Models Of Language Production Research Articles

The role of animacy in language production: evidence from bare noun naming.

According to Levelt's language production model, to name an object, speakers must first conceptualise and lexicalise the object before its name can be articulated. Conceptualisation is conducted through the semantic network that exists at the conceptual level, with the highly activated concept(s) activating lexical items at the lemma level, that is, lexicalisation. So far, research focused mostly on semantic categories (i.e., semantic interference) but less so on animacy-a concept that is correlated with semantic categories. To investigate the role of this semantic feature in language production, we conducted a picture-word interference study in Mandarin Chinese, varying animacy congruency and controlling for classifier congruency while recording behavioural and electrophysiological responses. We observed an animacy interference effect together with a larger N400 component for animacy-incongruent versus congruent picture-word pairs, suggesting animacy-congruent concepts may be in closer proximity and hence lead to a stronger spreading of activation relative to animacy-incongruent concepts. Furthermore, a larger P600 component was observed for classifier-incongruent versus classifier-congruent picture-word pairs, suggesting syntactically driven processing of classifiers at the lemma level.

Engineering conversation: Understanding the control requirements of language production in monologue and dialogue

Both artificial and biological systems are faced with the challenge of noisy and uncertain estimation of the state of the world, in contexts where feedback is often delayed. This challenge also applies to the processes of language production and comprehension, both when they take place in isolation (e.g., in monologue or solo reading) and when they are combined as is the case in dialogue. Crucially, we argue, dialogue brings with it some unique challenges. In this paper, we describe three such challenges within the general framework of control theory, drawing analogies to mechanical and biological systems where possible: (1) the need to distinguish between self- and other-generated utterances; (2) the need to adjust the amount of advance planning (i.e., the degree to which planning precedes articulation) flexibly to achieve timely turn-taking; (3) the need to track changing conversational goals. We show that message-to-sound models of language production (i.e., those that cover the whole process from message generation to articulation) tend to implement fairly simple control architectures. However, we argue that more sophisticated control architectures are necessary to build language production models that can account for both monologue and dialogue.

Four dimensions of naturalistic language production in aphasia after stroke.

There is a rich tradition of research on the neuroanatomical correlates of spoken language production in aphasia using constrained tasks (e.g., picture naming), which offer controlled insights into the distinct processes that govern speech and language (i.e., lexical-semantic access, morphosyntactic construction, phonological encoding, speech motor programming/execution). Yet these tasks do not necessarily reflect everyday language use. In contrast, naturalistic language production (also referred to as connected speech or discourse) more closely approximates typical processing demands, requiring the dynamic integration of all aspects of speech and language. The brain bases of naturalistic language production remain relatively unknown, however, in part because of the difficulty in deriving features that are salient, quantifiable, and interpretable relative to both speech-language processes and the extant literature. The present cross-sectional observational study seeks to address these challenges by leveraging a validated and comprehensive auditory-perceptual measurement system that yields four explanatory dimensions of performance-Paraphasia (misselection of words and sounds), Logopenia (paucity of words), Agrammatism (grammatical omissions), and Motor speech (impaired speech motor programming/execution). We used this system to characterize naturalistic language production in a large and representative sample of individuals with acute post-stroke aphasia (n = 118). Scores on each of the four dimensions were correlated with lesion metrics, and multivariate associations among the dimensions and brain regions were then explored. Our findings revealed distinct yet overlapping neuroanatomical correlates throughout the left-hemisphere language network. Paraphasia and Logopenia were associated primarily with posterior regions, spanning both dorsal and ventral streams, which are critical for lexical-semantic access and phonological encoding. In contrast, Agrammatism and Motor speech were associated primarily with anterior regions of the dorsal stream that are involved in morphosyntactic construction and speech motor planning/execution respectively. Collectively, we view these results as constituting a brain-behavior model of naturalistic language production in aphasia, aligning with both historical and contemporary accounts of the neurobiology of spoken language production.

Disfluency across the lifespan: an individual differences investigation

ABSTRACT This study had two research objectives. The first was to examine age-related differences in the fluency of speech outputs, as prior research contains conflicting findings concerning whether older adults produce more disfluency than younger adults. The second was to examine cognitive individual differences, and their relationship with the production of disfluency. One hundred and fifty-four adults completed a story re-telling task, and a battery of cognitive measures. Results showed that younger adults produced more um’s and fewer repetitions. For individual differences, results showed that inhibition and set shifting were related to the production of repetitions, and inhibition and working memory were related to uh production. Our results provide clarification about mixed findings with respect age and disfluency production. The individual differences provide clarification on theoretical arguments for disfluent speech in aging (e.g. Inhibition Deficit Hypothesis), and also sheds light on the role of executive functions in models of language production.

Bilingualism Influences How Articulation Enhances Verbal Encoding.

Learning information may benefit from movement: Items that are spoken aloud are more accurately remembered than items that are silently read (the production effect). Candidate mechanisms for this phenomenon suggest that speaking may enrich or improve the feature content of memory traces, yet research suggests that prior language skill also plays a role. Recent work showed a larger production effect in bilinguals for words in their second language (L2) compared to their first language (L1), potentially suggesting that bilinguals engage different or additional linguistic features when speaking L2 compared to L1 words. The current study examined whether the increased L2 production effect reduces for L2 and L1 pseudowords, which may similarly engage mainly phonological features. German (L1)-English (L2) bilinguals first read (out loud or silently) and subsequently recognized German or English words or pseudowords following German or English phonology. The production effect increased for L2 compared to L1 items and for words compared to pseudowords. Modest evidence suggested L2-L1 similarity in production effect scores for pseudowords, but different L2-L1 scores for words. Integrating feature models of memory with models of bilingual language production, we propose that speaking an L2 may engage more extensive and diverse linguistic features than an L1.

An Information-Theoretic Account of Availability Effects in Language Production.

I present a computational-level model of language production in terms of a combination of information theory and control theory in which words are chosen incrementally in order to maximize communicative value subject to an information-theoretic capacity constraint. The theory generally predicts a tradeoff between ease of production and communicative accuracy. I apply the theory to two cases of apparent availability effects in language production, in which words are selected on the basis of their accessibility to a speaker who has not yet perfectly planned the rest of the utterance. Using corpus data on English relative clause complementizer dropping and experimental data on Mandarin noun classifier choice, I show that the theory reproduces the observed phenomena, providing an alternative account to Uniform Information Density and a promising general model of language production which is tightly linked to emerging theories in computational neuroscience.

Information-theoretic principles in incremental language production

I apply a recently emerging perspective on the complexity of action selection, the rate-distortion theory of control, to provide a computational-level model of errors and difficulties in human language production, which is grounded in information theory and control theory. Language production is cast as the sequential selection of actions to achieve a communicative goal subject to a capacity constraint on cognitive control. In a series of calculations, simulations, corpus analyses, and comparisons to experimental data, I show that the model directly predicts some of the major known qualitative and quantitative phenomena in language production, including semantic interference and predictability effects in word choice; accessibility-based ("easy-first") production preferences in word order alternations; and the existence and distribution of disfluencies including filled pauses, corrections, and false starts. I connect the rate-distortion view to existing models of human language production, to probabilistic models of semantics and pragmatics, and to proposals for controlled language generation in the machine learning and reinforcement learning literature.

Dimensions of Explanatory Value in NLP Models

AbstractPerformance on a dataset is often regarded as the key criterion for assessing NLP models. I argue for a broader perspective, which emphasizes scientific explanation. I draw on a long tradition in the philosophy of science, and on the Bayesian approach to assessing scientific theories, to argue for a plurality of criteria for assessing NLP models. To illustrate these ideas, I compare some recent models of language production with each other. I conclude by asking what it would mean for institutional policies if the NLP community took these ideas onboard.

Language switching when writing: The role of phonological and orthographic overlap.

While language switching of bilinguals has been investigated extensively in the spoken domain, there has been little research on switching while writing. The factors that impact written language switching may differ from those that impact language switching while speaking. Thus, the study's goal was to test to what extent phonological and/or orthographic overlap impacts written language switching. In four experiments (NExp.1 = 34; NExp. 2 = 57; NExp. 3 = 39; NExp. 4 = 39), German-English bilinguals completed a cued language switching task where responses had to be typed. To-be-named translation-equivalent concepts were selected to be similar phonologically, orthographically or neither. Participants switching between languages while writing was facilitated by both phonological and orthographic overlap. Maximum orthographic overlap between translation-equivalent words with dissimilar pronunciations facilitated switching to the extent that no switch costs could be observed. These results imply that overlapping orthography can strongly facilitate written language switching and that orthography's role should be considered more thoroughly in models of bilingual language production.

Capturing writers’ typing while visually attending the emerging text: a methodological approach

AbstractKnowledge about writers’ eye movements and their effects on the writing process, and its product—the finally edited text—is still limited. Previous research has demonstrated that there are differences between reading texts written by someone else and reading one’s own emerging text and that writers frequently look back into their own texts (Torrance et al. in Psychol Res Psychologische Forschung 80(5):729–743, 2016. https://doi.org/10.1007/s00426-015-0683-8). For handwriting, Alamargot et al. (Writing and cognition: research and applications. Elsevier Science, pp 13–29, 2007) found support that these lookbacks could occur in parallel with transcription, but to our knowledge this type of parallel processing has not been explored further, and definitely not in the context of computer writing. Considering that language production models are moving away from previous sequential or serial models (e.g., Levelt in Speaking from intentions to articulation. MIT Press, 1989) towards models in which linguistic processes can operate in parallel (Olive in J Writ Res, 2014. https://doi.org/10.17239/jowr-2014.06.02.4), this is slightly surprising. In the present paper, we introduce a methodological approach to examine writers’ parallel processing in which we take our point of departure in visual attention rather than in the keystrokes. Capitalizing on New ScriptLog’s feature to link gaze with typing across different functional units in the writing task, we introduce and describe a method to capture and examine sequences of typing during fixations, outline how these can be examined in relation to each other, and test our approach by exploring typing during fixations in a text composition task with 14 competent adult writers.

Moving away from lexicalism in psycho- and neuro-linguistics

In standard models of language production or comprehension, the elements which are retrieved from memory and combined into a syntactic structure are “lemmas” or “lexical items.” Such models implicitly take a “lexicalist” approach, which assumes that lexical items store meaning, syntax, and form together, that syntactic and lexical processes are distinct, and that syntactic structure does not extend below the word level. Across the last several decades, linguistic research examining a typologically diverse set of languages has provided strong evidence against this approach. These findings suggest that syntactic processes apply both above and below the “word” level, and that both meaning and form are partially determined by the syntactic context. This has significant implications for psychological and neurological models of language processing as well as for the way that we understand different types of aphasia and other language disorders. As a consequence of the lexicalist assumptions of these models, many kinds of sentences that speakers produce and comprehend—in a variety of languages, including English—are challenging for them to account for. Here we focus on language production as a case study. In order to move away from lexicalism in psycho- and neuro-linguistics, it is not enough to simply update the syntactic representations of words or phrases; the processing algorithms involved in language production are constrained by the lexicalist representations that they operate on, and thus also need to be reimagined. We provide an overview of the arguments against lexicalism, discuss how lexicalist assumptions are represented in models of language production, and examine the types of phenomena that they struggle to account for as a consequence. We also outline what a non-lexicalist alternative might look like, as a model that does not rely on a lemma representation, but instead represents that knowledge as separate mappings between (a) meaning and syntax and (b) syntax and form, with a single integrated stage for the retrieval and assembly of syntactic structure. By moving away from lexicalist assumptions, this kind of model provides better cross-linguistic coverage and aligns better with contemporary syntactic theory.

A chimpanzee by any other name: The contributions of utterance context and information density on word choice

An important feature of language production is the flexibility of lexical selection; producers could refer to an animal as chimpanzee, chimp, ape, she, and so on. Thus, a key question for psycholinguistic research is how and why producers make the lexical selections that they do. Information theoretic approaches have argued that producers regulate the uncertainty of the utterance for comprehenders, for example using longer words like chimpanzee if their messages are likely to be misunderstood, and shorter ones like chimp when the message is easy to understand. In this work, we test for the relative contributions of the information theoretic approach and an approach more aligned with psycholinguistic models of language production. We examine the effect on lexical selection of whole utterance-level factors that we take as a proxy for register or style in message-driven production accounts. Using a modern machine learning-oriented approach, we show that for both naturalistic stimuli and real-world corpora, producers prefer words to be longer in systematically different contexts, independent of the specific message they are trying to convey. We do not find evidence for regulation of uncertainty, as in information theoretic approaches. We offer suggestions for modification of the standard psycholinguistic production approach that emphasizes the need for the field to specify how message formulation influences lexical choice in multiword utterances.

Workload and cognitive architecture in translation

Abstract This article endeavours to sketch out a consistent merger of models of language production and bilingualism that are currently used in the fields of Translation and Interpreting and of psycholinguistics. The workload resource of Cognitive Verbal Processing and the processing components that it serves are found to be of particular interest in this interdisciplinary perspective. Known psycholinguistic effects that involve workload (i.e., a function of activation energy) are outlined with particular emphasis on their possible relevance in translation processing. This suggests a number of potential lines of research for both fields.

Inter-individual variability in dorsal stream dynamics during word production.

The current standard model of language production involves a sensorimotor dorsal stream connecting areas in the temporo-parietal junction with those in the inferior frontal gyrus and lateral premotor cortex. These regions have been linked to various aspects of word production such as phonological processing or articulatory programming, primarily through neuropsychological and functional imaging group studies. Most if not all the theoretical descriptions of this model imply that the same network should be identifiable across individual speakers. We tested this hypothesis by quantifying the variability of activation observed across individuals within each dorsal stream anatomical region. This estimate was based on electrical activity recorded directly from the cerebral cortex with millisecond accuracy in awake epileptic patients clinically implanted with intracerebral depth electrodes for pre-surgical diagnosis. Each region's activity was quantified using two different metrics-intra-cerebral evoked related potentials and high gamma activity-at the level of the group, the individual and the recording contact. The two metrics show simultaneous activation of parietal and frontal regions during a picture naming task, in line with models that posit interactive processing during word retrieval. They also reveal different levels of between-patient variability across brain regions, except in core auditory and motor regions. The independence and non-uniformity of cortical activity estimated through the two metrics push the current model towards sub-second and sub-region explorations focused on individualized language speech production. Several hypotheses are considered for this within-region heterogeneity.

Understanding the Message and Formulation of Fake Online Reviews: A Language-production Model Perspective

Consumers have become ever more reliant on online reviews. Therefore, fake reviews have also become increasingly rampant and eroded online review platforms’ credibility. Previous literature suggests that particular linguistic styles can manifest in fake reviews with reference to the varying stages of the language-production process. Drawing on the language-production model as our theoretical foundation, we examine the psycholinguistic styles of fake online reviews at the message and formulation level. We performed a computational linguistic analysis on 66,940 reviews from Yelp. Our results suggest that fake reviews align more with deceptive writing in terms of the message-level variables such as length and psychological (affective, cognitive, social, and perceptual) cues. Interestingly, we found that they align less with deceptive writing in terms of the formulation-level variables such as readability, pronouns, and part-of-speech tags, which may be due to the fake review writers’ conscious attempt to follow the language styles that genuine reviews adopt.

The grammatical representation of expletive negation

Expletive negation refers to constructions where a negator in the complement of certain lexical items does not change the polarity of the complement proposition. Jin & Koenig (2021) show that expletive negation occurs rather widely in languages of the world and in very similar environments. They propose a language production model of why such apparently illogical uses of negation arise in language after language. But their study does not address the grammatical status and representation of expletive negation. In this paper, we argue that expletive negation is part of the lexical knowledge speakers have of their language and that the negator in expletive negation constructions contributes a negation to a non-at-issue content associated with expletive negation triggers. We provide a Lexical Resource Semantics analysis of how triggers combine in a non-standard manner with the standard semantic content of their complements: the negation (and in some cases an additional modal operator) of the content of their complement is part of the trigger’s non-at-issue content while the scope of the negation is an argument of the trigger’s MAIN content. Finally, we suggest that the expletive use of the French negator ne includes a lexical constraint that requires it to modify a verb that reverse selects for an expletive negation trigger.

Similarities between Cognitive Models of Language Production and Everyday Functioning: Implications for Development of Interventions for Functional Difficulties.

The loss of the ability to independently complete activities of daily living, such as meal preparation and household chores, is a defining characteristic of clinical dementia; however, minor difficulties in completing everyday activities emerge in the mild cognitive impairment stage, and even healthy older adults exhibit subtle functional difficulties compared to younger adults. These functional difficulties are associated with an array of negative outcomes, including decreased quality of life, higher costs of care, and increased frustration, depression, caregiver burden, and institutionalization. While cognitive models have been proposed to explain the functional deficits seen in dementia and subtler forms of cognitive decline, in contrast to other cognitive disorders such as aphasia, there are essentially no theoretically motivated interventions to address difficulties in everyday functioning. Proposed models of functional impairment share features with cognitive processing models of language, including hierarchical organization of representations and interactive, spreading activation; thus, an examination of aphasia interventions has the potential to inform the development of theoretically motivated interventions for everyday activities. This review first addresses the shared characteristics of cognitive models of everyday function and language, with a focus on linguistic production. Next, we will present aphasia interventions that target single-word production, sentence production, short-term memory, and semantics, and discuss their implications for everyday functioning interventions. We conclude with a discussion of limitations of the language-everyday functioning comparison as well as areas of future research.

A unified neurocomputational bilateral model of spoken language production in healthy participants and recovery in poststroke aphasia

Understanding the processes underlying normal, impaired, and recovered language performance has been a long-standing goal for cognitive and clinical neuroscience. Many verbally described hypotheses about language lateralization and recovery have been generated. However, they have not been considered within a single, unified, and implemented computational framework, and the literatures on healthy participants and patients are largely separated. These investigations also span different types of data, including behavioral results and functional MRI brain activations, which augment the challenge for any unified theory. Consequently, many key issues, apparent contradictions, and puzzles remain to be solved. We developed a neurocomputational, bilateral pathway model of spoken language production, designed to provide a unified framework to simulate different types of data from healthy participants and aphasic patients. The model encapsulates key computational principles (differential computational capacity, emergent division of labor across pathways, experience-dependent plasticity-related recovery) and provides an explanation for the bilateral yet asymmetric lateralization of language in healthy participants, chronic aphasia after left rather than right hemisphere lesions, and the basis of partial recovery in patients. The model provides a formal basis for understanding the relationship between behavioral performance and brain activation. The unified model is consistent with the degeneracy and variable neurodisplacement theories of language recovery, and adds computational insights to these hypotheses regarding the neural machinery underlying language processing and plasticity-related recovery following damage.

Lexical selection in bimodal bilinguals: ERP evidence from picture-word interference

ABSTRACT The picture word interference (PWI) paradigm and ERPs were used to investigate whether lexical selection in deaf and hearing ASL-English bilinguals occurs via lexical competition or whether the response exclusion hypothesis (REH) for PWI effects is supported. The REH predicts that semantic interference should not occur for bimodal bilinguals because sign and word responses do not compete within an output buffer. Bimodal bilinguals named pictures in ASL, preceded by either a translation equivalent, semantically-related, or unrelated English written word. In both the translation and semantically-related conditions bimodal bilinguals showed facilitation effects: reduced RTs and N400 amplitudes for related compared to unrelated prime conditions. We also observed an unexpected focal left anterior positivity that was stronger in the translation condition, which we speculate may be due to articulatory priming. Overall, the results support the REH and models of bilingual language production that assume lexical selection occurs without competition between languages.

Cognitive and pragmatic factors in language production: Evidence from source-goal motion events

In order to talk about an event they see in the world, speakers have to build a conceptual representation of that event and generate a message that selects the pragmatically appropriate (e.g., informative) parts of that event that they want to talk about. To further understand the relationship between a speaker's conceptual representations and the pragmatic factors that influence message generation, this work investigates the extent to which different aspects of an event could be affected by pragmatic constraints. We focus specifically on source-goal motion events (e.g., a butterfly flying from a lamppost to a chair) because the conceptual structure of these events is well-understood, but the role that those representations play in message generation is yet unclear. In Experiment 1, we manipulated the pragmatic status of the source (e.g., the lamppost) – in particular, whether starting points of motion were or were not already known to an addressee. We found that sources were mentioned significantly more in the latter case, where they provided new, previously unknown information to the addressee. In Experiment 2, we investigated whether the same pragmatic factors could affect goals, or endpoints of motion events (e.g., the chair), in the same way; results showed that they could not. We conclude that conceptually peripheral elements of an event (i.e., sources) are more susceptible to communicative factors than those elements which are conceptually privileged (i.e., goals). We consider the implications of our findings for the relationship between event cognition and pragmatics and discuss how theories of event cognition can be integrated into current models of language production. We also discuss the implications of our work for open issues in the domain of event cognition.