The Bayesian reader: explaining word recognition as an optimal Bayesian decision process

This article presents a theory of visual word recognition that assumes that, in the tasks of word identification, lexical decision, and semantic categorization, human readers behave as optimal Bayesian decision makers. This leads to the development of a computational model of word recognition, the Bayesian reader. The Bayesian reader successfully simulates some of the most significant data on human reading. The model accounts for the nature of the function relating word frequency to reaction time and identification threshold, the effects of neighborhood density and its interaction with frequency, and the variation in the pattern of neighborhood density effects seen in different experimental tasks. Both the general behavior of the model and the way the model predicts different patterns of results in different tasks follow entirely from the assumption that human readers approximate optimal Bayesian decision makers.     

What's in a name? access to information from people's names

Abstract

The processing of people's names is contrasted with face recognition and word recognition. The effects of the familiarity of initial and surnames and frequency of surnames (the number of people with the same surname) were investigated in several tasks. It was found that the effects of name familiarity and surname frequency were analogous to the effects of word frequency in tasks which did not require access to memory for individuals (a nationality decision and naming latency). In tasks which do require access to memory for individuals (familiarity decision and a semantic classification), the effect of surname frequency was analogous to the effect of distinctive-ness in face recognition. The results are discussed in terms of a functional model of name processing in which name recognition units mediate between the output of word recognition units and access to identity-specific semantics.     

BRINGING COMPUTATIONAL MODELS OF WORD NAMING DOWN TO THE ITEM LEVEL

Abstract— Early noncomputational models of word recognition have topically attempted to account for effects of categorical factors such as word frequency (high vs low) and spelling-to-sound regularity (regular vs irregular) More recent computational models that adhere to general connections: principles hold the promise of being sensitive to underlying item differences that are only approximated by these categorical factors In contrast to earlier models, these connectionist models provide predictions of performance for individual items In the present study, we used the item-level estimates from two connectionist models (Plaut, McClelland, Seidenberg, & Patterson, 1996, Seidenberg & McClelland, 1989) to predict naming latencies on the individual items on which the models were trained The results indicate that the models capture, at best, slightly more variance than simple log frequency and substantially less than the combined predictive power of log frequency, neighborhood density, and orthographic length. The discussion focuses on the importance of examining the item-level performance of word-naming models and possible approaches that may improve the models' sensitivity to such item differences     

Attentional resource demands of visual word recognition in naming and lexical decisions.

Attentional demands of lexical access were assessed with dual-task methodology. Subjects performed an auditory probe task alone (single-task) or combined (dual-task) with either a lexical decision or a naming task. In Experiment 1, probe performance showed a decrement from single- to dual-task conditions during recognition of words in both lexical decision and naming tasks. In addition, decrements in probe performance were larger during processing of low-frequency compared with high-frequency words in both of the word recognition tasks. Experiment 2 showed that the time course of frequency-sensitive demands was similar across lexical decision and naming tasks and that attention is required early in the word recognition sequence. The results support the assumption that lexical access is both frequency sensitive and attention demanding.     

Comparing naming, lexical decision, and eye fixation times: Word frequency effects and individual differences

Performance on three different tasks was compared: naming, lexical decision, and reading (with eye fixation times on a target word measured). We examined the word frequency effect for a common set of words for each task and each subject. Naming and reading (particularly gaze duration) yielded similar frequency effects for the target words. The frequency effect found in lexical decision was greater than that found in naming and in eye fixation times. In all tasks, there was a correlation between the frequency effect and average response time. In general, the results suggest that both the naming and the lexical decision tasks yield data about word recognition processes that are consistent with effects found in eye fixations during silent reading.     

The locus of word frequency effects in skilled spelling-to-dictation

In spelling-to-dictation tasks, skilled spellers consistently initiate spelling of high-frequency words faster than that of low-frequency words. Tainturier and Rapp's model of spelling shows three possible loci for this frequency effect: spoken word recognition, orthographic retrieval, and response execution of the first letter. Thus far, researchers have attributed the effect solely to orthographic retrieval without considering spoken word recognition or response execution. To investigate word frequency effects at each of these three loci, Experiment 1 involved a delayed spelling-to-dictation task and Experiment 2 involved a delayed/uncertain task. In Experiment 1, no frequency effect was found in the 1200-ms delayed condition, suggesting that response execution is not affected by word frequency. In Experiment 2, no frequency effect was found in the delayed/uncertain task that reflects the orthographic retrieval, whereas a frequency effect was found in the comparison immediate/uncertain task that reflects both spoken word recognition and orthographic retrieval. The results of this two-part study suggest that frequency effects in spoken word recognition play a substantial role in skilled spelling-to-dictation. Discrepancies between these findings and previous research, and the limitations of the present study, are discussed.     

Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage

Three experiments investigated the impact of five lexical variables (instance dominance, category dominance, word frequency, word length in letters, and word length in syllables) on performance in three different tasks involving word recognition: category verification, lexical decision, and pronunciation. Although the same set of words was used in each task, the relationship of the lexical variables to reaction time varied significantly with the task within which the words were embedded. In particular, the effect of word frequency was minimal in the category verification task, whereas it was significantly larger in the pronunciation task and significantly larger yet in the lexical decision task. It is argued that decision processes having little to do with lexical access accentuate the word-frequency effect in the lexical decision task and that results from this task have questionable value in testing the assumption that word frequency orders the lexicon, thereby affecting time to access the mental lexicon. A simple two-stage model is outlined to account for the role of word frequency and other variables in lexical decision. The model is applied to the results of the reported experiments and some of the most important findings in other studies of lexical decision and pronunciation.     

Adult age differences in letter-level and word-level processing

Older and young adults' letter detection and lexical decision performance were examined as word frequency varied to determine whether there were age differences in word recognition. Allen and Madden (1989) found that older adults' pattern of reaction time (RT) across word frequency categories was different from young adults' pattern for a letter detection task. In this study, for both letter detection and lexical decision tasks, older adults exhibited a monotonically decreasing RT function as word frequency increased. However, young adults exhibited a nonmonotonic RT function across word frequency for the letter detection task but a monotonically decreasing RT function as word frequency increased for the lexical decision task. An expanded parallel input serial analysis model of word processing was hypothesized.     

The Role of Orthographic Neighborhood Size Effects in Chinese Word Recognition

Previous studies about the orthographic neighborhood size (NS) in Chinese have overlooked the morphological processing, and the co-variation between the character frequency and the the NS. The present study manipulated the word frequency and the NS simultaneously, with the leading character frequency controlled, to explore their influences on word lexical decision (Experiment 1) and naming (Experiment 2). The results showed a robust effect that words with a larger NS produced shorter reaction time than those with a smaller NS, irrespective of the word frequency and the tasks. This facilitative effect may occur due to a semantic network formed by neighbor words, resulting in the semantic activation to accelerate the word recognition. Moreover, the comparison of the effect sizes of word frequency between the two tasks showed that lexical decision responses demonstrated a larger word frequency effect, indicating that the sub-word processing was involved in the multi-character word recognition.     

The Dimensionality of Memory Strength Between Levels of Both Serial Position and Word Frequency Category

In recognition tasks where Signal Detection Theory (SDT) is applied, a suitable decision variable is required. This variable is typically assumed to be a global assessment of memory strength. A further assumption is that factors such as old-new, serial position (SP), and word frequency (WF) simply act to change the value of strength across levels within the factor. In terms of SDT, the decision variable is assumed to be uni-dimensional. Four experiments are reported that provide a direct empirical assessment of this assumption for SP and WF. Judgements of items (JOI), recency (JOR), and primacy (JOP) were measured for SP; and subjective frequency (JOF) was measured for WF. Multiple dimensions imply non-additive effects across factor levels and different retrieval processes or the influence of an extra process. A single dimension is consistent with an additive effect of factor levels and current theorizing. The data are in fairly good agreement with the uni-dimensional assumption across all four factors with the exception of JOR and JOP. These showed less uni-dimensionality for comparisons between beginning vs. end list items. The implications of this are discussed.     

Individual differences in word recognition latency

Previous studies of the effects of word characteristics on word recognition have used orthogonal combinations of word variables and have failed to consider individual differences. The present study examined word naming (Experiment 1) and lexical decision (Experiment 2) tasks using an unrestricted set of words and a correlational analysis. Individual differences were considered using a measure of the subjects’ knowledge of the English vocabulary. The results of Experiment 1 indicated that log (RT) for word naming is affected by word length, word frequency, and the number of syllables in the word; the results of Experiment 2 confirmed the effects of length and frequency but also showed that log (RT) is a function of the age at which the word is introduced to a child’s reading vocabulary. Subjects with a high vocabulary score were more rapid in Experiment 1 but were slower in Experiment 2, compared to subjects with a low vocabulary score. More importantly, high-vocabulary subjects, in both studies, were less affected by word length than the low-vocabulary subjects. The results suggest that subjects do differ in their reading strategy and that word length and word frequency may affect different stages in the word recognition process.     

The word-frequency paradox for recall/recognition occurs for pictures

A yes-no recognition task and two recall tasks were conducted using pictures of high and low familiarity ratings. Picture familiarity had analogous effects to word frequency, and replicated the word-frequency paradox in recall and recognition. Low-familiarity pictures were more recognizable than high-familiarity pictures, pure lists of high-familiarity pictures were more recallable than pure lists of low-familiarity pictures, and there was no effect of familiarity for mixed lists. These results are consistent with the predictions of the Search of Associative Memory (SAM) model.     

Eye movements during the reading of compound words and the influence of lexeme meaning

We examined the use of lexeme meaning during the processing of spatially unified bilexemic compound words by manipulating both the location and the word frequency of the lexeme that primarily defined the meaning of a compound (i.e., the dominant lexeme). The semantically dominant and nondominant lexemes occupied either the beginning or the ending compound word location, and the beginning and ending lexemes could be either high- or low-frequency words. Three tasks were used--lexical decision, naming, and sentence reading--all of which focused on the effects of lexeme frequency as a function of lexeme dominance. The results revealed a larger word frequency effect for the dominant lexeme in all three tasks. Eye movements during sentence reading further revealed larger word frequency effects for the dominant lexeme via several oculomotor motor measures, including the duration of the first fixation on a compound word. These findings favor theoretical conceptions in which the use of lexeme meaning is an integral part of the compound recognition process.     

Regularity effects in word naming: What are they?

In a word-naming experiment, word-body consistency was crossed with grapheme-to-phoneme regularity to test predictions of current models of word recognition. In the latency and error data, a clear effect of consistency was observed, with the influence of regularity somewhat weaker. In addition, simulation data from three contemporary models of word recognition were obtained for the stimuli used in the experiment in order to compare the models' latencies with those of humans. The simulations showed that the human latency data are most consistent with the parallel-distributed-processing model of Plaut, McClelland, Seidenberg, and Patterson (1996), less so with the dual-process model (Zorzi, Houghton, & Butterworth, 1998), and least so with the dual-route-cascaded model (Coltheart & Rastle, 1994).     

The modulation of spatial congruency by object-based attention: Analysing the “locus” of the modulation

In a diffusion model, performance as measured by latency and accuracy in two-choice tasks is decomposed into different parameters that can be linked to underlying cognitive processes. Although the diffusion model has been utilized to account for lexical decision data, the effects of stimulus manipulations in previous experiments originated from just one parameter: the quality of the evidence. Here we examined whether the diffusion model can be used to effectively decompose the underlying processes during visual-word recognition. We explore this issue in an experiment that features a lexical manipulation (word frequency) that we expected to affect mostly the quality of the evidence (the drift rate parameter), and a perceptual manipulation (stimulus orientation) that presumably affects the nondecisional time (the T_er parameter, time of encoding and response) more than it affects the drift rate. Results showed that although the manipulations do not affect only one parameter, word frequency and stimulus orientation had differential effects on the model's parameters. Thus, the diffusion model is a useful tool to decompose the effects of stimulus manipulations in visual-word recognition.     

The time course of spoken word learning and recognition: studies with artificial lexicons

The time course of spoken word recognition depends largely on the frequencies of a word and its competitors, or neighbors (similar-sounding words). However, variability in natural lexicons makes systematic analysis of frequency and neighbor similarity difficult. Artificial lexicons were used to achieve precise control over word frequency and phonological similarity. Eye tracking provided time course measures of lexical activation and competition (during spoken instructions to perform visually guided tasks) both during and after word learning, as a function of word frequency, neighbor type, and neighbor frequency. Apparent shifts from holistic to incremental competitor effects were observed in adults and neural network simulations, suggesting such shifts reflect general properties of learning rather than changes in the nature of lexical representations.     

The overlap model: a model of letter position coding

Recent research has shown that letter identity and letter position are not integral perceptual dimensions (e.g., jugde primes judge in word-recognition experiments). Most comprehensive computational models of visual word recognition (e.g., the interactive activation model, J. L. McClelland & D. E. Rumelhart, 1981, and its successors) assume that the position of each letter within a word is perfectly encoded. Thus, these models are unable to explain the presence of effects of letter transposition (trial-trail), letter migration (beard-bread), repeated letters (moose-mouse), or subset/superset effects (faulty-faculty). The authors extend R. Ratcliff's (1981) theory of order relations for encoding of letter positions and show that the model can successfully deal with these effects. The basic assumption is that letters in the visual stimulus have distributions over positions so that the representation of one letter will extend into adjacent letter positions. To test the model, the authors conducted a series of forced-choice perceptual identification experiments. The overlap model produced very good fits to the empirical data, and even a simplified 2-parameter model was capable of producing fits for 104 observed data points with a correlation coefficient of .91.     

词素位置概率在中文阅读中的作用：词汇判断和眼动研究

中文词切分的认知机制一直是心理语言学关注的焦点问题之一,研究发现中文读者可使用词素位置概率等语言学线索帮助词切分,而首、尾词素位置概率的重要程度以往研究观点并不一致。本研究通过词汇判断和眼动实验,考察中文读者对首、尾词素位置概率的利用情况。实验1a和1b采用词汇判断任务,考察在不同词频条件下词首与词尾词素位置概率对词汇识别的影响。在错误率和反应时指标上,高频词条件下词首与词尾词素位置概率效应不显著;低频词条件下词首词素位置概率效应显著,词尾词素位置概率效应不显著。实验2a和2b采用句子阅读任务,考察在自然阅读情境中被试对词素位置概率的运用。在凝视时间、回视路径时间和总注视时间指标上,低频词条件下词首词素位置概率效应显著,词尾词素位置概率效应不显著。高频词条件下词首与词尾词素位置概率效应均不显著。词汇判断和眼动证据共同表明,词素位置概率信息是中文读者重要的语言词切分线索,且与词尾词素位置概率相比,词首词素位置概率在词汇切分与识别过程中发挥的作用更大。同时,词素位置概率线索的运用会受到词频的影响,研究结果支持复合词加工的混合通达表征模型。     

Word shape's in poor shape for the race to the lexicon

Current models of fluent reading often assume that fast and automatic word recognition involves the use of a supraletter feature corresponding to the envelope or shape of the word when it is printed in lowercase. The advantages of mixed case over pure case and of pure lowercase over pure uppercase have often been taken as evidence favoring the word-shape hypothesis. Alternative explanations for these phenomena are offered. Experiment 1 shows that previous demonstrations of word-shape effects during proofreading are better described as individual letter effects. Experiments 2-4 explore the possibility that word shape facilitates lexical access through uncertainty reduction. In all three experiments performance on words with rare shapes is compared to those with common shapes. There were no effects of shape frequency in either tachistoscopic recognition or lexical-decision tasks. This was true regardless of the degree to which the visual shape cue was supplemented by the nonvisual factors of familiarity and expectancy. Possible reasons why fluent readers ignore word shape are discussed within the framework of a model that assumes that automatic word recognition is mediated by the activation of abstract letter identities.     

Similarity, global matching, and judgments of frequency

In the test-pair similarity effect, forced-choice recognition is more accurate for similar test pairs, such as leopard-cheetah, than it is for unrelated test pairs, such as leopard-turnip. According to global matching models, this occurs because the retrieved familiarities of similar items are correlated. In the Minerva 2 model, global matching underlies frequency judgments as well as recognition memory. One implication of this model is that judged frequencies of similar items should be correlated. Another implication is that judgments of summed frequency for pairs of words (how many presentations were there of word1 and word2 combined?) should have higher variance when word1 and word2 are similar than when they are unrelated. These predictions were tested and confirmed in two experiments. A review of these and other results suggests that theories of recognition memory should also be applicable to frequency-judgment tasks.