Ardi Roelofs

A theory of lexical access in speech production

Preparing words in speech production is normally a fast and accurate process. We generate them two or three per second in fluent conversation; and overtly naming a clear picture of an object can easily be initiated within 600 msec after picture onset. The underlying process, however, is exceedingly complex. The theory reviewed in this target article analyzes this process as staged and feedforward. After a first stage of conceptual preparation, word generation proceeds through lexical selection, morphological and phonological encoding, phonetic encoding, and articulation itself. In addition, the speaker exerts some degree of output control, by monitoring of self-produced internal and overt speech. The core of the theory, ranging from lexical selection to the initiation of phonetic encoding, is captured in a computational model, called weaver ++. Both the theory and the computational model have been developed in interaction with reaction time experiments, particularly in picture naming or related word production paradigms, with the aim of accounting for the real-time processing in normal word production. A comprehensive review of theory, model, and experiments is presented. The model can handle some of the main observations in the domain of speech errors (the major empirical domain for most other theories of lexical access), and the theory opens new ways of approaching the cerebral organization of speech production by way of high-temporal-resolution imaging.

A spreading-activation theory of lemma retrieval in speaking

This paper presents a spreading-activation theory of conceptually driven lemma retrieval--the first stage of lexical access in speaking, where lexical items specified with respect to meaning and syntactic properties are activated and selected. The mental lexicon is conceived of as a network consisting of concept, lemma, and word-form nodes and labelled links, with each lexical concept represented as an independent node. A lemma is retrieved by enhancing the activation level of the node representing the to-be-verbalized concept. This activation then spreads towards the lemma level, and the highest activated lemma node is selected. The theory resolves questions such as the hypernym problem (Levelt, 1989). Furthermore, a computer model that implements the theory is shown to be able to account for many basic findings on the time course of object naming, object categorization, and word categorization in the picture-word interference paradigm. In addition, non-trivial predictions regarding the time course of semantic facilitation for hypernyms, hyponyms, and cohyponyms are experimentally tested, and shown to be valid.

The WEAVER model of word-form encoding in speech production

Lexical access in speaking consists of two major steps: lemma retrieval and word-form encoding. In Roelofs (Roelofs, A. 1992a. Cognition 42. 107-142; Roelofs. A. 1993. Cognition 47, 59-87.), I described a model of lemma retrieval. The present paper extends this work by presenting a comprehensive model of the second access step, word-form encoding. The model is called WEAVER (Word-form Encoding by Activation and VERification). Unlike other models of word-form generation, WEAVER is able to provide accounts of response time data, particularly from the picture-word interference paradigm and the implicit priming paradigm. Its key features are (1) retrieval by spreading activation, (2) verification of activated information by a production rule, (3) a rightward incremental construction of phonological representations using a principle of active syllabification, syllables are constructed on the fly rather than stored with lexical items, (4) active competitive selection of syllabic motor programs using a mathematical formalism that generates response times and (5) the association of phonological speech errors with the selection of syllabic motor programs due to the failure of verification.

Goal-referenced selection of verbal action: Modeling attentional control in the Stroop task.

This article presents a new account of the color-word Stroop phenomenon (J. R. Stroop, 1935) based on an implemented model of word production, WEAVER++ (W. J. M. Levelt, A. Roelofs, & A. S. Meyer, 1999b; A. Roelofs, 1992, 1997c). Stroop effects are claimed to arise from processing interactions within the language-production architecture and explicit goal-referenced control. WEAVER++ successfully simulates 16 classic data sets, mostly taken from the review by C. M. MacLeod (1991), including incongruency, congruency, reverse-Stroop, response-set, semantic-gradient, time-course, stimulus, spatial, multiple-task, manual, bilingual, training, age, and pathological effects. Three new experiments tested the account against alternative explanations. It is shown that WEAVER++ offers a more satisfactory account of the data than other models.

Role of inhibition in language switching: evidence from event-related brain potentials in overt picture naming.

How are bilinguals able to switch from one language to another? The prevailing inhibition hypothesis takes larger reaction-time (RT) costs for switching to the first language (L1) than to the second language (L2) as evidence for suppression of the non-target language. Switch cost asymmetries can alternatively be explained by an L1-repeat-benefit, assuming selective absence of language competition in repeating L1. To test the latter hypothesis, RTs and event-related brain potentials (ERPs) were recorded when unbalanced Dutch-English bilinguals switched between overt picture naming in L1 and L2. Preparation time (short versus long cue-stimulus intervals) modulated the degree to which top-down inhibitory control biased language competition, as indexed by the N2 component of the ERP. RT switch costs were asymmetrical on short intervals and symmetrical on long intervals. Preparation interval effects were observed in all conditions except for L1 repeat trials. This pattern was present both for RT and N2 data, thereby supporting the L1-repeat-benefit hypothesis. These results suggest that inhibition is not necessary, but can modulate the efficiency of language switching.

Metrical structure in planning the production of spoken words

According to most models of speech production, the planning of spoken words involves the independent retrieval of segments and metrical frames followed by segment-to-frame association. In some models, the metrical frame includes a specification of the number and ordering of consonants and vowels, but in the word-form encoding by activation and verification (WEAVER) model (A. Roelofs, 1997), the frame specifies only the stress pattern across syllables. In 6 implicit priming experiments, on each trial, participants produced 1 word out of a small set as quickly as possible. In homogeneous sets, the response words shared word-initial segments, whereas in heterogeneous sets, they did not. Priming effects from shared segments depended on all response words having the same number of syllables and stress pattem, but not on their having the same number of consonants and vowels. No priming occurred when the response words had only the same metrical frame but shared no segments. Computer simulations demonstrated that WEAVER accounts for the findings. Most theories of word production assume that the phonological representations constructed in planning utterances include separate representations of the segmental content of words and of their metrical properties, such as their syllable structure and stress pattern. This view is compatible with current linguistic theory, which allocates segmental and metrical information on separate representational tiers (e.g., Goldsmith, 1990; Kenstowicz, 1994). In addition, psycholinguistic models often contend that during speech planning, the metrical and segmental tiers are first retrieved, or generated, independently of each other and later combined (see Levelt, 1989, for an overview). Most of the evidence for this view comes from analyses of speech errors. The argument runs roughly as follows (see, for instance, Meyer, 1992, for a more extensive discussion). Speakers often commit sound errors, in which the intended and the actual utterance differ in a speech fragment smaller than a complete morpheme. Usually, these fragments correspond to individual segments or, less often, to clusters of two adjacent segments. This shows that, during speech planning, stored form representations are decomposed into their segments. Misplaced segments typically move from their target positions to corresponding positions in new syllables, for instance from one syllable onset (the prevocalic part of a

Word length effects in object naming: The role of a response criterion

According to Levelt, Roelofs, and Meyer (1999) speakers generate the phonological and phonetic representations of successive syllables of a word in sequence and only begin to speak after having fully planned at least one complete phonological word. Therefore, speech onset latencies should be longer for long than for short words. We tested this prediction in four experiments in which Dutch participants named or categorized objects with monosyllabic or disyllabic names. Experiment 1 yielded a length effect on production latencies when objects with long and short names were tested in separate blocks, but not when they were mixed. Experiment 2 showed that the length effect was not due to a difference in the ease of object recognition. Experiment 3 replicated the results of Experiment 1 using a within-participants design. In Experiment 4, the long and short target words appeared in a phrasal context. In addition to the speech onset latencies, we obtained the viewing times for the target objects, which have been shown to depend on the time necessary to plan the form of the target names. We found word length effects for both dependent variables, but only when objects with short and long names were presented in separate blocks. We argue that in pure and mixed blocks speakers used different response deadlines, which they tried to meet by either generating the motor programs for one syllable or for all syllables of the word before speech onset. Computer simulations using WEAVER++ support this view.

Testing a non-decompositional theory of lemma retrieval in speaking: Retrieval of verbs ☆

Theories of lexical access in speaking differ in whether they assume that words are accessed in a conceptually decomposed or non-decomposed way. In this paper, two experiments are reported that test the non-decompositional theory and computer model proposed by Roelofs (1992a). Subjects had to name pictured actions using verbs and ignore distractor verbs superimposed on the pictures. According to the theory, semantic inhibition should be obtained from distractor cohyponym verbs that are the names of other pictures in the experiment. By contrast, semantic facilitation should be obtained from hyponyms of the target verbs. Both predictions were empirically confirmed, both qualitatively and quantitatively. These findings support the proposed non-decompositional theory and computer model. Furthermore, they refute a recent attempt to deal with a class of retrieval problems within the decompositional framework. Bierwisch and Schreuder (1992) propose to solve the hyperonym problem (Levelt, 1989) by an inhibitory channel in the mental lexicon between a word and its hyperonyms. This predicts semantic inhibition by hyponyms, instead of the observed facilitation.

Anterior cingulate cortex activity can be independent of response conflict in Stroop-like tasks

Cognitive control includes the ability to formulate goals and plans of action and to follow these while facing distraction. Previous neuroimaging studies have shown that the presence of conflicting response alternatives in Stroop-like tasks increases activity in dorsal anterior cingulate cortex (ACC), suggesting that the ACC is involved in cognitive control. However, the exact nature of ACC function is still under debate. The prevailing conflict detection hypothesis maintains that the ACC is involved in performance monitoring. According to this view, ACC activity reflects the detection of response conflict and acts as a signal that engages regulative processes subserved by lateral prefrontal brain regions. Here, we provide evidence from functional MRI that challenges this view and favors an alternative view, according to which the ACC has a role in regulation itself. Using an arrow–word Stroop task, subjects responded to incongruent, congruent, and neutral stimuli. A critical prediction made by the conflict detection hypothesis is that ACC activity should be increased only when conflicting response alternatives are present. Our data show that ACC responses are larger for neutral than for congruent stimuli, in the absence of response conflict. This result demonstrates the engagement of the ACC in regulation itself. A computational model of Stroop-like performance instantiating a version of the regulative hypothesis is shown to account for our findings.

Anticipatory Activity in Anterior Cingulate Cortex Can Be Independent of Conflict and Error Likelihood

Previous studies have found no agreement on whether anticipatory activity in the anterior cingulate cortex (ACC) reflects upcoming conflict, error likelihood, or actual control adjustments. Using event-related functional magnetic resonance imaging, we investigated the nature of preparatory activity in the ACC. Informative cues told the participants whether an upcoming target would or would not involve conflict in a Stroop-like task. Uninformative cues provided no such information. Behavioral responses were faster after informative than after uninformative cues, indicating cue-based adjustments in control. ACC activity was larger after informative than uninformative cues, as would be expected if the ACC is involved in anticipatory control. Importantly, this activation in the ACC was observed for informative cues even when the information conveyed by the cue was that the upcoming target evokes no response conflict and has low error likelihood. This finding demonstrates that the ACC is involved in anticipatory control processes independent of upcoming response conflict or error likelihood. Moreover, the response of the ACC to the target stimuli was critically dependent on whether the cue was informative or not. ACC activity differed among target conditions after uninformative cues only, indicating ACC involvement in actual control adjustments. Together, these findings argue strongly for a role of the ACC in anticipatory control independent of anticipated conflict and error likelihood, and also show that such control can eliminate conflict-related ACC activity during target processing. Models of frontal cortex conflict-detection and conflict-resolution mechanisms require modification to include consideration of these anticipatory control properties of the ACC.