A neurally plausible parallel distributed processing model of event-related potential word reading data

The Parallel Distributed Processing (PDP) framework has significant potential for producing models of cognitive tasks that approximate how the brain performs the same tasks. To date, however, there has been relatively little contact between PDP modeling and data from cognitive neuroscience. In an attempt to advance the relationship between explicit, computational models and physiological data collected during the performance of cognitive tasks, we developed a PDP model of visual word recognition which simulates key results from the ERP reading literature, while simultaneously being able to successfully perform lexical decision—a benchmark task for reading models. Simulations reveal that the model’s success depends on the implementation of several neurally plausible features in its architecture which are sufficiently domain-general to be relevant to cognitive modeling more generally.     

Electrophysiological evidence for parallel and serial processing during visual search

Event-related potentials were recorded from young adults during a visual search task in order to evaluate parallel and serial models of visual processing in the context of Treisman's feature integration theory. Parallel and serial search strategies were produced by the use of feature-present and feature-absent targets, respectively. In the feature-absent condition, the slopes of the functions relating reaction time and latency of the P3 component to set size were essentially identical, indicating that the longer reaction times observed for larger set sizes can be accounted for solely by changes in stimulus identification and classification time, rather than changes in post-perceptual processing stages. In addition, the amplitude of the P3 wave on target-present trials in this condition increased with set size and was greater when the preceding trial contained a target, whereas P3 activity was minimal on target-absent trials. These effects are consistent with the serial self-terminating search model and appear to contradict parallel processing accounts of attention-demanding visual search performance, at least for a subset of search paradigms. Differences in ERP scalp distributions further suggested that different physiological processes are utilized for the detection of feature presence and absence.     

From parallel to serial processing: a computational study of visual search

A novel computational model of a preattentive system performing visual search is presented. The model processes displays of lines, reproduced from Wolfe, Friedman-Hill, Stewart, and O'Connell's (1992) and Treisman and Sato's (1990) visual-search experiments. The response times measured in these experiments suggest that some of the displays are searched serially, whereas others are scanned in parallel. Our neural network model operates in two phases. First, the visual displays are compressed via standard methods (principal component analysis), to overcome assumed biological capacity limitations. Second, the compressed representations are further processed to identify a target in the display. The model succeeds in fast detection of targets in experimentally labeled parallel displays, but fails with serial ones. Analysis of the compressed internal representations reveals that compressed parallel displays contain global information that enables instantaneous target detection. However, in representations of serial displays, this global information is obscure, and hence, a target detection system should resort to a serial, attentional scan of local features across the display. Our analysis provides a numerical criterion that is strongly correlated with the experimental response time slopes and enables us to reformulate Duncan and Humphreys's (1989) search surface, using precise quantitative measures. Our findings provide further insight into the important debate concerning the dichotomous versus continuous views of parallel/serial visual search.     

Individual differences in emotion word processing: A diffusion model analysis

The exploratory study investigated individual differences in implicit processing of emotional words in a lexical decision task. A processing advantage for positive words was observed, and differences between happy and fear-related words in response times were predicted by individual differences in specific variables of emotion processing: Whereas more pronounced goal-directed behavior was related to a specific slowdown in processing of fear-related words, the rate of spontaneous eye blinks (indexing brain dopamine levels) was associated with a processing advantage of happy words. Estimating diffusion model parameters revealed that the drift rate (rate of information accumulation) captures unique variance of processing differences between happy and fear-related words, with highest drift rates observed for happy words. Overall emotion recognition ability predicted individual differences in drift rates between happy and fear-related words. The findings emphasize that a significant amount of variance in emotion processing is explained by individual differences in behavioral data.     

Establishing the loci of serial and parallel effects in syntactic processing

Recent experimental work has demonstrated a robust effect of syntactic context on lexical decision times. Wright and Garrett (1984) report that targets belonging to syntactic categories required by the context produce faster RTs than either ungrammatical or optional continuations of the context string. The present study reports that structurally ambiguous contexts produced faster recognition times for syntactic categories required by only the putative nonpreferred reading of the ambiguities tested. These findings are argued to favor a parallel processing model in which syntactic representations are ranked according to structural criteria and shunted in a serial fashion to subsequent processors.     

A test of parallel versus serial processing applied to memory retrieval

Townsend has identified necessary and sufficient conditions that reaction time distributions meet if they are produced by two underlying processes that are parallel and within-stage independent. This paper describes an attempt to use one of these conditions to assess the assumption of independent parallel memory retrieval in Anderson's ACT theory. In the experimental test, subjects studied profession-location pairs, with each profession and location included in two pairs. One pair had a high frequency of presentation and the other had low. When cued with a profession (or location), subjects responded with the first of the two studied locations (or professions) that came to mind. The reaction times to these responses were used to test the ACT assumption of independent parallel search. The problems encountered in trying to apply Townsend's abstract test to a concrete situation included the need to consider mixtures and convolutions of distributions and empirical, rather than theoretical, distributions. These difficulties are discussed at length and some formal analyses concerning their effect on the test are given. Despite these problems, we were able to use Townsend's test and the results of the experiment support the within-stage independent parallel assumption of ACT.     

Evidence from mutual masking for serial processing of tachistoscopic letter patterns.

This experiment deomonstrates serial processing of tachistoscopic patterns when all potential artifacts are eliminated. Two nonsense arrays of six letters were tachistoscopically exposed successively at the same positions. Ss identified more letters from the temporally first string on the left of fixation and more from the second on the right, indicating left-to-right serial processing. Variations among Ss in the location of the crossover in curves indicate individual differences in processing time. Therefore, this study demonstrates definitively that tachistoscopic performance at various positions can reflect a sequential left-to-right processing of information, possibly at different rates.     

Arithmetic word problem solving: Evidence for the construction of a mental model

In the first experiment reported here, adults were given an unexpected task of problem recognition after a resolution task. During the recognition task, participants were presented with the original problems, inconsistent problems that had never been solved, and paraphrases, which respected the relational structure of the original problems but not their exact wording. More precisely, paraphrases were constructed by inversing the terms and the linguistic expressions in the original problems. Whereas the literal form of paraphrastic problems bore the least resemblance to original problems, paraphrastic problems were associated to higher recognition rates than inconsistent problems. A second experiment ruled out the interpretation that this result was due to a mere remembering of the exact values used in the problem text. Taken together, these results provide evidence that a non-propositional representation is built by individuals to solve arithmetic word problems and suggest that a mental model is constructed (Johnson-Laird, 1983).     

Testing the associative-link hypothesis in immediate serial recall: Evidence from word frequency and word imageability effects

Two immediate serial recall experiments were conducted to test the associative-link hypothesis (Stuart & Hulme, 2000). We manipulated interitem association by varying the intralist latent semantic analysis (LSA) cosines in our 7-item study word lists, each of which consists of high- or low-frequency words in Experiment 1 and high- or low-imageability words in Experiment 2. Whether item recall performance was scored by a serial-recall or free-recall criterion, we found main effects of interitem association, word imageability, and word frequency. The effect of interitem association also interacted with the word frequency effect, but not with the word imageability effect. The LSA-cosinexword frequency interaction occurred in the recency, but not primacy, portion of the serial position curve. The present findings set explanatory boundaries for the associative-link hypothesis and we argue that both item- and associative-based mechanisms are necessary to account for the word frequency effect in immediate serial recall.     

Testing the associative-link hypothesis in immediate serial recall: Evidence from word frequency and word imageability effects

Two immediate serial recall experiments were conducted to test the associative-link hypothesis (Stuart & Hulme, 2000 Stuart, G. and Hulme, C. 2000. The effects of word co-occurrence on short-term memory: Associative-links in long-term memory affect short-term memory performance. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26: 796–802. [Crossref], [PubMed], [Web of Science ®] , [Google Scholar]). We manipulated interitem association by varying the intralist latent semantic analysis (LSA) cosines in our 7-item study word lists, each of which consists of high- or low-frequency words in Experiment 1 and high- or low-imageability words in Experiment 2. Whether item recall performance was scored by a serial-recall or free-recall criterion, we found main effects of interitem association, word imageability, and word frequency. The effect of interitem association also interacted with the word frequency effect, but not with the word imageability effect. The LSA-cosine×word frequency interaction occurred in the recency, but not primacy, portion of the serial position curve. The present findings set explanatory boundaries for the associative-link hypothesis and we argue that both item- and associative-based mechanisms are necessary to account for the word frequency effect in immediate serial recall.     

Serial processing in reading aloud: no challenge for a parallel model

K. Rastle and M. Coltheart (1999) challenged parallel models of reading by showing that the cost of irregularity in low-frequency exception words was modulated by the position of the irregularity in the word. This position-of-irregularity effect was taken as strong evidence of serial processing in reading. This article refutes Rastle and Coltheart's theoretical conclusions in 3 ways: First, a parallel model, the connectionist dual process model (M. Zorzi, G. Houghton, & B. Butterworth, 1998b), produces a position-of-irregularity effect. Second, the supposed serial effect can be reduced to a position-specific grapheme-phoneme consistency effect. Third, the position-of-irregularity effect vanishes when the experimental data are reanalyzed using grapheme-phoneme consistency as the covariate. This demonstration has broader implications for studies aiming at adjudicating between models: Strong inferences should be avoided until the computational models are actually tested.     

Extrafoveal processing of objects in a naming task: Evidence from word probe experiments

In two experiments, we investigated the processing of extrafoveal objects in a double-object naming task. On most trials, participants named two objects; but on some trials, the objects were replaced shortly after trial onset by a written word probe, which participants had to name instead of the objects. In Experiment 1, the word was presented in the same location as the left object either 150 or 350 msec after trial onset and was either phonologically related or unrelated to that object name. Phonological facilitation was observed at the later but not at the earlier SOA. In Experiment 2, the word was either phonologically related or unrelated to the right object and was presented 150 msec after the speaker had begun to inspect that object. In contrast with Experiment 1, phonological facilitation was found at this early SOA, demonstrating that the speakers had begun to process the right object prior to fixation.     

Binocular information summation and the serial processing model

An experiment was performed to ascertain whether a particular member of the class of sequential processing models (Estes & Taylor, 1964; Townsend, 1966) should be further developed in terms of the contributions of information from the separate eyes. Using the detection paradigm (Estes & Taylor, 1964), nine Ss were each run 576 trials under each of three viewing conditions: (a) monocular left, (b) monocular right, and (c) binocular, after four days of practice and calibration. The serial processing model was used to make predictions for three possible cases of binocular information summation: (a) complete independence of the monocular channels, (b) partial independence of the monocular channels and (c) complete dependence in the monocular channels. Complete dependence provided the best fit to the data with a possible stress on use of a “best” eye, but a marginal level of significance was obtained between a simple average of monocular performance and binocular performance with transformed scores. Thus, although there seems to be little or no information summation in terms of the present model of multi-symbol perception, follow-up experiments were suggested to further delineate monocular-binocular relationships in the detection paradigm.     

Evidence for serial coercion: a time course analysis using the visual-world paradigm

Metonymic verbs like start or enjoy often occur with artifact-denoting complements (e.g., The artist started the picture) although semantically they require event-denoting complements (e.g., The artist started painting the picture). In case of artifact-denoting objects, the complement is assumed to be type shifted (or coerced) into an event to conform to the verb's semantic restrictions. Psycholinguistic research has provided evidence for this kind of enriched composition: readers experience processing difficulty when faced with metonymic constructions compared to non-metonymic controls. However, slower reading times for metonymic constructions could also be due to competition between multiple interpretations that are being entertained in parallel whenever a metonymic verb is encountered. Using the visual-world paradigm, we devised an experiment which enabled us to determine the time course of metonymic interpretation in relation to non-metonymic controls. The experiment provided evidence in favor of a non-competitive, serial coercion process.     

Evidence for word length coding during visual word recognition

In a masked priming procedure manipulating orthographic neighbourhood size, the priming word activates a number of word candidates of which the target could be one. Whether the target is one of the candidates or not determines how quickly it is recognised. However, the efficiency of lexical processing may be markedly less if all possible candidates are activated. One solution to this problem is if the visual system uses prime length information to reduce the number of candidates to a more manageable amount. Here, we investigated in two masked priming experiments whether prime length and orthographic information combine to facilitate target word recognition. In Experiment 1, we showed that the efficiency of visual word recognition is not influenced by the length of primes alone. However, when combined with orthographically related primes, word length coding is preserved. In Experiment 2, we investigated whether length priming affects recognition of short and long words differently. Results showed that only short words benefit from a same-length orthographically related prime, and that the priming effect does not generalise to longer words. These results suggest that the length of a word is not an essential feature in lexical processing, but that it can facilitate recognition by constraining the activation of orthographically related words.     

Syllabification strategies in spoken word processing: Evidence from phonological priming

Two experiments were conducted to examine which of two linguistic syllabification strategies, the maximal onset principle and the stress principle, is operative in auditory word recognition. Test stimuli consisted of bisyllabic words and nonwords containing two medial consonants, the second of which had a potential attachment to either the coda of the first syllable or the onset of the second syllable (e.g., MARKET). A phonological priming paradigm was used in which prime stimuli were artificially syllabified by inserting silence. In the critical conditions, the prime was consistent with the maximal onset principle (e.g., MAR-KET) or with the stress principle (e.g., MARK-ET). The results suggested that listeners prefer to attach a medial consonant to the onset of a syllable in a way that is characterized by the maximal onset principle. The results also indicate some use of the stress principle.     

Sentence processing: Evidence against the serial,independent stage assumption

The assumption that the processing stages involved in sentence comprehension are serially and independently executed was tested in two separate experiments using two new sentence types. One experiment was a sentence-picture verification task, the other was not. In Experiment I, subjects viewed sentences such as It's false that the dots are red and indicated, by pressing the appropriate response key, which of two colors would make the sentence true. In Experiment II, subjects verified sentences such as It isn't true that the dots aren't red against pictures of colored dots. On the strength of Sternberg's additive factor method, present findings challenge the validity of the serial, independent stage assumption, and results are alternatively discussed in terms of a capacity sharing model.This research was supported by PHS research grant MH 23401 to Sam Glucksberg, principal investigator.     

Evidence for processing of constituent singleand multiletter codes: Support for multilevel coding in word perception

This investigation assessed whether constituent codes are activated during word processing. In two experiments, a priming procedure was used to test whether printed word primes facilitate skilled readers’ identification of subsequently presented letter and letter bigram targets appearing in the primes (present condition), relative to the same targets not appearing in the primes (absent condition). Experiment 1 demonstrated the priming of single-letter targets and indicated that priming effects are facilitative and not inhibitory. In Experiment 2, high- and low-frequency bigram targets appearing in word primes were shown to have a processing advantage over bigram targets not appearing in word primes. Single-letter constituents in low-frequency bigrams also benefited from priming; however, single letters in high-frequency bigrams showed no such benefit. The results in general suggest that both single- and multiletter constituents are available during word processing, and hence support multicomponent models of word perception.