How useful are corpus-based methods for extrapolating psycholinguistic variables?

Subjective ratings for age of acquisition, concreteness, affective valence, and many other variables are an important element of psycholinguistic research. However, even for well-studied languages, ratings usually cover just a small part of the vocabulary. A possible solution involves using corpora to build a semantic similarity space and to apply machine learning techniques to extrapolate existing ratings to previously unrated words. We conduct a systematic comparison of two extrapolation techniques: k-nearest neighbours, and random forest, in combination with semantic spaces built using latent semantic analysis, topic model, a hyperspace analogue to language (HAL)-like model, and a skip-gram model. A variant of the k-nearest neighbours method used with skip-gram word vectors gives the most accurate predictions but the random forest method has an advantage of being able to easily incorporate additional predictors. We evaluate the usefulness of the methods by exploring how much of the human performance in a lexical decision task can be explained by extrapolated ratings for age of acquisition and how precisely we can assign words to discrete categories based on extrapolated ratings. We find that at least some of the extrapolation methods may introduce artefacts to the data and produce results that could lead to different conclusions that would be reached based on the human ratings. From a practical point of view, the usefulness of ratings extrapolated with the described methods may be limited.     

Effects of modality on subjective frequency estimates and processing of spoken and printed words

Previous research (Garber & Pisoni, 1991; Pisoni & Garber, 1990) has demonstrated that subjective familiarity judgments for words are not differentially affected by the modality (visual or auditory) in which the words are presented, suggesting that participants base their judgments on fairly abstract, modality-independent representations in memory. However, in a recent large-scale study in Japanese (Amano, Kondo, & Kakehi, 1995), marked modality effects on familiarity ratings were observed. The present research further examined possible modality differences in subjective ratings and their implications for word recognition. Specially selected words were presented to participants for frequency judgments. In particular, participants were asked how frequently they read, wrote, heard, or said a given spoken or printed word. These ratings were then regressed against processing times in auditory and visual lexical decision and naming tasks. Our results suggest modality dependence for some lexical representations.     

Interaction and representational integration: evidence from speech errors

We examine the mechanisms that support interaction between lexical, phonological and phonetic processes during language production. Studies of the phonetics of speech errors have provided evidence that partially activated lexical and phonological representations influence phonetic processing. We examine how these interactive effects are modulated by lexical frequency. Previous research has demonstrated that during lexical access, the processing of high frequency words is facilitated; in contrast, during phonetic encoding, the properties of low frequency words are enhanced. These contrasting effects provide the opportunity to distinguish two theoretical perspectives on how interaction between processing levels can be increased. A theory in which cascading activation is used to increase interaction predicts that the facilitation of high frequency words will enhance their influence on the phonetic properties of speech errors. Alternatively, if interaction is increased by integrating levels of representation, the phonetics of speech errors will reflect the retrieval of enhanced phonetic properties for low frequency words. Utilizing a novel statistical analysis method, we show that in experimentally induced speech errors low lexical frequency targets and outcomes exhibit enhanced phonetic processing. We sketch an interactive model of lexical, phonological and phonetic processing that accounts for the conflicting effects of lexical frequency on lexical access and phonetic processing.     

Lexical configuration and lexical engagement: when adults learn new words

People know thousands of words in their native language, and each of these words must be learned at some time in the person's lifetime. A large number of these words will be learned when the person is an adult, reflecting the fact that the mental lexicon is continuously changing. We explore how new words get added to the mental lexicon, and provide empirical support for a theoretical distinction between what we call lexical configuration and lexical engagement. Lexical configuration is the set of factual knowledge associated with a word (e.g., the word's sound, spelling, meaning, or syntactic role). Almost all previous research on word learning has focused on this aspect. However, it is also critical to understand the process by which a word becomes capable of lexical engagement--the ways in which a lexical entry dynamically interacts with other lexical entries, and with sublexical representations. For example, lexical entries compete with each other during word recognition (inhibition within the lexical level), and they also support the activation of their constituents (top-down lexical-phonemic facilitation, and lexically-based perceptual learning). We systematically vary the learning conditions for new words, and use separate measures of lexical configuration and engagement. Several surprising dissociations in behavior demonstrate the importance of the theoretical distinction between configuration and engagement.     

Lexical configuration and lexical engagement: When adults learn new words

People know thousands of words in their native language, and each of these words must be learned at some time in the person’s lifetime. A large number of these words will be learned when the person is an adult, reflecting the fact that the mental lexicon is continuously changing. We explore how new words get added to the mental lexicon, and provide empirical support for a theoretical distinction between what we call lexical configuration and lexical engagement. Lexical configuration is the set of factual knowledge associated with a word (e.g., the word’s sound, spelling, meaning, or syntactic role). Almost all previous research on word learning has focused on this aspect. However, it is also critical to understand the process by which a word becomes capable of lexical engagement—the ways in which a lexical entry dynamically interacts with other lexical entries, and with sublexical representations. For example, lexical entries compete with each other during word recognition (inhibition within the lexical level), and they also support the activation of their constituents (top-down lexical-phonemic facilitation, and lexically-based perceptual learning). We systematically vary the learning conditions for new words, and use separate measures of lexical configuration and engagement. Several surprising dissociations in behavior demonstrate the importance of the theoretical distinction between configuration and engagement.     

The Effect of Inhibition of Return on Lexical Access

Abstract—Inhibition of return refers to a bias against returning attention to a location or object that has been recently attended. Recent research has shown that inhibition of return can be found not only in simple detection tasks, but also in tasks requiring relatively simple discrimination judgments. The present experiments examined whether inhibition of return occurs in tasks that require complex discriminations such as lexical decision and categorization tasks. Not only was inhibition of return found in both experiments, but a greater amount of inhibition was found for low-frequency words than for high-frequency words. These findings suggest that inhibition of return not only inhibits returning attention to previously attended locations, but can also affect the processing that is required for lexical access.     

Lexical access of function versus content words

There has been a simmering debate as to whether evidence exists for differential processes of lexical access for function and content words. This has centered around the frequency effect (higher word frequency reducing access times for content words but not function words). Previous work has used the lexical decision paradigm, which has been shown to reflect more than lexical access times. We measured naming times for words in sentences read for meaning. Our findings confirm that lexical access for function words is indeed faster than for content words as predicted by neurolinguistic theory and electrophysiological evidence, but that this difference can be attributed to word predictability (Cloze value) and word familiarity (log frequency). We also show that differences in frequency effect for the two word types holds only for the lower frequency words and not at all for the higher frequency words. We discuss the implications of the results for neurolinguistic theory.     

At-lexical, articulatory interference in silent reading: The “upstream” tongue-twister effect

In two experiments, we investigated the interpretation and boundary conditions of the tongue-twister (TT) effect in silent reading. Previously, McCutchen, Bell, France, and Perfetti (1991) observed a TT effect when students made semantic acceptability judgments on sentences, but not when they made lexical decisions on lists of words. Using similar methodology in Experiment 1, along with two changes (using “better” TTs and longer word lists), we observed a TT effect in a lexical decision task. In Experiment 2, a memory span task revealed that students recalled fewer words from TT lists than from control lists. These results suggest that the basic mechanism of the TT effect may be articulatory, rather than working-memory, interference that occurs during lexical access and resurfaces post-lexically, inhibiting efforts to maintain the temporal order of several words.     

Combining feature norms and text data with topic models

Many psychological theories of semantic cognition assume that concepts are represented by features. The empirical procedures used to elicit features from humans rely on explicit human judgments which limit the scope of such representations. An alternative computational framework for semantic cognition that does not rely on explicit human judgment is based on the statistical analysis of large text collections. In the topic modeling approach, documents are represented as a mixture of learned topics where each topic is represented as a probability distribution over words. We propose feature-topic models, where each document is represented by a mixture of learned topics as well as predefined topics that are derived from feature norms. Results indicate that this model leads to systematic improvements in generalization tasks. We show that the learned topics in the model play in an important role in the generalization performance by including words that are not part of current feature norms.     

Orthography and familiarity effects in word processing

Both orthographic regularity and visual familiarity have been implicated as contributors to the efficiency of processing visually presented words. Our studies sought to determine which of the internal codes representing words in the nervous system are facilitated by these two variables. To do this, sets of letter strings in which orthography and familiarity were factorially combined were used as the basis for physical, phonetic, semantic, and lexical judgments. The data indicated consistent effects of orthography on the activation of all codes. These effects were seen in same-different matching and in judgments of stimulus orientation, which are based on visual codes; in judgments of pronounceability based on phonetic codes; in judgments of meaningfulness based on semantic codes; and in lexical decisions, which are based on phonetic and semantic codes together. Familiarity, on the other hand, had a clear influence on the activation of semantic codes and to a lesser extent affected phonetic codes. Despite previous positive results found in matching letter strings, however, no influence of familiarity occurred in judgments based on visual codes once evidence for criterion shifting was eliminated. Our negative results included direct tests of facilitation in matching acronyms (e.g., FBI) and in matching both regular and irregular strings familiarized by specific training. It now appears that earlier findings of visual familiarity effects may be attributed to response biases resulting from the activation of higher level codes sensitive to familiarity, and to the use of small sets of training stimuli that allowed subjects to induce orthographic-like rules. The results obtained so far with our methods seem to reconcile an inconsistent literature by showing that speeded decisions based on visual codes are most strongly influenced by rule-governed processing mechanisms sensitive to orthographic structure, whereas decisions based on phonetic and semantic codes are affected about equally by rule-governed mechanisms and by stimulus-specific mechanisms sensitive to familiarity. This conclusion may lead to changes in notions of how effective various kinds of visual training are likely to be at different stages in the acquisition of reading skill.     

WINDSOR: Windsor improved norms of distance and similarity of representations of semantics

Lexical co-occurrence models of semantic memory form representations of the meaning of a word on the basis of the number of times that pairs of words occur near one another in a large body of text. These models offer a distinct advantage over models that require the collection of a large number of judgments from human subjects, since the construction of the representations can be completely automated. Unfortunately, word frequency, a well-known predictor of reaction time in several cognitive tasks, has a strong effect on the co-occurrence counts in a corpus. Two words with high frequency are more likely to occur together purely by chance than are two words that occur very infrequently. In this article, we examine a modification of a successful method for constructing semantic representations from lexical co-occurrence. We show that our new method eliminates the influence of frequency, while still capturing the semantic characteristics of words.     

Shock and awe: Distinct effects of taboo words on lexical decision and free recall

Taboo stimuli are highly arousing, but it has been suggested that they also have inherent taboo-specific properties such as tabooness, offensiveness, or shock value. Prior studies have shown that taboo words have slower response times in lexical decision and higher recall probabilities in free recall; however, taboo words often differ from other words on more than just arousal and taboo properties. Here, we replicated both of these findings and conducted detailed item analyses to determine which word properties drive these behavioural effects. We found that lexical-decision performance was best explained by measures of lexical accessibility (e.g., word frequency) and tabooness, rather than arousal, valence, or offensiveness. However, free-recall performance was primarily driven by emotional word properties, and tabooness was the most important emotional word property for model fit. Our results suggest that the processing of taboo words is influenced by distinct sets of factors and by an intrinsic taboo-specific property.     

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。     

Freeze or flee? Negative stimuli elicit selective responding

Humans preferentially attend to negative stimuli. A consequence of this automatic vigilance for negative valence is that negative words elicit slower responses than neutral or positive words on a host of cognitive tasks. Some researchers have speculated that negative stimuli elicit a general suppression of motor activity, akin to the freezing response exhibited by animals under threat. Alternatively, we suggest that negative stimuli only elicit slowed responding on tasks for which stimulus valence is irrelevant for responding. To discriminate between these motor suppression and response-relevance hypotheses, we elicited both lexical decisions and valence judgments of negative words and positive words. Relative to positive words (e.g., kitten), negative words (e.g., spider) elicited slower lexical decisions but faster valence judgments. Results therefore indicate that negative stimuli do not cause a generalized motor suppression. Rather, negative stimuli elicit selective responding, with faster responses on tasks for which stimulus valence is response-relevant.     

Strategies for visual word recognition and orthographical depth: a multilingual comparison

We investigated the psychological reality of the concept of orthographical depth and its influence on visual word recognition by examining naming performance in Hebrew, English, and Serbo-Croatian. We ran three sets of experiments in which we used native speakers and identical experimental methods in each language. Experiment 1 revealed that the lexical status of the stimulus (high-frequency words, low-frequency words, and nonwords) significantly affected naming in Hebrew (the deepest of the three orthographies). This effect was only moderate in English and nonsignificant in Serbo-Croatian (the shallowest of the three orthographies). Moreover, only in Hebrew did lexical status have similar effects on naming and lexical decision performance. Experiment 2 revealed that semantic priming effects in naming were larger in Hebrew than in English and completely absent in Serbo-Croatian. Experiment 3 revealed that a large proportion of nonlexical tokens (nonwords) in the stimulus list affects naming words in Hebrew and in English, but not in Serbo-Croatian. These results were interpreted as strong support for the orthographical depth hypothesis and suggest, in general, that in shallow orthographies phonology is generated directly from print, whereas in deep orthographies phonology is derived from the internal lexicon.     

Virtual experiments in megastudies: A case study of language and emotion

A recent dramatic increase in the number and scope of chronometric and norming lexical megastudies offers the ability to conduct virtual experiments—that is, to draw samples of items with properties that vary in critical linguistic dimensions. This paper introduces a bootstrapping approach, which enables testing of research hypotheses against a range of samples selected in a uniform, principled manner and evaluates how likely a theoretically motivated pattern is in a broad distribution of possible outcome patterns. We apply this approach to conflicting theoretical and empirical accounts of the relationship between the psychological valence (positivity) of a word and its speed of recognition. To this end, we conduct three sets of multiple virtual experiments with a factorial and a regression design, drawing data from two lexical decision megastudies. We discuss the influence that criteria for stimuli selection, statistical power, collinearity, and the choice of dataset have on the efficacy and outcomes of the bootstrapping procedure.     

Subcategorical phonetic mismatches and lexical access.

The place of phonetic analysis in the perception of words is unclear. While some theories assume fully specified phonemic strings as input, other theories assume that little analysis occurs. An earlier experiment by Streeter and Nigro (1979) produced evidence, based on auditorily presented words with misleading acoustic cues, that lexical decisions were based on mostly unanalyzed patterns, since word judgments were delayed by misleading information whereas nonword judgments were not. The present studies expand that work to a different set of cues, and to cases in which the overriding cue came first. An additional task, auditory naming, was used to examine the effects when the decision stage is less demanding. For the lexical decision task, misleading information slowed the responses, for both words and nonwords. In the auditory naming task, only the slower responses were affected. These results suggest that phonetic conflicts are resolved prior to lexical access.     

Subcategorical phonetic mismatches and lexical access

The place of phonetic analysis in the perception of words is unclear. While some theories assume fully specified phonemic strings as input, other theories assume that little analysis occurs. An earlier experiment by Streeter and Nigro (1979) produced evidence, based on auditorily presented words with misleading acoustic cues, that lexical decisions were based on mostly unanalyzed patterns, since word judgments were delayed by misleading information whereas non word judgments were not. The present studies expand that work to a different set of cues, and to cases in which the overriding cue came first. An additional task, auditory naming, was used to examine the effects when the decision stage is less demanding. For the lexical decision task, misleading information slowed the responses, for both words and nonwords. In the auditory naming task, only the slower responses were affected. These results suggest that phonetic conflicts are resolved prior to lexical access.     

Developmental bias for number words in the intraparietal sulcus

Children and adults show behavioral evidence of psychological overlap between their early, non‐symbolic numerical concepts and their later‐developing symbolic numerical concepts. An open question is to what extent the common cognitive signatures observed between different numerical notations are coupled with physical overlap in neural processes. We show that from 8 years of age, regions of the intraparietal sulcus (IPS) that exhibit a numerical ratio effect during non‐symbolic numerical judgments also show a semantic distance effect for symbolic number words. In both children and adults, the IPS showed a semantic distance effect during magnitude judgments of number words (i.e. larger/smaller number) but not for magnitude judgments of object words (i.e. larger/smaller object size). The results provide novel evidence of conceptual overlap between neural representations of symbolic and non‐symbolic numerical values that cannot be explained by a general process, and present the first demonstration of an early‐developing dissociation between number words and object words in the human brain.     

Semantic involvement in reading aloud: evidence from a nonword training study

The current research uses a novel methodology to examine the role of semantics in reading aloud. Participants were trained to read aloud 2 sets of novel words (i.e., nonwords such as bink): some with meanings (semantic) and some without (nonsemantic). A comparison of reading aloud performance between these 2 sets of novel words was used to provide an indicator of the importance of semantic information in reading aloud. In Experiment 1, in contrast to expectations, reading aloud performance was not better for novel words in the semantic condition. In Experiment 2, the training of novel words was modified to reflect more realistic steps of lexical acquisition: Reading aloud performance became faster and more accurate for novel words in the semantic condition, but only for novel words with inconsistent pronunciations. This semantic advantage for inconsistent novel words was again observed when a subset of participants from Experiment 2 was retested 6-12 months later (in Experiment 3). These findings provide support for a limited but significant role for semantics in the reading aloud process.