonological distracting task in Experiment 2). To assess memory performance for both item identity and item-item bindings, young and older adults completed immediate and delayed recognition tests in which they had to reconstruct previously presented word pairs. In the second experiment, participants also reported their memory strategies on a trial-by-trial basis. Except for the articulatory suppression manipulation, which failed to promote the use of semantic strategies, the main results revealed that both passive and active semantic strategies enhanced WM performance for both item identity and bindings. Moreover, the findings emphasized that encouraging passive semantic strategies was more beneficial for older adults than for young adults, reducing age-related differences in WM, whereas encouraging active semantic strategies did not modulate these differences. Thus, distinguishing semantic strategies by their cognitive cost appears essential for understanding cognitive aging in verbal WM. These findings are discussed in light of the potential cognitive mechanisms underlying the observed semantic effects in verbal WM.
Competing hypotheses on the coordination of breathing and language structure in speech originate from vertical versus horizontal perspectives on the speech-language relationship. The vertical hypothesis assumes that inhalations occur where long pauses are licensed by language factors (e.g., boundary strength) with the probability of inhalation increasing with pause duration. The horizontal hypothesis assumes that inhalation is tied to the cognitive demands of language planning such that the probability of inhalation increases with these demands. These predictions were tested in spontaneous and rote speech. In Experiment 1, mini-narratives were elicited from 40 speakers under two planning conditions: one where causal links between pictured events were either obvious (baseline) or not (planning-intensive). Pause duration increased with planning demands; inhalation frequency did not. When pauses with inhalation were removed from the analysis, pause duration only varied with linguistic boundary strength in the planning-intensive condition. Experiment 2 was designed to test the for a link between the cognitive demands of planning and inhalation. Twenty four speakers produced 12 repetition of a sequence of 20 long and short sentences grouped into several thematic chunks. No matter the prosodification of target sentences during elicitation, the probability of inhalation was higher before long target sentences than before short ones; it was also higher at the onset of a thematic group of sentences than within the group. Taken together, the study findings suggest that inhalations are coordinated with language structure based on a link between breathing and cognitive processing.
Memory for an earlier event can be impaired by later studying a related event (retroactive interference). However, when later events cue retrievals of earlier events that also enable integrative encoding, memory for the earlier event can be enhanced (retroactive facilitation). Two experiments using continuous A-B, A-C paired-associate tasks examined retroactive memory effects across lags and retention intervals. We indexed study-phase retrieval and integrative encoding by examining associations between awareness of changes and memory for first responses. Participants studied A-B, A-C, and retention-interval-matched A-B pairs and indicated when they detected response changes. The lag between A-B and A-C and the retention interval between A-C and test trials varied factorially across 4, 16, and 48 intervening items. Cued-recall test formats differed: Experiment 1 prompted direct retrieval of first responses, with participants reporting recollected changes. Experiment 2 used modified-modified free recall in which participants reported responses as they came to mind and indicated whether they were the first, second, or only responses; the first two decisions indicated recollection of changes. Both experiments showed the greatest retroactive facilitation at the longest retention interval and intermediate lag. Retroactive facilitation co-occurred with and was statistically associated with detection and recollection of changes, suggesting these processes contributed to the observed effect. Negatively curvilinear, non-monotonic lag functions occurred for A-B, A-C recall at the longest retention interval and were associated with detected changes, especially those not recollected. These findings may be accounted for by a memory-for-change framework with a role for contextual variability that determines study-phase retrieval success.
The probability of a word in context, as captured by large language models, is predictive of both behavioral and neural measures of human language processing. Intuitively, language models that are better at next-word prediction might better model predictability effects in human language comprehension. Yet recent work suggests that language models can become too good at next-word prediction to model reading time, implying that the aspects of human comprehension indexed by reading time do not track perfectly with predictability from language statistics alone. However, it is unknown whether this decoupling is true of reading time only, or whether it is intrinsic to online measures of comprehension more generally. To address this question, we turn to another robust and well-studied measure of online processing, the N400 component of the event-related brain potential. We compare how a language model’s size, number of training tokens, and performance on natural language benchmarks correlate with its ability to predict both reading time and N400 amplitude. Based on an analysis of 4 reading time datasets and 9 N400 datasets, we replicate past results for reading time, but find that larger language models, models that are trained on more data, and models that perform better at next-word prediction and other more complex natural language tasks are better able to predict N400 amplitude. We interpret this difference between the N400 and reading time measures as potentially revealing the comparative importance of semantic prediction in the neurocognitive processes indexed by the N400.
The current study critically examined a common theoretical view in which there exists a distinct working memory (WM) system that not only temporarily stores information but also acts as an attentional controller (the Working Memory—Attentional Control account). A central prediction of this account is that an increase in cognitive load should take up some of the system’s capacity for attentional control and thereby render short-term storage more vulnerable to distraction by task-irrelevant stimuli. The present results directly contradict this prediction: The requirement to engage in an additional concurrent articulation task during short-term serial recall abolished not only the classical changing-state irrelevant sound effect but also attentional capture by an auditory deviant (Experiments 1a and 1b). The deviation effect was also abolished under concurrent articulation in a task involving little if any serial rehearsal (the missing-item task), as indicated independently by the fact this task was immune to a detrimental main effect of concurrent articulation and to a changing-state effect (Experiment 2). We argue that concurrent articulation not only specifically impedes serial rehearsal in the context of serial recall but also imposes a nonspecific cognitive load (regardless of task). However, rather than drain capacity from a distinct WM system, such increased load boosts task engagement which in turn shields performance against forms of distraction that are due to task-disengagement. The task-engagement view may also provide an alternative account of numerous findings that are typically cited in support of the Working Memory—Attentional Control view.
