Word-shape and word-lexical-frequency effects in lexical-decision and ...

0 downloads 0 Views 690KB Size Report
between shape frequency and lexical frequency, indicating that, for rare words, having a rare shape speeded up lexical decisions. Experiment 2 primed shape.
VISUAL COGNITION, 2003, 10 (8), 913±948

Word-shape and word-lexical-frequency effects in lexical-decision and naming tasks Bernard LeÂte INRP and Universite de Provence, France

JoeÈl Pynte CNRS and Universite de Provence, France The present study investigated the use of word-shape information in visual word recognition. Word-shape frequency was computed for lexically frequent and rare words. Experiment 1 contrasted lowercase and uppercase presentations in a lexicaldecision task. The observed latencies indicated a shape-frequency effect in lowercase presentation, i.e., responses were given faster for words with a lowfrequency shape than for words with a high-frequency shape, and an interaction between shape frequency and lexical frequency, indicating that, for rare words, having a rare shape speeded up lexical decisions. Experiment 2 primed shape information with a 400 ms SOA. The results showed that high-frequency shape words benefited more from the priming procedure than did low-frequency ones. Priming was also used in a naming task (Experiment 3). The results indicated a strong priming effect for all four target types. When all words were given an up± down configuration (Experiment 4), the same pattern of results as in Experiment 1 was found, rejecting a letter-confusability explanation. Taken together, the results suggest that shape information affects word recognition. Having a rare shape seems to shorten lexical decision times on lexically rare words and to lengthen naming times on lexically frequent words.

Is word-shape information used by skilled readers during word recognition? Currently, most theorists suggest instead that word recognition is mediated by letter unit identification, and that there is no need to postulate that word shape enters independently into word identification. Consequently, most models of word recognition assume today that words are formed from their component letters (Adams, 1979; Besner & Johnston, 1989; Grainger & Jacobs, 1996; McClelland & Rumelhart, 1981; Paap, Newsome, McDonald, & Schvaneveldt, 1982; Rumelhart & McClelland, 1982; Seidenberg & McClelland, 1989).

Please address all correspondence to: B. LeÂteÂ, Laboratoire Parole et Langage, 29 ave. Robert Schuman, 13621 ± Aix-en-Provence, cedex 1, France. E-mail: [email protected] # 2003 Psychology Press Ltd http://www.tandf.co.uk/journals/pp/13506285.html DOI:10.1080/13506280344000112

914

LEÂTEÂ AND PYNTE

However, many theorists agree that perception involves the rapid analysis of stimuli at a number of levels, often regarded as roughly hierarchical. The sensory systems transduce the physical event and supply various sources of information called features. The preliminary stages of word recognition analyse these physical or sensory features, which include lines, angles, brightness, pitch, and loudness, while later stages recognize patterns and extract meaning. For example, in the interactive-activation model of McClelland and Rumelhart (1981) and Rumelhart and McClelland (1982), processing is organized into the feature stage, the letter stage, and the word stages. When a string of letters is presented to the visual system, each feature activates all nodes of letters that are consistent with it, and inhibits all nodes of letters that are not. Features can be studied at the word level or at the letter level. At the word level, features are often called transletter or supraletter features because they are assumed to be composed of multiletter patterns and even whole-word patterns. Word recognition is therefore viewed as a comparison process wherein a pattern is compared with a set of stored templates (one of each word known). In one of the stronger versions of this viewpoint, the holistic view of word recognition, the shape (or envelope) of the word may be sufficient for identifying words (this belief was the basis for the global or whole-word method for teaching reading). For the advocates of this view (Allport, 1977; Dechant & Smith, 1977), a reader sees a word as a unified symbol rather than as a collection of related letters. If we consider features at the letter level, we are taking a more analytic view of word processing, because letters can be described in terms of their distinctive features. One approach is to assume that letter features used in recognition are general and abstract. Gibson's (1969) feature list, for example, includes general binary contrasts such as the presence or absence of straight (vertical, horizontal, diagnonal), curve (closed, open vertically, or open horizontally), intersection. Such contrast may be important for word recognition because they are generally applicable and easy for the visual system to detect. Another approach is to consider that letter features are global. Global features involve detailed relationships between parts of letters. The importance of global features was demonstrated in Bouma's (1971) study of confusions between lowercase letters. Bouma found that letters with similar shapes were more likely to be confused. Sanocki (1991) found that typical patterns (i.e., letters in ``normal'' fonts) were easier to recognize than less typical patterns (i.e., letters in ``abnormal'' fonts). Other examples of word features include the shape of the spaces between letters (Wheeler, 1970) and density, which requires a more detailed resolution of the stimulus' high spatial frequency, a salient cue that may contribute to the recognition process. Empirical evidence for word features as a cue to word recognition has come from studies using tasks with high ecological validity, such as reading continuous texts or proofreading for misspelled words. Haber, Haber, and Furlin (1983) had subjects read short texts of about 150 words. On each trial, one or

WORD-SHAPE EFFECT

915

two text lines were displayed that terminated randomly in the middle of a sentence. The reader had to guess the next word on the basis of the preceding context or via additional cues like word length and word shape. Guessing was improved by about 10% when word-length information was available and another 10% when shape information was given. These results seem to imply that word-shape information is a cue for word recognition. However, an alternative explanation can be proposed. As argued by Besner and Johnston (1989; see also Paap, Newsome, & Noel, 1984), subjects may generate a word hypothesis from the preceding context alone and then match it against subsequent shape information, rather than using shape information to constrain the guess. Monk and Hulme (1983) created misspelled words by deleting or substituting letters in critical words in such a way that word shape was either altered or maintained. The results showed that substitutions were more likely to be noticed than deletions, and for both types of alterations, those that modified word shape (changes in the presence or absence of ascenders and descenders) were more often noticed than alterations that preserved word shape. When mixed-case stimuli were used, the shape effect disappeared (see also Haber & Schindler, 1981). As argued by Oden (1984), these results may mean only that ascenders and descenders are particularly important to letter identification, and not necessarily that the distinctive word shapes produced by ascenders and descenders are used in word recognition. Criticisms of the last two experiments were taken into account by Healy and Cunningham (1992), who tried to extend the Monk and Hulme (1983) findings by modifying their procedure in three crucial ways. First, so as to avoid disrupting normal reading processes, they used uppercase letters in the baseline condition rather than mixed-case stimuli. Second, words were misspelled solely by deleting one of four letters, s, c, k, or p, which have similar features in the lowercase and uppercase. Third, they tested children of various ages and reading abilities as well as college-age adults, in order to examine the relationship between age, reading level, and word shape. The results showed that all subjects were sensitive to changes in word shape (in the lowercase condition, subjects missed more words on s and c deletions, which retained word shape, than on k and p deletions, where word shape was altered) and that proofreading errors decreased with age and increased with reading ability. However, the overall error pattern did not vary as a function of age or reading level. These results show that subjects searching for misspelled words tolerate misspellings involving missing letters if word shape is maintained. Evidence from neuropsychological studies (Marsolek, 1995; Marsolek, Kosslyn, & Squire, 1992; Marsolek, Schacter, & Nicholas, 1996) suggests that two relatively independent visual-form subsystems operating in the brain can process word shape information: (1) an abstract visual-form subsystem that underlies recognition of abstract categories of form (e.g., A/a vs. S/s) and

916

LEÂTEÂ AND PYNTE

operates more effectively in the left hemisphere than in the right; and (2) a specific visual-form subsystem that underlies recognition of instances of forms (e.g., A vs. a) and operates more effectively in the right hemisphere than in the left. Following this line of reasoning, Faust, Babkoff, and Avidor-Reiss (2000) recently showed that shape information is processed differently by the two cerebral hemispheres. They used word shape and congruent sentence context as coprimes, and observed that lexical decision latencies were speeded up for subsequent Hebrew target words. This joint effect varied orthogonally when targets activated the right hemisphere, and interactively when targets activated the left hemisphere. They assumed that the subsystem that processes abstract forms (left hemisphere) is dedicated to specifying the identity of the word or cluster of letters, and their arrangement, whereas the other subsystem, which processes form-specific information (right hemisphere), is dedicated to distinguishing between different forms. Other evidences for the use of shape information in word recognition can be found in experiments using MiXeD-cAsE stimuli. Although some studies have shown that breaking up the shape by case alternation does not impair recognition (McClelland, 1976; McConkie & Zola, 1979; Smith, 1969; Smith, Lott, & Cronnell, 1969), a large number of experiments have shown that case mixing has a strong effect on performance (Besner, 1989; Besner & Johnston, 1989; Besner & McCann, 1987; Kinoshita, 1987; Mayall & Humphreys, 1996). However, the results remain unclear because they depend on the task and the type of stimuli used. For example, in a naming task, case alternation was found to disrupt lexically rare words more than frequent ones, whereas in a lexical task, case mixing and word frequency effects were cumulative (Besner & McCann, 1987). In other cases, case mixing was found to affect lexical decision more than naming (Besner & McCann, 1987; Mayall & Humphreys, 1996) and to have a greater effect on nonwords than on words in a naming task, while affecting words more than nonwords in a lexical-decision task (Besner & Johnston, 1989; Besner & McCann, 1987; but see Kinoshita, 1987). The question of what kinds of visual information are disrupted by case mixing in naming tasks was addressed by Mayall, Humphreys, and Olson (1997). They showed that case-mixing disruption effects are due to at least two factors: (1) The disruption of transletter features caused by the insertion of entirely different letter features, by the introduction of differing relative sizes of letters, or by the distortion of the shapes of the spaces between letters; and (2) the introduction of inappropriate grouping between letters with the same size and case (for example, the A and E in the word ArEa may be grouped together and may form an inappropriate unit for visual lexical access). One of the strongest arguments against the word-shape idea is its small potential contribution to reading. Overall word shape does not sufficiently differentiate the words of a language (Groff, 1975) and to be useful, shape would have to be used in conjunction with other cues, such as orthographic features

WORD-SHAPE EFFECT

917

(Walker, 1987) or syntactic and semantic features (Haber et al., 1983). Walker (1987) determined the distinctiveness of different word shapes (i.e., wordshape frequency) by analysing the three- to seven-letter words in the KucÏera and Francis (1967) corpus, and found that only a small percentage of the words had a unique shape, with the percentage increasing steadily from 0.5 to 2.9 as word length increased (see below for a similar analysis in French). So word shape can rarely be used to completely specify the identity of a word. However, when shape information is combined with other cues, such as other orthographic features (e.g., the identity of the word's first letter), or sentence context (as in Faust et al., 2000), the number of candidate words is considerably reduced. Paap et al.'s (1984) findings are often said to provide the basic arguments against the use of word-shape information in the word-recognition process. They limit the definition of word-shape information to the holistic pattern formed by the sequence of ascending, descending, and neutral letters which ``includes information about both the length of the word (and hence the number of letters) and the smooth outline, or envelope, that surrounds the contours of the word'' (p. 414). Paap et al. argued that all apparent effects of word shape in their proofreading task were due to the confounding factor of letter confusability (their Exp. 1). In another three experiments, they explored the possibility that word shape facilitates lexical access through uncertainty reduction. They assumed that a critical determinant of the utility of a word's shape is the number of other words that share the same shape, i.e., shape frequency. On this basis, they compared performance on words with rare shapes to those with common shapes, displayed in lower- or uppercase. They matched their two sets of words (common vs. rare shape) by looking at their constituent letters: (1) the two words had the same number of ascending, descending, and neutral letters; (2) the two words had at least three letters in common for four- to five-letter words, and at least four letters in common for six-letter words; and (3) the two words had the same number of repeated letters (recall paired with cellar). Searching for the word-superiority effect (their Exp. 2), they predicted that it would be greater for words with rare shapes than common ones. They found that performance on word stimuli was not affected by shape frequency, and that shape frequency had no effect on the magnitude of the word-superiority effect. In a lexical-decision task (their Exp. 3), they compared three shape frequencies (rare, common, and intermediate) and two lexical frequencies (high vs. low), and found no significant tendency for rare-shape words to be accessed faster than common-shape words. In their fourth experiment, targets were preceded by a semantically related or unrelated prime. If subjects' expectations about targets include information about their shapes, then words with rare shapes ought to be identified faster than words with common shapes. No effect of shape frequency was found. The result pattern reported by Paap et al. led them to consider that information about supraletter features is lost early in the word recognition

918

LEÂTEÂ AND PYNTE

process. However, they agreed that their experiments did not provide a critical test for the use of letter features in word recognition. Many other studies have addressed the question of the usefulness of shape information in word recognition, and yet there is no consensus about the final answer. For example, recently, Perea and Rosa (2002) distorted stimuli by size alternation (letters were alternately in 10- and 14-point Courier). In a lexical decision task (their Exp. 1), they found that the effect was greater for low-frequency words than for high-frequency words. Manipulating case type (lowercase vs. uppercase; their Exp. 2) they found an effect only for low-frequency words, as in Mayall and Humphreys (1996, Exp. 1). In both experiments, nonwords were not affected by visual familiarity. These results showed that recognition of familiar words did not rely on word-shape information. However, Perea and Rosa also found main effects of size alternation and case type (both factors lengthening response latencies), illustrating that visual familiarity played a role in lexical access. The fact that both factors occurred for words rather than for nonwords suggests that the locus of the visual familiarity effect occurs late in word processing. As pointed out by McClelland (1977), one difficulty in shape information research is that word-shape information does not exist without letter-shape information. Thus, it is not possible to set up experimental conditions with words that have word-shape information but do not have letter-shape information or vice versa. So, the term ``shape information'' will be used here to refer to two sets of word features: Supreletter features and letter features. Supraletter features will include the shape of the whole word and all features larger than single letters but smaller than whole words (so-called transletter features). Letter features will include distinctive features of letters like curvature, relative size of letters, and the presence or absence of ascenders and descenders. This dichotomy makes the distinction between global shape information furnished by overall word shape and local information furnished by overall letter shape. What is critical for our purposes is to distinguish between two theoretical positions, a radical one and a safer one. The radical position claims that all words can be recognized solely on the basis of whole-word features, and that supraletter features are used directly to find a word because perceptual characteristics are stored as learned units. The safer position, which we shall adopt here, contends that printed words are recognized via a variety of units ranging from word and/or letter features and sublexical letter units to single letters or activated abstract letter identities. This leads to a more flexible view of word recognition, as in Allen and Maden's (1990) parallel input serial analysis model, Drewnowski and Healy's (1977) unitization model, and Massaro and Cohen's (1994) fuzzy logical model of word perception, where reading letters and words is basically a pattern recognition problem. The general aim of the four experiments described in the present paper was to provide a detailed evaluation of the effects of shape information on word

WORD-SHAPE EFFECT

919

recognition, in an attempt to shed light on the issues raised by the above findings. These effects were studied in a lexical-decision task (Experiments 1, 2, and 4) and in a naming task (Experiment 3).

EXPERIMENT 1: CASE, WORD-SHAPE, AND WORD-LEXICALFREQUENCY EFFECTS ON LEXICAL DECISIONS As pointed out by Walker (1987), as a prerequisite to any comparison of common versus rare word shapes, normative information about the distinctiveness of word shapes is required. So, a detailed analysis of how word-shape frequency was computed in our study will be presented below. In Experiment 1, high-frequency shape words and low-frequency shape words were presented in a lexical-decision task, either in lowercase or in uppercase letters. In order to determine how shape information is used on common or rare words, we also combined shape frequency with lexical frequency, keeping it within a clearly delineated frequency range. Because competition between words is now a critical requirement in wordrecognition models (Andrews, 1989, 1992; Grainger, 1990; Grainger & Segui, 1990; Sears, Hino, & Lupker, 1995), we decided to control the orthographic neighbourhood of targets, and other information provided by their orthographic form. The targets were strictly controlled for bigram frequencies and number of syllables, and none of the targets had any neighbours, as defined by Coltheart, Davelaar, Jonasson, and Besner (1977). The number of phonemes and the number of two-letter neighbours were less accurately controlled.

Predictions If supraletter features, as defined above, are useful in word identification, there should be an advantage for lowercase words because they produce the most discriminable global shape. Following Paap et al.'s (1984) line of reasoning, we can assume that a critical determinant of the utility of a word's shape is the number of other words that share the same shape. If the shape of a word matches only one word, fast lexical access can occur with little additional information about the individual letters. However, if the shape matches many candidates, its usefulness is reduced and readers must rely upon letter identification. Thus, word shape facilitates word recognition only to the extent that it serves to preclude all but a relatively small set of candidate words, what Paap et al. called the uncertainty-reduction hypothesis. This leads to the prediction of a shape-frequency effect in lowercase presentation, i.e., words with infrequent shapes should be easier to recognize than words with frequent shapes. The uncertainty-reduction hypothesis clearly places the locus of the shapefrequency effect at the level of lexical identification, where candidates are

920

LEÂTEÂ AND PYNTE

compared to find a lexical entry that best matches the input as in Paap et al.'s (1982) activation-verification model. In this model, the lexical-identification process has two stages. The activation stage constructs a posticonic visual representation of the target and then generates a sensory set from the activated words that are orthographically similar to the target. Word units are activated to the extent that their constituent letters are activated in the alphabetum. The alphabetum possesses a feature list for each letter of the alphabet. The activation of level of each letter is determined by the number of matching and mismatching features detected in the target. During the second stage, the verification stage, the set of candidates that share features with the target undergoes a frequencyordered, sequential comparison process. This conception leads to a second prediction about the lexical-frequency effect. In the case of a target with a frequent shape, the set of candidates generated during the activation stage should be greater than for a target with an infrequent shape, thereby slowing down the comparison process and leading to a stronger lexical-frequency effect.

Method Subjects. Fifty first- and second-year Psychology students at the University of Bourgogne took part in experiment for course credit. All were native speakers of French and unaware of the purpose of the experiment. Word-shape frequency computation. Word-shape frequency was computed in French using Content, Mousty, and Radeau's (1990) computerized word pool. All two- to thirteen-letter words in the database were examined (excepted compound words), which made 30,095 words in all. Separate statistics were computed for each word length. To derive information regarding the distribution of word shapes in printed lowercase, each word was first transformed into a character string according to the pattern created by its ascending letters (b ± d ± f ± h ± k ± l ± t), descending letters (g ± j ± p ± q ± y), neutral letters (a ± c ± cË ± e ± i ± m ± n ± o ± r ± s ± u ± v ± w ± x ± z), and marked letters (aÁ ± aà ± aÈ ± Ãõ ± Èõ ± e ± eÁ ± eà ± eÈ ± oÈ ± oà ± uÁ ± uà ± uÈ). The codes for the character strings were A, D, N, M, respectively. (Note that in Walker's, 1987 count, there were no marked letters, since English has none.) The letter ``i'' was put in the neutral-letter set because we considered that only one dot above the vertical line did not make it stand out sufficiently to be classified as a marked letter. The decision to put the letter ``cË'' in the neutral-letter set was more difficult, and can be considered arbitrary and perhaps unjustified. Note, however, that this decision had no impact on the results of the present study because no target with that letter was chosen. Then, for each shape encountered, the words sharing that shape were counted and this gave us a measure of word-shape frequency. For example, the character string NAANN is shared by six five-letter words (album, aller, atlas, obtus, offre,

WORD-SHAPE EFFECT

921

TABLE 1 Analysis of word-shape frequency

Length 2 3 4 5 6 7 8 9 10 11 12 13 Total

Number of words

Number of shapes

61 269 897 2,120 3,481 4,529 4,933 4,567 3,700 2,688 1,778 1,072 30,095

7 25 78 207 438 730 1,155 1,468 1,619 1,443 1,164 803 9,137

Percentage of shapes that were unique (one candidate)

Percentage of words with a unique shape

14.3 12.0 24.4 34.8 44.1 46.2 52.7 59.1 66.6 70.9 77.5 81.2

1.6 1.1 2.1 3.4 5.5 7.4 12.3 19.0 29.2 38.1 50.7 60.8

ultra), which make up the set of candidates for that shape. These words were thus assigned a shape frequency count of 6. Table 1 provides a summary of the analysis. For each word length, the table gives the number of words examined (second column), the number of different shapes encountered (third column), the percentage of shapes that were unique (fourth column), and the percentage of words with a unique shape (fifth column). As pointed out by Walker (1987) for English, only a very small percentage of words with seven letters or fewer (maximum length analysed by Walker) have a unique shape. However, for words with between eight and thirteen letters, the percentage of words with a unique shape increases continuously from 12.3 to 60.8. As in English orthography, words with a sequence of neutral letters have the most frequent shape. The percentage of words with such a sequence ranged from 44% to 9% for two- to six-letter words, and from 6% to 1% for seven- to twelve-letter words. An important point to be noted is that all unmarked vowels belong to the set of neutral letters, whereas the ascending and descending letter sets contain only consonants. The set of marked letters, on the other hand, is made up entirely of vowels. Materials. Four categories of words were selected by taking all combinations of the two variables: Lexical frequency (high vs. low) and shape frequency (high vs. low). There were 15 words in each category with 5 words for each word length (seven, eight, and nine letters).

922

LEÂTEÂ AND PYNTE

The written frequency of occurrence of 24,121 words, as specified in the Imbs count (1971), was included in Content et al.'s (1990) word pool. For each word, the Imbs count gives the number of times it occurs per 100 million words in a corpus of texts from the second half of the twentieth century. The log of this count, multiplied by 100, was used to choose targets. This produced a range of 0±675 with a median of 222. We considered a word to be a rare word if its count was below 186, which represented 33% of the words in the pool. We considered a word to be a frequent word if its count was above 260, which again represented 33% of the words. Then from these two sets of words, targets with high- and lowfrequency shapes were selected. Lexically frequent targets ranged from 265 to 402, with a mean of 331. Lexically rare targets ranged from 0 to 180, with a mean of 118. High shape-frequency targets averaged 215 candidates (290 for seven-letter targets, 219 for eight-letter targets, and 135 for nine-letter targets). Low shape-frequency targets had a unique shape (just one candidate, the target itself). The four categories were controlled for bigram frequency (mean of 2.7 in each set, as per the Content et al., 1990 word pool) and number of syllables (mean of 2.9 in each set). No targets had any neighbours, as defined by Coltheart et al. (1977). (See Appendix 1 for a detailed list of the words and descriptive statistics.) Sixty nonwords, seven, eight, and nine letters long (twenty per length) were created. Half of them were illegal nonwords (random consonant strings) and half were legal nonwords created from a set of words by substituting two letters. Procedure and apparatus. Subjects were randomly assigned to either the uppercase group or the lowercase group. Stimuli were presented in a different random order to each subject. Subjects were given 20 practice trials (2 blocks of 10 trials) followed by 12 blocks of 10 experimental trials. The stimuli were displayed on a colour VGA monitor with a resolution of 640 6 480 pixels driven by a 80286-based microcomputer. They were presented in white on a black background, and were preceded by a 500 ms fixation point (an asterisk) in the middle of the screen followed by a 500 ms empty screen. The computer keyboard was used for responding. Errors were signaled by the message ``Wrong answer'' displayed under the target for 2 s. Subjects were allowed to rest after each block of trials. Mean decision latencies (on correct words and nonwords) and error rate were displayed during this time. Stimulus display and event timing were controlled by the TSCOPE unit developed by Haussmann (1992), whose timer offers better than millisecond precision. RTs were measured between the onset of target display on the screen and the subject's button pressing response. Testing lasted approximately 35 min.

WORD-SHAPE EFFECT

923

Results Response latency means and standard deviations for correct responses were first calculated for each subject across items over the experimental conditions. Any response that was more than 2.5 SDs above or below the subject mean was treated as an outlier. Outliers accounted for 3% of word responses and for 2% of nonword responses. For each cell (excluding outliers), we then calculated mean subject and item RTs, and mean subject and item error rates. Given that no experimental factor related to shape frequency was manipulated on nonword decisions, nonword data were not analysed in ANOVAs. The mean response latency was 826 ms for legal nonwords and 532 ms for illegal nonwords. Participants made fewer errors on illegal nonwords (1%) than on legal nonwords (21%). Word response latencies. For each of the two dependent variables a 2 (case) 6 2 (shape frequency) 6 2 (lexical frequency) ANOVA was conducted with subjects (F1) and items (F2) as random factors. Case (in the by-subject analyses) and shape frequency and lexical frequency (in the by-item analyses) were treated as between factors. Separate 2 (shape frequency) 6 2 (lexical frequency) ANOVAs were also conducted on the uppercase and lowercase groups. The mean correct response latencies and the mean error rates are shown in Table 2. The main effect of case was not significant on response latencies in either the subject or item analysis, F1 < 1; F2(1, 56) = 1.47, p > .10, MSE = 1402. Averaged response latencies on words with a low-frequency shape were 20 ms faster than those on words with a high-frequency shape. The main effect of shape frequency was highly significant in the by-subject analysis and approached significance in the by-item analysis, F(1, 48) = 19.40, p < .001, MSE = 1060; F2(1, 56) = 3.75, p = .058, MSE = 4571. Lexically frequent words were responded to 75 ms faster than lexically rare words, F1(1, 48) = 97.13, p < .001, MSE = 2909; F2(1, 56) = 38.94, p < .001, MSE = 4571. There was a 64-ms lexical TABLE 2 Mean lexical-decision latencies (by-subject analysis, in ms) and error rates (in parentheses) for words in Experiment 1, as a function of case, shape frequency, and lexical frequency Uppercase Lexical frequency

High-F shape

Low-F shape

Frequent Rare Effect

586 (2) 652 (13) 66 (11)

578 (3) 641 (9) 63 (6)

Lowercase Effect 78 (1) 711 (74)

High-F shape

Low-F shape

568 (2) 680 (11) 112 (9)

562 (2) 623 (11) 61 (9)

Effect 76 (0) 757 (0)

924

LEÂTEÂ AND PYNTE

frequency effect in the uppercase group, and a 86 ms effect in the lowercase group. The interaction did not reach significance, F1(1, 48) = 2.02, p > .10, MSE = 2909; F2(1, 56) = 2.33, p > .10, MSE = 4571. Finally, the three-way interaction between case, shape frequency, and lexical frequency was significant in the by-subject analysis and approached significance in the by-item analysis, F1(1, 48) = 5.07, p < .05, MSE = 1405; F2(1, 56) = 3.11, p = .08, MSE = 1402. This interaction reflects the fact that shape frequency and lexical frequency interacted in lowercase presentation, whereas the two factors had an additive effect in uppercase presentation. These relations between shape and lexical frequency were examined in more detail by analysing the data from each group separately. The analysis of the uppercase group's data revealed that the main effect of lexical frequency was significant (+64 ms), F1(1, 24) = 42.47, p < .001, MSE = 2436; F2(1, 56) = 22.04, p < .001, MSE = 3155, but none of the effects involving shape frequency were. In contrast, the analysis of the lowercase group's data revealed significant main effects of shape frequency (732 ms for words with a low-frequency shape), F1(1, 24) = 21.76, p < .001, MSE = 1141; F2(1, 56) = 6.84, p < .05, MSE = 2819, and lexical frequency (+86 ms), F1(1, 24) = 54.69, p < .001, MSE = 3382, F2(1, 56) = 39.34, p < .001, MSE = 2819, as well as a significant Shape frequency 6 Lexical frequency interaction, F1(1, 24) = 13.96, p < .001, MSE = 1184; F2(1, 56) = 5.08, p < .05, MSE = 2819. This interaction reflected the fact that the shape-frequency effect was facilitatory only for rare words. Alternatively, one can consider this interaction in terms of lexical frequency, in which lowercase words with a high-frequency shape exhibited a 112 ms lexical frequency effect, whereas words with a low-frequency shape were affected less by the lexical frequency manipulation (61 ms). A Newman-Keuls test on the Case 6 Shape frequency 6 Lexical frequency interaction showed that performance on rare words with high-frequency shapes was not as good in lowercase presentation as in uppercase presentation (+28 ms), p < .01 by subjects; p = .08 by items. No significant differences were found in the other three conditions. Word error rates. ANOVAs on error rates yielded a main effect of lexical frequency, F1(1, 48) = 109.23, p < .001, MSE = 37; F2(1, 56) = 29.44, p < .001, MSE = 83. Participants made fewer mistakes on frequent words (2% vs. 11%). No significant main effects or interactions involving case, shape frequency, or lexical frequency were obtained. Separate analyses of the data for the upper- and lowercase groups revealed a similar pattern of results.

Discussion Failure to obtain significant differences between uppercase words and lowercase words is inconsistent with the view that readers rely upon whole word shape in word recognition. Additionally, two main findings of the present experiment

WORD-SHAPE EFFECT

925

showed that features smaller than whole word shape are processed and mediated the identification of lowercase words. First, this shows that shape frequency contributed to word identification in a lexical-decision task: Words with a lowfrequency shape were recognized more quickly than words with a high-frequency shape only when shape information was available. Second, it showed that the processing of rare words with a high-frequency shape in lowercase presentation was at a disadvantage compared to rare words with a low-frequency shape. In other words, a high-frequency shape tended to jeopardize the least familiar words the most, which is again at odds with the idea that recognition rely wholly upon outline word shape, but is consistent with our hypothesis, which views shape frequency as an uncertainty-reducing factor in the process of lexical access (Paap et al., 1984). The first finding deviates from Paap et al.'s (1984) results. One could hypothesize that the discrepancy between the present results and Paap et al.'s lies in differences between the materials used in each case. Averaged across word length, the mean shape-frequency log of our 30 words classified as having high-frequency shapes was 2.33 compared to 1.66 for the 120 words in Paap et al. Their set of low shape-frequency words had a mean log of 0.78, whereas we chose only words with a unique shape (log of zero). Another interpretation, which does not rule out the first, can be derived from Walker (1987). He noted that each character in Paap et al. was created from a 5 6 9 dot matrix, with ascending and descending line segments extending just two dots above or below the 5 6 5 matrix used to generate the neutral letters. The matrix in the present experiment was a 7 6 15 dot matrix, with ascending line segments extending four dots above the 7 6 7 matrix used to generate the neutral letters, and with descending line segments extending three dots below. Consequently, the physical characteristics of the letters of our words were more enhanced, especially for the set of words with a low-frequency shape. The frequent words produced fast lexical access across the various case and shape-frequency conditions. We can speculate for our subjects that on familiar words, they did not delay their responses until the unique lexical code of a word had been accessed but responded instead on the basis of a reasonable level of lexical activation, while on rare words, they may have processed each item until lexical access was achieved. In other words, for skilled readers and familiar words, the extraction of visual letter information may be so quick that other types of information do not have any effect. However, with unfamiliar words, shape information may be important for inhibiting a set of candidates. Until now, we have ruled out the shape of the word outline as a source of information because no effects were found on familiar words. Letter features and/or multiletter features seem to be better candidates for producing a shape frequency effect, given that word-shape information is a resulting property of letter-shape information. Although it seems mandatory to postulate the processing of such visual features, it is not clear at what stage this processing

926

LEÂTEÂ AND PYNTE

occurs. Perhaps features contribute to initial lexical access, or perhaps they play a role during a subsequent stage of processing, such as verification (Paap et al., 1982). The aim of Experiment 2 was to study expectancy mechanisms produced by shape priming. We compared the performance of a shape priming group with a 400 ms SOA to that of a neutral-shape priming group. Shape information was given via priming with the shape of each target letter (see below for a description of the primes used).

EXPERIMENT 2: SHAPE PRIMING EFFECT ON LEXICAL DECISIONS If shape information enters into play at a relatively early stage in letter encoding, we can assume it is used to determine which letters are present in the word at each letter position, thereby limiting the set of possible letters. The results of this process may be passed on to the word-identification process, which would attempt to find a word that is consistent with one of the possible letters in each position. Computation of letter identities may be delayed by the informativeness of the visual features: A word with a low-frequency shape has a small set of candidate words, whereas a word with a high-frequency shape has a large set of candidate words, necessitating additional information processing and slowing down word identification. The uncertainty-reduction hypothesis (Paap et al., 1984) postulates that the locus of the shape-frequency effect is in the process that discriminates a word from among a set of candidates in the lexical system after computation of letter identities. In this case, giving shape information before the target is processed should be of little use. So a Shape frequency 6 Lexical frequency interaction is expected when there is shape priming, with the same pattern of results as in Experiment 1 for lowercase presentation. The informativeness hypothesis postulates that the locus of the shapefrequency effect is in the process that computes letter identities. Shape information is accumulated at a faster rate in a low-frequency shape word than in a high-frequency shape word because there are fewer candidate words. Consequently, shape priming should facilitate the encoding of the most ambiguous stimuli. Thus, if expectancies are used to process the upcoming target, words with a frequent shape should benefit more from priming than ones with an infrequent shape, giving a Prime condition 6 Shape frequency interaction. The informativeness hypothesis also predicts a Shape frequency 6 Lexical frequency interaction in the neutral-prime group, giving the same pattern of results as in Experiment 1, along with the disappearance of the interaction in the shape±prime group, giving a three-way Priming condition 6 Shape frequency 6 Lexical frequency interaction.

WORD-SHAPE EFFECT

927

Method Subjects. Fifty first- and second-year Psychology students at the University of Bourgogne took part in the experiment for course credit. All were native speakers of French and unaware of the purpose of the experiment. Materials. The targets were the words and nonwords used in Experiment 1. Primes for the neutral letters, ascending letters, descending letters, and marked letters (see Figure 1) were generated by creating a set of four ASCII characters from a matrix 7 pixels wide (columns) and 15 pixels high (rows). This matrix is the one used to generate the 255 characters of an IBM computer. The twelfth row (from the top) was the position of the letter on the line. For each character, the seven columns were always turned on but the on/off switching of the rows varied: For neutral letters, rows 6 to 12 were on (so a total of 49 pixels were on); for marked letters, rows 3 and 4 and rows 6 to 12 were on (63 pixels); for ascending letters, rows 3 to 12 were on (70 pixels); and for descending letters, rows 6 to 15 were on (70 pixels). Neutral primes were constructed from the same 7 6 15 pixel matrix. An ASCII character was created by turning on rows 3 to 15 (91 pixels were on).

Figure 1. The five letter primes used in Experiments 2 and 3. The matrix is 7 pixels wide (columns) and 15 pixels high (rows).

928

LEÂTEÂ AND PYNTE

Procedure and apparatus. The procedure and the timing of events were the same as in Experiment 1, except for the following points. In the shape-prime group, they were told they would see two items on each trial, one presented after the other. They were also told that the second item would be a word or a nonword, and that the first would be a string of four types of characters, each representing one letter of the second item. Then they were shown what kinds of letters each of the four characters represented. In the neutral-prime group, subjects were told that the first item would be a string with one character representing each letter in the second item. Finally, subjects were instructed to decide whether the second item was a word they knew and to respond as quickly and accurately as possible. The presentation sequence was the same as in Experiment 1: A 500 ms fixation point and a 500 ms empty interval before the appearance of the prime. The prime was displayed for 400 ms and followed by a 100 ms empty interval to reduce forward masking. The target then appeared in lowercase letters and remained on the screen until the subject responded. The fixation marker, the prime, and the target appeared in the centre of the screen.

Results Outliers (2.3% of word and nonword responses), mean subject and item RTs, and mean subject and item error rates were calculated as in Experiment 1. Word response latencies. For each of the two dependent variables a 2 (priming condition: Neutral prime vs. word-shape prime) 6 2 (shape frequency) 6 2 (lexical frequency) ANOVA was conducted with subjects (F1) and items (F2) as random factors. Priming condition (in the by-subject analyses) and shape and lexical frequency (in the by-item analyses) were treated as between factors. The mean correct response latencies and the mean error rates are shown in Table 3. TABLE 3 Mean lexical-decision latencies (by-subject analysis, in ms) and error rates (in parentheses) for words in Experiment 2, as a function of priming condition, shape frequency, and lexical frequency Neutral prime Lexical frequency

High-F shape

Low-F shape

Frequent Rare Effect

541 (4) 628 (10) 87 (6)

536 (2) 586 (10) 50 (8)

Shape prime Effect 75 (0) 742 (0)

High-F shape

Low-F shape

524 (2) 606 (11) 82 (9)

520 (1) 602 (6) 82 (5)

Effect 74 (71) 74 (75)

WORD-SHAPE EFFECT

929

The shape-priming group exhibited a 9 ms facilitatory effect on RTs data. This effect did not reach significance, F1 < 1, MSE = 23,903; F2(1, 56) = 2.46, p > .10, MSE = 3526. Averaged response latencies on words with a low-frequency shape were 14 ms faster than those on words with a high-frequency shape. The main effect of shape frequency was only significant in the by-subject analysis, F1(1, 48) = 10.01, p < .005, MSE = 947; F2(1, 56) = 2.47, p > .10, MSE = 579. A significant Priming condition 6 Shape frequency interaction confirmed that words with a high-frequency shape had benefited more from the shape priming than words with a low-frequency shape (19 ms vs. 0 ms, respectively), F1(1, 48) = 5.13, p < .05, MSE = 947; F2(1, 56) = 4.05, p < .05, MSE = 579. Frequent words were responded to 75 ms faster than rare words, F(1, 48) = 112.76, p < .001, MSE = 2498; F2(1, 56) = 52.76, p < .001. MSE = 3526. No significant interaction was found with the prime condition factor. Finally, the three-way interaction between prime condition, shape frequency, and lexical frequency revealed that the Shape frequency 6 Lexical frequency interaction in the neutral-prime condition disappeared when word shape was primed, F1(1, 48) = 6.21, p < .05, MSE = 695; F2(1, 56) = 4.11, p < .05, MSE = 579. As in Experiment 1 (lowercase group), in the neutral-prime group, recognition of a rare word with a high-frequency shape was poorer than it was for a rare word with a low-frequency shape (628 ms vs. 586 ms, respectively). However, when shape information was given to the subject, this disadvantage no longer existed (606 ms vs. 602 ms, respectively). A separate ANOVA on the shape-prime group's data revealed a significant main effect of lexical frequency (+82 ms), F1(1, 24) = 96.53, p < .001, MSE = 1732; F2(1, 56) = 57.30, p < .001, MSE = 1987, but none of the effects involving shape frequency were significant. In contrast, the analysis of the neutral-prime group's data revealed significant main effects of shape frequency (723 ms for words with a low-frequency shape), F1(1, 24) = 13.27, p < .005, MSE = 1051; F2(1, 56) = 4.73, p < .05, MSE = 2118, and lexical frequency (+69 ms), F1(1, 24) = 35.77. p < .001, MSE = 3265; F2(1, 56) = 35.06, p < .001, MSE = 2118. The Shape frequency 6 Lexical frequency interaction was significant in the by-subject analysis, F1(1, 24) = 11.97, p < .005, MSE = 708; and was marginally significant in the by-item analysis, F2(1, 56) = 3.00, p = .09, MSE = 2118. As in Experiment 1 (lowercase group), this interaction reflected the fact that the shape-frequency effect was facilitatory only for rare words (42 ms). Alternatively, one can consider this interaction in terms of lexical frequency, in which case words with a highfrequency shape exhibited a 87 ms lexical frequency effect, whereas words with a low-frequency shape were affected less by the lexical frequency manipulation (50 ms). Word error rates. The error data yielded a main effect of lexical frequency, F1(1, 48) = 52.82, p < .001, MSE = 51; F2(1, 56) = 18.15, p < .001, MSE = 89. The main effect of shape frequency was significant in the by-subject analysis

930

LEÂTEÂ AND PYNTE

(fewer mistakes were made for words with a low-frequency shape), F(1, 48) = 4.91, p < .05, MSE = 36; F2(1, 56) = 1.18, p > .10, MSE = 89. The critical Priming condition 6 Lexical frequency 6 Shape frequency interaction approached significance in the by-subject analysis, F1(1, 48) = 3.65, p = .062, MSE = 35; and was significant in the by-item analysis; F2(1, 56) = 4.27, p < .05, MSE = 18. In the neutral-prime group, there were no differential effects of shape frequency on accuracy for either of the lexical-frequency conditions. In the shape-prime group, rare words benefited from prior shape information only when a low-frequency shape was displayed. Nonword data. A 2 (priming condition) 6 2 (legality) ANOVA was conducted with subjects (F1) and items (F2) as random factors. Priming condition (in the by-subject analysis) and legality (in the by-item analysis) were treated as between factors. The response-latency ANOVA yielded a main effect of nonword legality (495 ms vs. 739 ms), F1(1, 48) = 206.09, p < .001, MSE = 6353; F2(1, 58) = 485.97, p < .001, MSE = 3833. Nonwords were rejected more rapidly in the shape-prime group than in the neutral-prime group (712 ms by subjects, and 714 ms by items), giving a main effect of priming only in the by-item analysis, F1 < 1; F2(1, 58) = 6.43, p < .05, MSE = 906. The priming effect was only present for legal nonwords (724 ms by subjects, and 728 ms by items), but the interaction was only significant in the analysis by items, F1 < 1; F2(1, 58) = 6.05, p < .05, MSE = 906. Error rates showed a significant effect of legality with fewer errors for illegal nonwords (1% vs. 16%), F1(1, 48) = 49.36, p < .001, MSE = 111; F2(1, 58) = 88.73, p < .001, MSE = 86. There was also a main effect of priming, as well as an interaction between the two factors, but these effects were only significant in the by-item analysis, F1(1, 48) = 1.63, p > .10, MSE = 118; F2(1, 58) = 7.92, p < .01, MSE = 29; F1(1, 48) = 1.24, p > .10, MSE = 111; F2(1, 58) = 5.54, p < .05, MSE = 29, respectively. There were fewer errors in the shape-prime group (7%) than in the neutral-prime group (10%); and the facilitative effect of priming was only present for legal nonwords (75.5% vs. 70.5%).

Discussion Experiment 2 revealed that shape information tended to facilitate target identification for words presented in lowercase, especially when the shape corresponded to a larger set of candidate words. Indeed, high-frequency shapes profited more from the priming procedure than low-frequency ones. There was an early contribution of shape information because the shape effect disappeared on target response latencies when shape information was primed during 400 ms

WORD-SHAPE EFFECT

931

SOA. This is at odds with the uncertainty-reduction hypothesis, which stipulates that the shape-frequency effect lies in the process which discriminates a word among a candidate set in the lexical system, after computation of letter identities. It can be hypothesised that the 400 ms SOA was used by subjects to generate an appropriate expectancy about the identities of the letters in the upcoming word. Computations of letter identities may therefore be facilitated and passed on to the word identification process, which then attempts to find a word consistent with one of the possible letters in each position. This causes the shapefrequency effect to disappear without affecting the lexical-frequency effect, because lexical access is accomplished during the presentation of the target. Low-frequency words with rare shapes suffered a 16 ms disruption of shape priming. One possible explanation is that rare shapes enabled participants to generate expectancies that were valid for high-frequency words, but not for lowfrequency words because they were not easily accessible during the 400 ms SOA. High-frequency words with a similar shape were likely to be activated and there were some costs involved in rejecting these candidates in favour of the actual target. Overall, the word data of Experiment 2 suggested the early contribution of shape information to word recognition. This is consistent with the nonword data, which exhibited a shape-priming effect in the by-item analyses: Participants rejected legal nonwords more rapidly and with more accuracy than illegal nonwords when the shape of the upcoming target was primed. However, these results are not conclusive, because it is difficult to be certain whether the observed effects took place at the perceptual or inferential level. It would be useful to obtain further evidence by manipulating a short SOA. Moreover, caution must be taken regarding the task used. Indeed, Besner (1983) claimed that the lexical decision task involves a type of recognition mechanism that simply monitors the visual familiarity of the target without uniquely specifying the word (see Balota & Chumbley, 1984 for a similar assumption based on the familiarity/meaningfulness dimension). Although the present lexical-decision tasks required excellent information about letter identities because half of the nonwords were constructed from a set of words, this task can be suspected to require only shallow processing; namely, it does not necessarily involve the completion of lexical access and it treats printed stimuli as visual patterns rather than as individual letters. Thus, to be able to contend that the effects observed in our previous experiments were not task dependent, we decided to test for the use of shape information in a task (naming) that requires mapping letters to speech and uniquely specifying the stimulus. As in Experiment 2, we looked at expectancy mechanisms produced by shape priming with a baseline condition generated by a neutral prime.

932

LEÂTEÂ AND PYNTE

EXPERIMENT 3: SHAPE PRIMING EFFECT ON NAMING Method Subjects. Thirty-six first- and second-year Psychology students at the University of Provence took part in the experiment. All were native speakers of French and unaware of the purpose of the experiment. Materials.

The stimuli were the 60 words used in Experiment 1.

Procedure and apparatus. Procedure and apparatus were the same as in Experiment 2. Nonwords were excluded from stimuli. Participants were given one block of 10 practice trials followed by six blocks of 10 experimental trials. They were instructed to pronounce the presented words as quickly and as accurately as possible. Naming latencies were measured from the onset of target display on the screen to the onset of naming. Testing lasted approximately 15 min.

Results Outliers accounted for 3% of the naming latencies. ANOVAs were conducted as in Experiment 2. The mean naming latencies are shown in Table 4. Shape priming in the naming task had a 74 ms facilitatory effect. This effect was marginally significant in the by-subject analysis and was highly significant in the by-item analysis, F1(1, 34) = 3.69, p = .063, MSE = 52873; F2(1, 56) = 200.30, p < .001, MSE = 850. No main effect of shape frequency was found in either the by-subject analysis (78 ms), F1(1, 34) = 3.05, p > .09, MSE = 662, or the by-item analysis (75 ms), F2 < 1. A significant main effect of lexical TABLE 4 Mean naming latencies (by-subject analysis, in ms) in Experiment 3, as a function of priming condition and lexical frequency Neutral prime Lexical frequency

High-F shape

Low-F shape

Frequent Rare Effect

592 693 101

620 689 69

Shape prime Effect 28 74

High-F shape

Low-F shape

552 596 57

553 601 50

Effect 1 5

WORD-SHAPE EFFECT

933

frequency showed that frequent words were named 66 ms faster than rare words, F1(1, 34) = 76.22, p < .001, MSE = 2020; F2(1, 56) = 35.40, p < .001, MSE = 3861. Word-shape priming produced a 54 ms facilitatory effect on frequent words and a 93 ms facilitatory effect on rare words. Consequently, the lexical-frequency effect was lower in the shape-prime group (+46 ms) than in the neutralprime group (+85 ms), F1(1, 34) = 6.83, p < .05, MSE = 2020; F2(1, 56) = 17.05, p < .001, MSE = 850. The Lexical frequency 6 Shape frequency interaction was modulated by the type of prime used because the three-way interaction between group, shape frequency, and lexical frequency was reliable in the by-subject analysis and marginally significant in the by-item analysis, F1(1, 34) = 4.55, p < .05, MSE = 664; F2(1, 56) = 3.37, p = .071, MSE = 850. Separate ANOVAs revealed that, for the shape-prime group, the main effect of lexical frequency was significant, F1(1, 17) = 31.56, p < .001, MSE = 1198; F2(1, 56) = 24.78, p < .001, MSE = 1254, but none of the effects involving shape frequency were. The analysis of the neutral-prime group's data yielded no main effect of shape frequency in either the subject analysis or the item analysis, F1(1, 17) = 2.59, p > .10, MSE = 921; F2 < 1, but a significant main effect of lexical frequency, F1(1, 17) = 45.73, p < .001, MSE = 2842; F2(1, 56) = 34.74, p < .001, MSE = 3457. The interaction between shape frequency and lexical frequency was significant in the subject analysis, F1(1, 17) = 5.85, p < .05, MSE = 822; F2(1, 56) = 1.18, p > .10, MSE = 3457. This interaction reflected the fact that frequent words were subject to an inhibitory shape-frequency effect (+28 ms), whereas rare words were not affected by the shape frequency manipulation. If we consider this interaction in terms of lexical frequency, words with a high-frequency shape exhibited a 101 ms lexical frequency effect, whereas words with a low-frequency shape were less affected by the lexical frequency manipulation (69 ms), giving the same pattern of results as in Experiment 1 for lowercase presentation.

Discussion Shape information that appeared in the primes produced a strong facilitation effect. Newman-Keuls tests showed that the facilitatory effect was reliable for all four target categories (p < .001, by subjects and items). We also found a lexical-frequency effect that was about twice as small in the shape-prime group as in the neutral-prime group (54 ms vs. 85 ms, respectively). The 85 ms effect was about the same as in Experiment 1 for lowercase presentation (87 ms). The present finding is inconsistent with previous research in English showing that the lexical-frequency effect is much larger in a lexicaldecision task than in a naming task (Balota & Chumbley, 1984, 1985; Chumbley

934

LEÂTEÂ AND PYNTE

& Balota, 1984; Forster & Chambers, 1973; Frederiksen & Kroll, 1976). Furthermore, it goes against the controversial accounts proposed by Balota and Chumbley (1984, 1985) and other authors (Besner & McCann, 1987; McCann & Besner, 1987; McCann, Besner, & Davelaar, 1988), who contend that processes subsequent to lexical identification, even ones that follow access to meaning or pronunciation, might be the locus of lexical-frequency effects. The neutral-prime group exhibited an interaction between shape frequency and lexical frequency with no shape effect for rare words and a shape effect for frequent words. The interaction reached significance in the by-subject analysis but not in the by-item analysis. A Newman-Keuls test performed across subjects showed that frequent words with a high-frequency shape were named faster than frequent words with a low-frequency shape (p < .01). The naming process is commonly viewed as involving either a lexical transcoding process (the addressed route) or a sublexical transcoding process (the assembled route). Lexical transcoding refers to identifying a word from its learned orthographic pattern and retrieving its learned pronunciation; sublexical transcoding refers to assigning a phonological representation to the orthographic subcomponents of the letter string and assembling them into a phonological code. The time needed for sublexical transcoding is commonly thought to be longer than that required for addressing the phonological representation in the lexicon. Several studies have suggested that the pronunciation of rare words relies on sublexical transcoding, whereas familiar words and words with exceptional spellings are read primarily through the addressed route (Seidenberg, 1985; Seidenberg, Waters, Barnes, & Tanenhaus, 1984; Taraban & McClelland, 1987). How can shape priming facilitate the recognition of a word in a naming task? In the light of the above theoretical considerations, it could be hypothesized that when a prime with a low-frequency shape is displayed, the pattern of ascending, descending, and marked letters supplies partial information about the syllabic structure of the upcoming word (such as syllable boundaries), which can be useful for grouping letters in order to generate the articulatory-motor program. For example, an (ascender/descender)± neutral±(ascender/descender) sequence of letters can easily be translated into a (consonant)±(vowel)±(consonant) sequence, which provides information about the syllabic structure of the word. Consequently, word-shape priming information in this experiment could have promoted the use of a sublexical-transcoding strategy because the prime forced reliance upon the syllabic structure of the upcoming target. In contrast, in the neutral-prime group, a lexical transcoding strategy can be hypothesized. This syllable boundary argument is in line with Mayall et al.'s (1997) results, which showed that one possible influence on case-mixing effects in a naming task results from inappropriate letter grouping, which suggests that reading aloud uses visually based letter clusters.

WORD-SHAPE EFFECT

935

EXPERIMENT 4: CONTROLLING FOR THE UP±DOWN CONFIGURATION OF LOW-FREQUENCY SHAPE WORDS IN A LEXICAL-DECISION TASK Until now, we contended that the up±down configuration is the essential property of words with a low-frequency shape, and we did not match our stimuli with respect to the factors as Paap et al. (1984) did: (1) The two words (common vs. rare-shape word) had the same number of ascending, descending, and neutral letters; (2) the two words had at least three letters in common for four- to fiveletter words, and at least four letters in common for six-letter words; and (3) the two words had the same number of repeated letters (e.g., recall paired with cellar). The problem with this type of control is that it can produce items that have inhibitory connections due to shared orthographic neighbourhoods. Indeed, restrictions (2) and (3) lead to enhanced orthographic similarity between the sets of words. (Note that with four- to six-letter words orthographic similarities are even higher than with our seven- to nine-letter stimuli). Landauer and Streeter (1973) already pointed out that such a similarity neighbourhood would have an effect on the ease of word recognition, and that the number and characteristics of such neighbours, e.g., their frequency of usage in the language, would affect recognition. Since then, some investigators have reported that large neighbourhoods (in terms of the size of a word's orthographic neighbourhood) facilitate lexical access (Andrews, 1989, 1992; Pynte, 2000; Sears et al., 1995), while others have reported that higher frequency neighbours (in terms of the frequency of a word's orthographic neighbours) delay lexical access (Grainger, 1990; Grainger, O'Regan, Jacobs, & Segui, 1989; Grainger & Segui, 1990). Consequently, because competition between words is now a critical requirement in word-recognition models, the orthographic neighbourhood of targets was controlled here, along with other information provided by their orthographic form. The targets were strictly controlled for bigram frequencies and number of syllables, and none of the targets had any neighbours, as defined by Coltheart et al. (1977). This minimized the possibility of a type 2 error (i.e., failure to reject an incorrect null hypothesis). However, it can be argued that the up±down configuration in our word sets may have influenced the perceptibility of individual letters because ascending and descending letters may be easier to discriminate than neutral ones. Indeed, letter recognition studies have generally found that letters that ``look alike'' are often confused, and letters that do not look alike are rarely confused (for a review, see Appelman & Mayzner, 1982). In other words, any effect of word shape in our experiments may have been due to letter-linked differences. So, instead of a shape-frequency factor, a letter-confusability factor could have been manipulated with high letter-confusability words (e.g., high-frequency shape words) vs. low letter-confusability words (i.e., low-frequency shape words).

936

LEÂTEÂ AND PYNTE

An important finding to keep in mind in advocating this confounding-variable idea, is the Shape frequency 6 Lexical frequency interaction found in lowercase presentation. We must explain why confusability jeopardizes the least familiar words the most. The hypothesized letter-confusability effect could be related to a visual degradation effect, and previous studies with manipulation of stimulus quality and lexical frequency have always yielded additive effects: Degradation impairs performance equally for low- and high-lexical-frequency words (Balota & Abrams, 1995; Becker & Killion, 1977, Exps. 3 and 4; Borowsky & Besner, 1993; Plourde & Besner, 1997). Nevertheless, to test whether the word-shape effect could be linked to a letterconfusability effect, a lexical-decision task was run with number of descending, ascending, neutral, and marked letters matched across the four sets of experimental words, as in Paap et al.'s (1984) study. In order to minimize orthographic similarity between the target words, six-, seven-, and eight-letter words were chosen.

Method Subjects. Twenty first- and second-year Psychology students at the University of Provence took part in the experiment for course credit. All were native speakers of French and unaware of the purpose of the experiment. Materials. Table 5 presents characteristics of the stimuli (see Appendix 2 for a detailed list of the words). The four sets of words were selected from Content et al.'s (1990) word pool by taking all combinations of the lexical frequency (frequent vs. rare) and shape frequency (high vs. low) variables. There were 18 words in each set with 6 words for each word length (six, seven, and eight letters). The letter range covered about 43% of the entries in the Content et al. word pool (47% in the previous word sets). Number of words per syntactic class was not strictly controlled across sets. There were about 10 nouns and 8 verbs in each set. The sets were also controlled for bigram frequency (mean TABLE 5 Characteristics of the words tested in Experiment 4 High-frequency shape

Mean bigram frequency Mean shape frequency (range) Mean lexical frequency (range) Number of A-M-D-N letters

Frequent

Rare

2.9 42 (27±53) 367 (263±501) 20-2-16-88

2.8 41 (18±53) 115 (0±190) 20-2-16-88

See Appendix 2 for all sets of words.

Low-frequency shape Frequent

Rare

2.7 2.5 4 (1±9) 4 (1±9) 371 (269±474) 148 (108±183) 20-2-16-88 20-2-16-88

WORD-SHAPE EFFECT

937

2.4±2.9). Words with orthographic neighbours were not strictly excluded as in the previous experiments because of difficulty obtaining a large enough word set. The average number of orthographic neighbours per target word in each cell of the experimental design was 0.8. The value was considered as acceptable to prevent orthographic neighbourhood effect on response latencies. Words were matched across the four sets on number of ascending, descending, neutral, and marked letters. In each set, there was the same number of neutral letters (88), ascending letters (20), descending letters (2), and marked letters (2). High shape-frequency targets averaged 40 candidate words. Low shape-frequency targets averaged four candidate words. The frequency of lexically frequent targets ranged from 263 to 501 (as defined in Experiment 1), with a mean of 369. Lexically rare targets ranged in frequency from 0 to 190, with a mean of 131. Seventy-two six-, seven-, and eight-letter nonwords (twenty-four per length) were created. Half of them were illegal nonwords (random consonant strings) and half were legal nonwords. Procedure and apparatus. The procedure and apparatus were the same as in Experiment 1 in the lowercase presentation condition.

Results For each cell of the experimental design, mean subject and item RTs, and mean subject and item error rates were calculated. For each of the two dependent variables, a 2 (shape frequency) 6 2 (lexical frequency) ANOVA was conducted with subjects (F1) and items (F2) as random factors. Shape frequency and lexical frequency (in the by-item analysis) were treated as between factors. Given that no experimental factor related to shape frequency was manipulated on nonword decisions, the nonword data were not analysed in ANOVAs. The mean response latency were 785 ms for legal nonwords and 508 ms for illegal nonwords. Participants made fewer errors on illegal nonwords (1%) than on legal nonwords (21%). For word decisions, the mean correct response latencies and the mean error rates are shown in Table 6.

TABLE 6 Mean lexical-decision latencies (by-subject analysis, in ms) and error rates (in parentheses) in Experiment 4, as a function of shape frequency and lexical frequency

Lexical frequency Frequent Rare Effect

High-F shape

Low-F shape

556 (2) 697 (21) 141 (19)

552 (2) 650 (11) 98 (9)

Effect 74 (0) 747 (710)

938

LEÂTEÂ AND PYNTE

Response latencies. Averaged response latencies on words with a lowfrequency shape were 25 ms faster than those on words with a high-frequency shape. The main effect of shape frequency was significant, F1(1, 19) = 7.85, p < .05, MSE = 1653; F2(1, 68) = 5.01, p < .05, MSE = 1914. Lexically frequent words were responded to 120 ms faster than lexically rare words, F1(1, 19) = 152.84, p < .001, MSE = 1871; F2(1, 68) = 150.82, p < .001; MSE = 1914. Finally, the interaction between shape frequency and lexical frequency was significant in both analyses, F1(1, 19) = 9.45, p < .01, MSE = 1014; F2(1, 68) = 4.47, p < .05, MSE = 1914. As in the previous experiments, this interaction reflected the fact that the shape-frequency effect was facilitatory only for rare words (747 ms). Alternatively, if we consider this interaction in terms of lexical frequency, words with a high-frequent shape exhibited a 141 ms lexical frequency effect, whereas words with a low-frequency shape were affected less by the lexical frequency manipulation (98 ms). Error rates. The main effect of lexical frequency was highly significant, F1(1, 19) = 74.65, p < .001, MSE = 52; F2(1, 68) = 22.04, p < .001, MSE = 158. Participants made fewer mistakes on frequent words (2% vs. 16%). Contrary to the lexical-decision tasks in Experiments 1 and 2, a main effect of shape frequency was found, as well as a significant interaction between shape frequency and lexical frequency in the by-subject analysis, F1(1, 19) = 8.79, p < .01, MSE = 55; F2(1, 68) = 2.76, p = .10, MSE = 158; and F1(1, 19) = 14.79, p < .01, MSE = 32; F2(1, 68) = 2.71, p = .10, MSE = 158, respectively. Participants made fewer mistakes on words with a low-frequency shape (6%) than on words with a high-frequency shape (11%). As for the RT data, the shape-frequency effect was facilitatory only for rare words (710%). Shape and length effects. An ANOVA was conducted on RT data with length (i.e., six, seven, or eight letters) as a third factor. It was treated as a between factor in the by-item analysis. The ANOVA yielded the same effects as in the previous analyses, i.e., a main effect of lexical frequency, F1(1, 18) = 155.05, p < .001, MSE = 5359; F2(1, 60) = 153.92, p < .001, MSE = 1876; a main effect of shape frequency, F1(1, 18) = 7.95, p < .05, MSE = 5086; F2(1, 60) = 5.11, p < .05, MSE = 1876; as well as an interaction between the two factors, F1(1, 18) = 6.35, p < .05, MSE = 3820; F2(1, 60) = 4.56, p < .05, MSE = 1876. The critical interaction between Shape frequency 6 Length, F1(2, 36) = 1.06, p > .10, MSE = 2948; F2 < 1, showed that the shape effect did not vary across lengths. On rare words, there was a shape effect of 731 ms for six-letter words, 747 ms for seven-letter words, and 763 ms for eightletter words. On frequent words, the shape effect was 71 ms, 735 ms, and +17 ms, respectively.

WORD-SHAPE EFFECT

939

Discussion Experiment 4 replicated the results obtained in Experiment 1 in the lowercase presentation condition, and was consistent with the findings obtained for shape priming in a lexical-decision task (Experiment 2) and in a naming task (Experiment 3). These results rule out the view that the up±down configuration of low-frequency shape words can explained the shape effect. When word-shape distinctiveness was controlled, a shape-frequency effect was obtained only for low-frequency words. In line with Paap et al.'s (1984) view that a critical determinant of the effect of a word's shape is the number of other words that share the same shape, it seems that the shape effect is related to the shape information carried by the letters at each position for discriminating among the alternative candidate words.

GENERAL DISCUSSION The present results can be summarized as follows. In a lexical-decision task (Experiment 1, lowercase condition), a rare word with a pattern of ascending and descending letters tended to be responded to faster than a rare word with a pattern of neutral letters (757 ms on average); no difference was found for frequent words. The results cannot be explained by the up±down configuration of low-frequency shape words because when this factor was controlled, the same pattern of results was obtained (Experiment 4). In a naming task (Experiment 3, neutral condition), a frequent word with a pattern of ascending and descending letters tended to be responded to more slowly than a frequent word with a pattern of neutral letters (+28 ms on average); no difference was found for rare words. In a lexical-decision task (Experiment 2), shape priming information tended to facilitate the processing of the most impoverished words in terms of their features (words with a neutral letter pattern). In a naming task (Experiment 3), shape priming information tended to facilitate the processing of the most difficult words (rare words). In short, in a task that requires good knowledge of letter identities (lexical-decision task), priming facilitates the processing of the most ambiguous source of information (frequent-shape words), whereas in a task that requires good knowledge of word identities (naming task), priming facilitates the processing of the least common orthographic patterns (rare words). Taken together, the present results argue in favour of the idea that shape information plays a role in the word-recognition process. As far as we know, this is the first demonstration of a shape-frequency effect in word recognition. The fact that word recognition takes advantage of shape information is of theoretical interest because most models are based on the assumption that performance is influenced at any given time by a single source of informationÐthe letter.

940

LEÂTEÂ AND PYNTE

We stated here that shape information can be seen as the set of all visual features of a word, including single-letter patterns and/or sublexical multiletter patterns, but excluding the global characteristics of the word such as its outline, because no effect was found on familiar words in the lexical-decision task used here. We cannot assume that shape information serves as an access unit in the word-recognition process, but it does seem to furnish a certain amount of information that facilitates word parsing. Letter features and/or multiletter features seem to be candidates for producing a shape-frequency effect, given that word-shape information is a resulting property of letter-shape information. Our results are consistent with Paap et al.'s (1984) line of reasoning, which assumes that a critical determinant of the utility of a word's shape is the number of other words that share the same shape. If the shape of a word matches only one word, fast lexical access can occur with little additional information about the individual letters. If the shape matches many candidates, its usefulness is reduced and readers must rely upon letter identification. Thus, word shape facilitates word recognition only to the extent that it serves to preclude all but a relatively small set of candidate words. Paap et al.'s (1984) uncertainty-reduction hypothesis clearly places the locus of the shape-frequency effect at the level of lexical identification, where candidates are compared to find a lexical entry that best matches the input, as in Paap et al.'s (1982) activation-verification model. However, the results of Experiment 2 suggest that shape information enters into play at a relatively early stage in letter encoding. We postulated that it is used to determine which letter is present in the word at each letter position, thereby limiting the set of possible letters. The output of this process may be passed on to the word-identification process, which would attempt to find a word that is consistent with one of the possible letters in each position. Given that a word with a low-frequency shape has a small set of candidate words, whereas a word with a high-frequency shape has a large set of candidate words, additional information processing is needed in the latter case, slowing down the word-identification process. Of primary interest for students of today's reading research is the problem of case manipulation (Experiment 1), which leads to different patterns for lexical and shape effects. Underwood and Bargh (1982) using a naming task showed that differences in naming latencies between regular and irregular words were observed for uppercase words only. They suggested that the removal of the word envelope leads to greater dependency upon other intrinsic and extrinsic information. The richer graphemic information of lowercase letters may allow for identification on the basis of visual features alone, whereas identification of words in uppercase letters may depend upon left-to-right scanning. In an alphabetic decision task, Whiteley and Walker (1994, 1997) showed that bigram priming no longer had an effect when targets appeared in lowercase or in alternated case, and was replaced by letter priming. They argued that the disappearance of bigram priming was due to the removal of supraletter features,

WORD-SHAPE EFFECT

941

which normally contribute to the direct activation of bigram units. As stated in Experiment 1 and pointed out by Whiteley and Walker (1997), one way to reconcile our results with preliminary letter identification models of word recognition is to consider that case manipulation takes effect at the letter level because the computation of uppercase letter identities is made more difficult by the reduced salience of the individual letters, and, as Paap et al. (1984) proposed, by the greater degree of uncertainty associated with letter identification. In summary, these experiments show that the shape-frequency effect must be understood as a complex variable, and thereby suggest that there are many different ways in which shape information can participate in the activation of lexical representations. Here, we contend that printed words are recognized via a variety of units, ranging from word and/or letter features and sublexical letter units, to single letters or activated abstract letter identities. In this view, reading letters and words is basically a pattern-recognition problem. The key assumption is that shape information generates a sufficient level of discriminability among competing letters and sublexical units. Future experimentation should help provide more details for understanding the processes that are competing in a shape neighbourhood.

REFERENCES Adams, M. J. (1979). Models of word recognition. Cognitive Psychology, 11, 133±176. Allen, P. A., & Madden, D. J. (1990). Evidence for a parallel input serial analysis model of word processing. Journal of Experimental Psychology: Human Perception and Performance, 16, 48±64. Allport, D. A. (1977). On knowing the meaning of words we are unable to report: The effects of visual masking. In S. Dornicà (Ed.), Attention and performance VI. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Andrews, S. (1989). Frequency and neighborhood effects on lexical access: Activation or search? Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 802±814. Andrew, S. (1992). Frequency and neighborhood effects on lexical access: Lexical similarity or orthographic redundancy? Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 234±254. Appelman, I. B., & Mayzner, M. S. (1982). Application of geometric models to letter recognition: Distance and density. Journal of Experimental Psychology: General, 111, 60±100. Balota, D. A., & Abrams, R. A. (1995). Mental chronometry: Beyond onset latencies in the lexical decision task. Journal of Experimental Psychology: Learning, Memory and Cognition, 21, 1289±1302. Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage. Journal of Experimental Psychology: Human Perception and Performance, 10, 340±357. Balota, D. A., & Chumbley, J. I. (1985). The locus of word-frequency effects in the pronunciation task: Lexical access and/or production? Journal of Memory and Language, 24, 89±106. Becker, C.A., & Killion, T. H. (1977). Interaction of visual and cognitive effects in word recognition. Journal of Experimental Psychology: Human Perception and Performance, 3, 389±401. Besner, D. (1983). Basic decoding components in reading: Two dissociable feature extraction processes. Canadian Journal of Psychology, 37, 429±438.

942

LEÂTEÂ AND PYNTE

Besner, D. (1989). On the role of outline shape and word-specific visual pattern in the identification of function words: NONE. Quarterly Journal of Experimental Psychology, 41A, 91±105. Besner, D., & Johnston, J. C. (1989). Reading and the mental lexicon: On the uptake of visual information. In W. Marslen-Wilson (Ed.), Lexical representation and process (pp. 291±316). Cambridge, MA: MIT Press. Besner, D., & McCann, R. S. (1987). Word frequency and pattern distortion in visual word identification and production: An examination of four classes of models. In M. Coltheart (Ed.), Attention and performance XII: The psychology of reading. Hove, UK: Lawrence Erlbaum Associates Ltd. Borowsky, R., & Besner, D. (1993). Visual word recognition: A multistage activation model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 813±840. Bouma, H. (1971). Visual recognition of isolated lower-case letters. Vision Research, 11, 459±474. Chumbley, J. I., & Balota, D. A. (1984). A word's meaning effects the decision in lexical decision. Memory and Cognition, 12, 590±606. Coltheart, M., Davelaar, E., Jonasson, J. T., & Besner, D. (1977). Access to the internal lexicon. In S. Dornicà (Ed.), Attention and performance VI (pp. 535±555). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Content, A., Mousty, P., & Radeau, M. (1990). Brulex: Une base de donneÂes lexicale informatiseÂe pour le lexique francËais eÂcrit et parle [Brulex: A computerized lexical database for oral and written French lexicon]. L'AnneÂe Psychologique, 90, 551±566. Dechant, E. V., & Smith, H. P. (1977). Psychology in teaching reading. Englewood Cliffs, NJ: Prentice-Hall. Drewnowski, A., & Healy, A. F. (1977). Detection errors on the and and: Evidence for reading units larger than the words. Memory and Cognition, 5, 636±647. Faust, M., Babkoff, H., & Avidor-Reiss, I. (2000). Sentence and word outline shape as co-primes for target words presented to the two visual hemifields. Brain and Language, 73, 50±61. Forster, K. I., & Chambers, S. M. (1973). Lexical access and naming time. Journal of Verbal Learning and Verbal Behavior, 12, 627±635. Frederiksen, J. R., & Kroll, J. F. (1976). Spelling and sound: Approaches to the internal lexicon. Journal of Experimental Psychology: Human Perception and Performance, 2, 361±379. Gibson, E. J. (1969). Principles of perceptual learning and development. New York: Academic Press. Grainger, J. (1990). Word frequency and neighborhood frequency effects in lexical decision and naming. Journal of Memory and Language, 29, 228±244. Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual word recognition: A multiple read-out model. Psychological Review, 93, 46±74. Grainger, J., O'Regan, K., Jacobs, A. M., & Segui, J. (1989). On the role of competing word units in visual word recognition: The orthographic neighborhood effect. Perception and Psychophysics, 45, 189±195. Grainger, J., & Segui, J. (1990). Neighborhood frequency effects in visual word recognition: A comparison of lexical decision and masked identification latencies. Perception and Psychophysics, 47, 191±198. Groff, P. (1975). Research in brief: Shape as cues to word recognition. Visible Language, 9, 67±75. Haber, L. R., Haber, R. N., & Furlin, K. R. (1983). Word length and word shape as sources of information in reading. Reading Research Quarterly, 18, 165±189. Haber, R. N., & Schindler, R. M. (1981). Error in proofreading: Evidence of syntactic control of letter processing? Journal of Experimental Psychology: Human Perception and Performance, 7, 573±579. Haussmann, R. E. (1992). Tachistoscopic presentation and millisecond timing on the IBM/XT/AT and PS/2: A Turbo Pascal unit to provide general-purpose routines for CGA, Hercules, EGA, and VGA monitors. Behavior Research Methods Instruments and Computers, 24, 303±310.

WORD-SHAPE EFFECT

943

Healy, A. F., & Cunningham, T. F. (1992). A developmental evaluation of the role of word shape in word recognition. Memory and Cognition, 20, 141±150. Imbs, P. (1971). EÂtudes statistiques sur le vocabulaire francËais. Dictionnaire des freÂquencies. Vocabulaire litteÂraire des XIXeÁme et XXeÁme sieÁcles [Statistical analysis of French vocabulary: Frequency counts of literary works from the 19th and 20th centuries]. Nancy, France: Marcel Didier. Kinoshita, S. (1987). Case alternation effect: Two types of word recognition? Quarterly Journal of Experimental Psychology, 39A, 701±720. KucËera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Landauer, T. K., & Streeter, L. A. (1973). Structural differences between common and rare words: Failure of equivalence assumptions for theories of word recognition. Journal of Verbal Learning and Verbal Behavior, 12, 119±131. Marsolek, C. J. (1995). Abstract-visual-form representations in the left cerebral hemisphere. Journal of Experimental Psychology: Human Perception and Performance, 21, 375±386. Marsolek, C. J., Kosslyn, S. M., & Squire, L. R. (1992). Form-specific visual priming in the right cerebral hemisphere. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 492±508. Marsolek, C. J., Schacter, D. L., & Nicholas, C. (1996). Form-specific visual priming for new associations in the right cerebral hemisphere. Memory and Cognition, 24, 539±556. Massaro, D. W., Cohen, M. M. (1994). Visual, orthographic, phonological, and lexical influences in reading. Journal of Experimental Psychology: Human Perception and Performance, 20, 1107±1128. Mayall, K., & Humphreys, G. W. (1996). Case mixing and the task-sensitive disruption of lexical processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 278±294. Mayall, K., Humphreys, G. W., & Olson, A. (1997). Disruption to word or letter processing? The origins of case-mixing effects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 1275±1286. McCann, R. S., & Besner, D. (1987). Reading pseudo-homophones: Implications for models of pronunciation assembly and the locus of word-frequency effects in naming. Journal of Experimental Psychology: Human Perception and Performance, 13, 14±24. McCann, R. S., Besner, D., & Davelaar, E. (1988). Word recognition and identification: Does word frequency reflect lexical access? Journal of Experimental Psychology: Human Perception and Performance, 14, 693±706. McClelland, J. L. (1976). Preliminary letter recognition in the perception of words and non-words. Journal of Experimental Psychology: Human Perception and Performance, 2, 80±91. McClelland, J. L. (1977). Letter and configuration information in word identification. Journal of Verbal Learning and Verbal Behavior, 16, 137±150. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88, 375±407. McConkie, G. W., & Zola, D. (1979). Is visual information integrated across successive fixations in reading? Perception and Psychophysics, 25, 221±224. Monk, A. F., & Hulme, C. (1983). Errors in proofreading: Evidence for the use of word shape in word recognition. Memory and Cognition, 11, 16±23. Oden, G. C. (1984). Dependence, independence, and emergence of word features. Journal of Experimental Psychology: Human Perception and Performance, 10, 394±405. Paap, K. R., Newsome, S. L., McDonald, J. E., & Schvaneveldt, R. W. (1982). An activationverification model for letter and word recognition. The word-superiority effect. Psychological Review, 89, 573±594.

944

LEÂTEÂ AND PYNTE

Paap, K. R., Newsome, S. L., & Noel, R. W. (1984). Word shape's in poor shape for the race to the lexicon. Journal of Experimental Psychology: Human Perception and Performance, 10, 313±428. Perea, M., & Rosa, E. (2002). Does ``whole-word shape'' play a role in visual word recognition? Perception and Psychophysics, 64, 785±794. Plourde, C. E., & Besner, D. (1997). On the locus of the word frequency effect in visual word recognition. Canadian Journal of Experimental Psychology, 51, 181±194. Pynte, J. (2000). Words likely to activate many lexical candidates are granted an advantage: Evidence from within-word eye movements. In A. Kennedy, R. Radach, D. Heller, & J. Pynte (Eds.), Reading as a perceptual process (pp. 47±64). Oxford, UK: Elsevier. Rumelhart, D. E., & McClelland, J. L. (1982). An interactive activation model of context effects in letter perception: Part 2. The contextual enhancement effect and some tests and extensions of the model. Psychological Review, 89, 60±94. Sanocki, T. (1991). Intra- and inter-pattern relations in letter recognition. Journal of Experimental Psychology: Human Perception and Performance, 17, 924±941. Sears, C. R., Hino, Y., & Lupker, S. J. (1995). Neighborhood size and neighborhood frequency effects in word recognition. Journal of Experimental Psychology: Human Perception and Performance, 21, 876±900. Seidenberg, M. S. (1985). The time course of phonological code activation in two writing systems. Cognition, 19, 1±30. Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word recognition and naming. Psychological Review, 96, 523±568. Seidenberg, M. S., Waters, G. S., Barnes, M. A., & Tanenhaus, M. K. (1984). When does irregular spelling or pronunciation influence word recognition? Journal of Verbal Learning and Verbal Behavior, 23, 383±404. Smith, F. (1969). Familiarity of configuration versus discriminability of features in the visual identification of words. Psychonomic Science, 14, 261±262. Smith, F., Lott, D., & Cronnell, B. (1969). The effect of type size and case alternation on word identification. American Journal of Psychology, 82, 248±253. Stanners, R. F., Jastrzembski, J. E., & Westbrook, A. (1975). Frequency and visual quality in a word± nonword classification task. Journal of verbal Learning and Verbal Behavior, 14, 259±264. Taraban, R., & McClelland, J. L. (1987). Conspiracy effects in word pronunciation. Journal of Memory and Language, 26, 608±631. Underwood, G., & Bargh, K. (1982). Word shape, orthographic regularity, and contextual interactions in a reading task. Cognition, 12, 197±209. Walker, P. (1987). Word shape as a cue to the identity of a word: An analysis of the KucÃera and Francis (1967) word list. Quarterly Journal of Experimental Psychology, 39A, 675±700. Wheeler, D. (1970). Processes in word recognition. Cognitive Psychology, 1, 59±85. Whiteley, H. E., & Walker, P. (1994). The activation of multiletter units in visual word recognition. Visual Cognition, 1, 433±473. Whiteley, H. E., & Walker, P. (1997). Multiletter units in visual word recognition: Direct activation by supraletter visual features. Visual Cognition, 4, 69±110. Manuscript received March 2002 Revised manuscript received April 2003

WORD-SHAPE EFFECT

945

APPENDICES APPENDIX 1 WORDS USED IN EXPERIMENTS 1, 2, AND 3 Lexical frequency

Bigram frequency

Number of candidates

High-frequency shapeÐfrequent words reconnu AJ murmure NO rancune NO excuser VB revivre VB soucieux AJ sarcasme NO musicien NO examiner VB survivre VB convaincu AJ excursion NO raccourci NO massacrer VB conserver VB Mean

379 339 341 374 315 338 265 355 371 331 351 267 286 274 376 331

2.7 2.7 2.8 2.6 2.8 3.0 2.6 2.7 2.7 2.7 2.7 2.6 2.6 2.8 3.0 2.7

290 290 290 290 290 219 219 219 219 219 135 135 135 135 135 215

High-frequency shapeÐrare words veineux AJ urinoir NO verseau NO envaser VB emmurer VB omnivore AJ mocassin NO excision NO croasser VB escrimer VB circoncis AJ armurerie NO annonceur NO immuniser VB concasser VB Mean

174 180 60 123 146 146 0 90 132 162 132 123 0 180 123 118

2.8 2.7 2.7 2.8 2.8 2.6 2.7 2.7 2.6 2.7 2.8 2.7 2.8 2.7 3.0 2.7

290 290 290 290 290 219 219 219 219 219 135 135 135 135 135 215

Item

Class

(continued overleaf)

946

LEÂTEÂ AND PYNTE APPENDIX 1 (cont.) Lexical frequency

Bigram frequency

Number of candidates

Low-frequency shapeÐfrequent words steÂrile AJ preÂlude NO volupte NO digeÂrer VB frapper VB eÂloquent AJ controÃle NO mareÂchal NO peÂneÂtrer VB palpiter VB deÂfinitif AJ poleÂmique NO sinceÂrite NO appreÂcier VB conqueÂrir VB Mean

323 285 338 276 402 300 347 360 388 287 366 275 348 338 329 331

2.7 2.7 2.6 2.7 2.7 2.7 2.7 2.7 2.7 2.7 2.7 2.7 2.7 2.7 2.7 2.7

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Low-frequency shapeÐrare words eÂmeÂrite AJ pompage NO peÂnates NO reÂeÂlire VB eÂcreÂmer VB pigmente AJ eÂchalote NO filtrage NO eÂgrapper VB treÂpider VB domicilie AJ magnitude NO pointille NO peÂtarader VB torreÂfier VB Mean

123 90 153 90 123 123 140 158 0 123 166 90 166 108 123 118

2.5 2.8 2.6 2.4 2.7 2.6 2.7 2.7 2.5 2.6 2.6 2.8 3.0 2.7 2.8 2.7

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Item

Class

947

WORD-SHAPE EFFECT APPENDIX 2 Words used in Experiment 4 Bigram Frequency

Number of Candidates

A

M

D

N

High-frequency shapeÐfrequent words banque NO 340 pardon NO 369 quatre NO 444 barque NO 338 beÃtise NO 341 deÂtour NO 348 grandir VB 352 gronder VB 335 gueuler VB 287 joindre VB 356 pivoter VB 263 prendre VB 501 gendarme NO 348 peinture NO 387 poitrine NO 387 produire VB 404 prudence NO 349 question NO 453

2.80 2.84 3.23 2.73 2.63 2.91 3.01 3.01 2.90 3.07 2.87 3.19 2.79 2.99 3.24 2.80 2.79 3.09

51 53 53 51 27 27 43 43 43 43 43 43 41 38 41 41 41 38

1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1

0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 1 0 0 1 1 1 1 1 1 1 1 1 1 1 1

4 4 4 4 3 3 5 5 5 5 5 5 6 6 6 6 6 6

High-frequency shapeÐrare words bisque NO 0 friper VB 158 jucher VB 174 langer VB 0 heÂlium NO 140 torche AJ 132 goinfre NO 158 guinder VB 180 piastre NO 185 picoler VB 140 picotin NO 123 tripier NO 153 ganterie NO 60 goinfrer VB 108 parolier NO 0 poudrier NO 166 quatrain NO 190 questeur NO 0

2.86 2.68 2.77 3.11 1.71 2.57 2.72 2.83 2.84 2.78 2.63 2.79 2.94 2.75 3.08 3.07 3.07 3.09

51 51 53 51 27 18 43 43 43 43 43 41 41 38 38 41 41 38

1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1

0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 1 0 0 1 1 1 1 1 1 1 1 1 1 1 1

4 4 4 4 3 3 5 5 5 5 5 5 6 6 6 6 6 6

Item

Class

Lexical Frequency

(continued overleaf)

948

LEÂTEÂ AND PYNTE APPENDIX 2 (cont.) Bigram Frequency

Number of Candidates

A

M

D

N

Low-frequency shapeÐfrequent words agiter VB 369 change NO 319 charge NO 367 emploi NO 370 blesse NO 382 eÂtoile NO 396 analyse NO 383 aveugle NO 386 changer VB 424 charger VB 372 concept NO 345 exemple NO 440 accepter VB 420 analyser VB 326 beaucoup AV 474 champion NO 288 escargot NO 269 vulgaire NO 349

2.80 2.81 2.66 2.53 2.58 2.98 2.31 2.81 2.92 2.80 2.59 2.66 2.60 2.46 2.55 2.75 2.22 2.64

4 2 2 9 1 3 3 6 6 6 1 6 5 2 2 4 1 1

1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1

0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 1 0 0 1 1 1 1 1 1 1 1 1 1 1 1

4 4 4 4 3 3 5 5 5 5 5 5 6 6 6 6 6 6

Low-frequency shapeÐrare words aiglon NO 162 aphone AJ 177 calque NO 132 oxyder VB 108 caÃbler VB 140 troõÈka NO 123 abroger VB 140 aphasie NO 153 clivage NO 177 occiput NO 177 scalper VB 108 spolier VB 123 aguicher VB 158 ajusteur NO 183 congeler VB 140 inculper VB 140 opticien NO 180 rempiler VB 140

2.61 2.68 2.75 1.45 2.03 1.36 2.60 2.48 2.52 2.16 2.31 2.76 2.63 2.53 3.00 2.56 2.63 2.95

9 5 2 9 1 1 6 3 1 2 3 4 6 4 4 2 1 4

1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1

0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 1 0 0 1 1 1 1 1 1 1 1 1 1 1 1

4 4 4 4 3 3 5 5 5 5 5 5 6 6 6 6 6 6

Item

Class

Lexical Frequency

A: Number of ascending letters; M: Number of marked letters; D: Number of descending letters; N: Number of neutral letters.