- Email: [email protected]
Is banara really a word?
Cognition 113 (2009) 254–257
Contents lists available at ScienceDirect
Cognition journal homepage: www.elsevier.com/locate/COGNIT
Brief article
Is banara really a word? q Xiaomei Qiao 1, Kenneth Forster *, Naoko Witzel Department of Psychology, University of Arizona, Tucson, AZ 85721, United States
a r t i c l e
i n f o
Article history: Received 24 January 2009 Revised 18 July 2009 Accepted 5 August 2009
Keywords: Masked priming Lexical decision Lexical acquisition Competition Visual word recognition Lexical access
a b s t r a c t Bowers, Davis, and Hanley (Bowers, J. S., Davis, C. J., & Hanley, D. A. (2005). Interfering neighbours: The impact of novel word learning on the identification of visually similar words. Cognition, 97(3), B45–B54) reported that if participants were trained to type nonwords such as banara, subsequent semantic categorization responses to similar words such as banana were delayed. This was taken as direct experimental support for a process of lexical competition during word recognition. This interpretation assumes that banara has been lexicalized, which predicts that masked form priming for items such as banara– banana should be reduced or eliminated. An experiment is reported showing that the trained novel words produced the same amount of priming as untrained nonwords on both the first and the second day of training, suggesting that the interference observed by Bowers et al was not due to word-on-word competition. Ó 2009 Elsevier B.V. All rights reserved.
1. Introduction The role played by competition in visual word recognition is the defining feature of models based on the interactive activation model of McClelland and Rumelhart (1981). Without this competitive process, a cascaded parallel model (which has no identification thresholds) has no way to uniquely identify the stimulus, since the system has to decide which word unit is the most strongly activated, and competition is the only way to establish this. Hence any evidence either for or against such a process is keenly sought. The most obvious prediction from such a model is that it should take longer to identify a word that has many neighbors. However, the evidence here is very mixed. For English at least, it appears that having many neighbors leads to faster lexical decision times, although
q This paper is based on an experiment reported in Xiaomei Qiao’s dissertation (Qiao, 2009). * Corresponding author. E-mail addresses: [email protected] (X. Qiao), kforster@ u.arizona.edu (K. Forster), [email protected] (N. Witzel). 1 Present address: Shanghai University of Finance and Economics, Shanghai, China.
0010-0277/$ - see front matter Ó 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.cognition.2009.08.006
inhibitory effects have been observed in French and Spanish (see Andrews (1997) for a review). Bowers et al. (2005) pointed to the methodological problems that plague research in this area, namely the difficulty of finding two sets of words that differ on the number of neighbors but are matched on all other variables known to affect lexical processing. In an elegant experiment they avoided this problem by selecting words that had no neighbors (e.g., banana), and creating neighbors for these words by training participants to type similar nonwords (e.g., banara). In a subsequent semantic categorization experiment (where unique identification is required), this group took longer to categorize words like banana than a group that had not been required to learn to type banara. This result was taken as strong support for a competitive process, but there is one issue that needs to be settled before that conclusion can be accepted, namely the question of how banara is represented. Is it represented in the lexicon, or is it merely represented in episodic memory? A reasonable working assumption would be that competition would be limited to lexically represented words only, that is, only real words would compete with words, in which case a competitive effect would imply that banara had been lexicalized. But what might be happening instead
X. Qiao et al. / Cognition 113 (2009) 254–257
is that the training establishes a strong episodic trace for the nonword banara, so that when the target word banana is presented, it not only activates the lexical representation, but also the episodic trace of banara. This forces a more intensive post-access orthographic check in order to decide whether the input was banana or banara. There is a test that may help us decide whether the banara effect represents a case of genuine word-on-word competition. In a lexical decision task using a masked priming paradigm (e.g., Forster & Davis, 1984), priming between two similar forms (form priming) depends on whether the prime is a word or a nonword. When the prime is a nonword (e.g., contrapt–CONTRACT), there is a strong facilitatory effect, but when it is a word (e.g., contrast–CONTRACT), there is sometimes no priming, or even an inhibitory effect (Davis & Lupker, 2006; Forster & Veres, 1998; Nakayama, Sears, & Lupker, 2008; Segui & Grainger, 1990). This phenomenon is known as a prime lexicality effect (henceforth PLE). Elgort (2007) used this phenomenon to investigate the effectiveness of vocabulary acquisition in a second language. If a newly learned word in the new language has become lexicalized, then it should behave like a word in a masked priming experiment. Thus, if banara is represented lexically, then it should fail to prime its lexical neighbor BANANA. The following experiment was designed to test this hypothesis. As in the Bowers et al. (2005) study, participants were trained on two successive days to type a set of nonwords that were one-letter-different from a real word. After training, a lexical decision task was used with trained and untrained nonwords as masked primes, rather than a semantic categorization task, as in the Bowers study. If the trained nonwords function as real words, then they should produce less priming than the untrained nonword primes. The reason for using lexical decision rather than semantic categorization is simply that it is unknown whether prime lexicality is relevant to masked priming in a semantic categorization task. 2. Method 2.1. Participants A total of 39 undergraduate students at the University of Arizona participated and received course credit. 2.2. Materials and design The 40 critical to-be-learned items and their base words (e.g., banara and its base word BANANA) used in the original Bowers et al. (2005) study were used as the experimental stimuli. For list counterbalancing purposes, eight more items were added, which were six letters in length and had no substitution, transition, deletion or addition neighbors, and a CELEX written frequency count between 1 and 17 per million. For the typing task, 24 of these nonwords were used for half the participants, and the other 24 were used for the remaining participants. For the lexical decision task, each base word was paired with a related nonword prime (banara–BANANA), or a completely unrelated non-
255
word prime that was also one-letter-different from a word (agenty–BANANA). Half of the related nonword primes had been trained, half had not. An additional set of 48 words that have a real word neighbor were included as a test of whether the conditions of the experiment were conducive to a real prime lexicality effect. These word targets were primed either by a related word prime (passive–MASSIVE), or an unrelated word prime (logical–MASSIVE) in two lists of the four, and were primes by a related NONWORD prime (cassive–MASSIVE), or an unrelated word prime (logical– MASSIVE) in the other two lists. These items were omitted on Day 2, along with a corresponding number of nonword targets. Nonword distractors for the lexical decision task consisted of 96 orthographically legal nonwords that were one-letter-different from a word, primed with either a word neighbor (marvel–MARVIL), or an unrelated nonword prime (e.g., ganger–MARVIL). Four counterbalanced lists were constructed so that each base word appeared in each condition across lists, but only once for each participant, and so that each nonword was used as a trained and as an untrained prime. Trained items were only used as related primes, unrelated primes were untrained for all participants. 2.3. Procedure The training procedure on each day was exactly the same as in the Bowers study. Altogether there were ten blocks of typing. At the end of each block the full list of the typed items appeared on the screen for participants to read through. After the typing task was completed, participants performed the lexical decision task. A typical masked priming paradigm was used consisting of a forward mask (########) presented for 500 ms, followed by the prime which was presented in lower case letters for 50 ms, which was then replaced by the upper-case target which was displayed for 500 ms. After each trial the participants were given feedback about speed and accuracy of their response. All items were presented in a different pseudorandom order for each participant. After the participant responded and the feedback was displayed, the next trial was initiated automatically. Prior to the experiment beginning, 12 practice items were used. The experiment was controlled by a Pentium PC, using the Windows DMDX software developed by J.C. Forster at the University of Arizona (Forster & Forster, 2003). Participants were assigned to lists in order of appearance at the laboratory. 3. Results Three participants did not appear for the Day 2 session, and their performance on Day 1 was excluded from the analysis. Five participants were rejected because their overall error rates exceeded 30% on at least one of the two days. In order to get an equal number of participants on each of the four lists, the data from an additional three participants were discarded, which yielded a total of 28 participants. Reaction times (RT) that were above
256
X. Qiao et al. / Cognition 113 (2009) 254–257
Table 1 Mean lexical decision times and error rates (in parentheses) for word targets that have trained nonword neighbors on the first and second day of training. Training on related primes
Related prime (banara–BANANA)
Unrelated prime (agenty–BANANA)
Priming
Day 1
Trained Untrained
588 (9.7) 612 (15.2)
602 (14.7) 626 (15.6)
14 14
Day 2
Trained Untrained
589 (8.7) 585 (10.4)
611 (12.7) 618 (15.6)
22 33
1500 ms or below 300 ms were also excluded (5.9% of the trials). Four items (one in each list) were removed from the analysis as they had an error rate exceeding 50%, yielding a total of 44 items. The excluded items were JACKAL, TARMAC, MOSAIC, and SORDID. Table 1 shows the mean lexical decision times and error rates for the word targets as a function of whether the prime was related or not, and whether it was trained or not. These means were analyzed in a 2 4 2 2 ANOVA, the factors being Days (Day 1 vs. Day 2), Groups (subject groups in the subject analysis, item groups in the item analysis), Training (trained vs. untrained prime), and Priming (related vs. unrelated prime). The overall effect of Priming (21 ms) was significant, F1(1,24) = 13.67, p < .001, F2(1,40) = 13.20, p < .001, but the interaction of Training and Priming was not (both Fs < 1), indicating that there was no difference between the priming obtained with trained and untrained primes. Nor was the three-way interaction of Day, Training, and Priming (both Fs < 1). The only other effect of interest was the main effect of Training, where training on a nonword such as banara produced faster overall RTs to a word target such as BANANA than to a word target for which there was no corresponding trained nonword, F1(1,24) = 4.89, p < .05, F2(1,40) = 10.85, p < .01. In the analysis of error rates, there was a significant effect of Priming, with fewer errors in the related condition, F1(1,24) = 9.77, p < .01, F2(1,40) = 4.81, p < .05. Also, trained primes tended to produce fewer errors, this effect being significant in the subject analysis, F1(1,24) = 4.20, p = .05, but not quite in the item analysis, F2(1,40) = 3.82, p = .06. No other effects of interest were significant. Table 2 shows the mean lexical decision times and error rates for the word targets with real word primes (passive– MASSIVE). As can be seen, there was no priming ( 1 ms) in this condition. This result establishes an important prerequisite for the success of the experiment, confirming that the conditions of this experiment were favorable for a prime lexicality effect. The effect with a nonword prime (25 ms) was significant in the subject analysis, F1(1, 24) = 7.12, p = .01, but somewhat weaker in the item analysis, F2(1,44) = 3.45, p = .07. Although the interaction between priming and lexical status of the prime was not significant, F1(1,24) = 3.56, p = .07, F2(1,88) = 1.18, p > .05,
it would not make sense to therefore conclude that the same amount of priming is observed in the two conditions, or that there is no priming in either condition, especially since there are multiple instances of priming with a nonword prime in the present experiment, and in the literature. A more likely explanation is that there is considerable variance when the prime is a word. Finally, there was no significant priming for the nonword targets on either day. For Day 1, the priming effect was 5 ms, and on Day 2, it was 2 ms (all Fs < 1). 4. Discussion The results show that the training technique used by Bowers et al. (2005) did not reduce the capacity of nonwords to prime their real word neighbors in a lexical decision task. The only hint of a reduction in priming was on Day 2, when the priming with trained primes was 11 ms less than with untrained primes. However, this difference was not at all reliable, and the priming was in fact stronger than the priming observed on Day 1, despite an extra session of training. The implication is that the trained nonwords did not compete with their real word neighbors, as they should have if they had been lexicalized. An important element in this argument is the demonstration that there was no priming for real word pairs such as passive–MASSIVE. Without this, it could have been argued that the design of the lexical decision task used in this experiment was not capable of producing a prime lexicality effect, since it is known that such an effect does not occur unless the nonword distractors are one-letter-different from words (Forster & Veres, 1998). Inspection of the results in Table 1 shows that in the trained condition on Day 1, where the targets resembled the trained nonwords, the lexical decision times are slightly faster than in the untrained condition. This is quite the opposite of what was found in the Bowers study, where training on banara interfered with the categorization response to BANANA. One possibility is that there was a bias to respond ‘‘Yes” to any word that closely resembled one of the trained nonwords, since none of the nonword targets had this feature. However, this bias would be irrelevant to priming, because it would be present for both the related and unrelated conditions. Also, it should be noted
Table 2 Mean lexical decision times and error rates (in parentheses) on Day 1 for word targets that have real word neighbor(s).
Word targets
Real word primes Nonword primes
Related prime
Unrelated prime
Priming
(passive–MASSIVE) 577 (5.4) (cassive–MASSIVE) 571 (6.5)
(logical–MASSIVE) 576 (7.0) (logical–MASSIVE) 596 (6.7)
1 25
X. Qiao et al. / Cognition 113 (2009) 254–257
that this main effect is absent on Day 2, and it is not clear why such a bias would be abandoned with further training. In any event, the evidence on Day 2 (when there was no bias) indicates that the training did not eliminate priming. This raises the question of why interference should be obtained in a semantic categorization task, as in the Bowers study. One possible reason is that in such a task, participants normally assume that all target items will be correctly spelled words, and hence the importance of a post-access spelling check is downgraded. But when many of the targets are so similar in form to a previously trained nonword (and presumably participants infer that there must be some connection between the training and test phases of the experiment), greater weight may be given to a spelling check, leading to longer RTs for those items. One might think that exactly the same argument applies to lexical decision, but if one accepts that under the conditions of this experiment, where all nonword targets were one-letter-different from words, a spelling check is always required, regardless of the similarity to a previously trained nonword. What alternative issues remain to be explored? One is whether some other form of training might be sufficient to produce an effect of prime lexicality. It could be that the typing task establishes a lexical representation at the level of form only, and that more is required to produce a PLE. In a similar study, Elgort (2007) taught her participants meanings for the new words, and obtained a clear PLE. However, this experiment used visible primes and a long prime-target interval (522 ms), which makes it difficult to rule out possible effects of anticipation and other strategic influences. If a masked PLE can be obtained by training participants to associate a meaning with the novel words, the implication would be that competition occurs at a semantic level rather than at a form level. Another possibility is that a PLE might be obtained if a semantic categorization task was used (as in the Bowers study) rather than lexical decision. However, the failure to obtain a PLE in a semantic task would not be informative, since there have been no reports of a PLE using such a task, and the reason may be that a PLE only occurs when a detailed post-access spelling check is required. Finally, we need to clarify what we mean by saying that an item has become lexicalized. Put simply, it would mean that the item is represented in a neural structure that is specialized for language. At a purely functional level, an item can be said to have become lexicalized when it has the same properties as a word. One of those properties is revealed in the PLE, and it is argued here that a failure to exhibit a PLE indicates that banara has not been lexicalized. Another property is the presence of strong masked repetition priming (nonwords do not show such an effect), and hence we could have tested to see whether repetition
257
priming could be obtained for items such as banara–BANARA. (Indeed, Qiao (2009) has actually obtained such an effect in a lexical decision task for trained pseudo-neighbors such as BALTERY, even though participants were instructed to respond ‘‘No” to these items.) However, the problem here is that it has been shown that strong repetition priming is obtained for nonwords when tested in a speeded old/ new recognition memory task (Forster, 1985), indicating that episodic memory traces can be primed just like words. So this test is not diagnostic of lexicality. The conclusion, then is that the longer categorization time observed by Bowers et al. (2005) can be explained by the post-access spelling check triggered by the episodic memory traces of the novel words. In addition, this argument also explains why a prime lexicality effect was not observed in the current experiment. Real words do have competitive effects on their word neighbors, but items represented solely in episodic memory cannot compete with a word represented in lexical memory. Therefore, we maintain that the banara effect should not be taken as unambiguous support for a competitive process. References Andrews, S. (1997). The effect of orthographic similarity on lexical retrieval: Resolving neighborhood conflicts. Psychonomic Bulletin & Review, 4(4), 439–461. Bowers, J. S., Davis, C. J., & Hanley, D. A. (2005). Interfering neighbours: The impact of novel word learning on the identification of visually similar words. Cognition, 97(3), B45–B54. Davis, C. J., & Lupker, S. J. (2006). Masked inhibitory priming in English: Evidence for lexical inhibition. Journal of Experimental Psychology: Human Perception and Performance, 32(3), 668–687. Elgort, I. (2007). The role of intentional decontextualised learning in second language vocabulary acquisition: Evidence from primed lexical decision tasks with advanced bilinguals. Unpublished thesis. Victoria University of Wellington, New Zealand. Forster, K. I. (1985). Lexical acquisition and the modular lexicon. Language and Cognitive Processes, 1(2), 87–108. Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10(4), 680–698. Forster, K. I., & Forster, J. C. (2003). DMDX: A Windows display program with millisecond accuracy. Behavior Research Methods, Instruments & Computers, 35, 116–124. Forster, K. I., & Veres, C. (1998). The prime lexicality effect: Form-priming as a function of prime awareness, lexical status, and discrimination difficulty. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24(2), 298–314. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: I. An account of basic findings.. Psychological Review, 88(5), 375–407. Nakayama, M., Sears, C. R., & Lupker, S. J. (2008). Masked priming with orthographic neighbors: A test of the lexical competition assumption. Journal of Experimental Psychology: Human Perception and Performance, 34(5), 1236–1260. Qiao, X. (2009). The representation of newly-learned words in the mental lexicon. Unpublished dissertation. University of Arizona. Segui, J., & Grainger, J. (1990). Priming word recognition with orthographic neighbors: Effects of relative prime-target frequency. Journal of Experimental Psychology: Human Perception & Performance, 16(1), 65–76.
Contents lists available at ScienceDirect
Cognition journal homepage: www.elsevier.com/locate/COGNIT
Brief article
Is banara really a word? q Xiaomei Qiao 1, Kenneth Forster *, Naoko Witzel Department of Psychology, University of Arizona, Tucson, AZ 85721, United States
a r t i c l e
i n f o
Article history: Received 24 January 2009 Revised 18 July 2009 Accepted 5 August 2009
Keywords: Masked priming Lexical decision Lexical acquisition Competition Visual word recognition Lexical access
a b s t r a c t Bowers, Davis, and Hanley (Bowers, J. S., Davis, C. J., & Hanley, D. A. (2005). Interfering neighbours: The impact of novel word learning on the identification of visually similar words. Cognition, 97(3), B45–B54) reported that if participants were trained to type nonwords such as banara, subsequent semantic categorization responses to similar words such as banana were delayed. This was taken as direct experimental support for a process of lexical competition during word recognition. This interpretation assumes that banara has been lexicalized, which predicts that masked form priming for items such as banara– banana should be reduced or eliminated. An experiment is reported showing that the trained novel words produced the same amount of priming as untrained nonwords on both the first and the second day of training, suggesting that the interference observed by Bowers et al was not due to word-on-word competition. Ó 2009 Elsevier B.V. All rights reserved.
1. Introduction The role played by competition in visual word recognition is the defining feature of models based on the interactive activation model of McClelland and Rumelhart (1981). Without this competitive process, a cascaded parallel model (which has no identification thresholds) has no way to uniquely identify the stimulus, since the system has to decide which word unit is the most strongly activated, and competition is the only way to establish this. Hence any evidence either for or against such a process is keenly sought. The most obvious prediction from such a model is that it should take longer to identify a word that has many neighbors. However, the evidence here is very mixed. For English at least, it appears that having many neighbors leads to faster lexical decision times, although
q This paper is based on an experiment reported in Xiaomei Qiao’s dissertation (Qiao, 2009). * Corresponding author. E-mail addresses: [email protected] (X. Qiao), kforster@ u.arizona.edu (K. Forster), [email protected] (N. Witzel). 1 Present address: Shanghai University of Finance and Economics, Shanghai, China.
0010-0277/$ - see front matter Ó 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.cognition.2009.08.006
inhibitory effects have been observed in French and Spanish (see Andrews (1997) for a review). Bowers et al. (2005) pointed to the methodological problems that plague research in this area, namely the difficulty of finding two sets of words that differ on the number of neighbors but are matched on all other variables known to affect lexical processing. In an elegant experiment they avoided this problem by selecting words that had no neighbors (e.g., banana), and creating neighbors for these words by training participants to type similar nonwords (e.g., banara). In a subsequent semantic categorization experiment (where unique identification is required), this group took longer to categorize words like banana than a group that had not been required to learn to type banara. This result was taken as strong support for a competitive process, but there is one issue that needs to be settled before that conclusion can be accepted, namely the question of how banara is represented. Is it represented in the lexicon, or is it merely represented in episodic memory? A reasonable working assumption would be that competition would be limited to lexically represented words only, that is, only real words would compete with words, in which case a competitive effect would imply that banara had been lexicalized. But what might be happening instead
X. Qiao et al. / Cognition 113 (2009) 254–257
is that the training establishes a strong episodic trace for the nonword banara, so that when the target word banana is presented, it not only activates the lexical representation, but also the episodic trace of banara. This forces a more intensive post-access orthographic check in order to decide whether the input was banana or banara. There is a test that may help us decide whether the banara effect represents a case of genuine word-on-word competition. In a lexical decision task using a masked priming paradigm (e.g., Forster & Davis, 1984), priming between two similar forms (form priming) depends on whether the prime is a word or a nonword. When the prime is a nonword (e.g., contrapt–CONTRACT), there is a strong facilitatory effect, but when it is a word (e.g., contrast–CONTRACT), there is sometimes no priming, or even an inhibitory effect (Davis & Lupker, 2006; Forster & Veres, 1998; Nakayama, Sears, & Lupker, 2008; Segui & Grainger, 1990). This phenomenon is known as a prime lexicality effect (henceforth PLE). Elgort (2007) used this phenomenon to investigate the effectiveness of vocabulary acquisition in a second language. If a newly learned word in the new language has become lexicalized, then it should behave like a word in a masked priming experiment. Thus, if banara is represented lexically, then it should fail to prime its lexical neighbor BANANA. The following experiment was designed to test this hypothesis. As in the Bowers et al. (2005) study, participants were trained on two successive days to type a set of nonwords that were one-letter-different from a real word. After training, a lexical decision task was used with trained and untrained nonwords as masked primes, rather than a semantic categorization task, as in the Bowers study. If the trained nonwords function as real words, then they should produce less priming than the untrained nonword primes. The reason for using lexical decision rather than semantic categorization is simply that it is unknown whether prime lexicality is relevant to masked priming in a semantic categorization task. 2. Method 2.1. Participants A total of 39 undergraduate students at the University of Arizona participated and received course credit. 2.2. Materials and design The 40 critical to-be-learned items and their base words (e.g., banara and its base word BANANA) used in the original Bowers et al. (2005) study were used as the experimental stimuli. For list counterbalancing purposes, eight more items were added, which were six letters in length and had no substitution, transition, deletion or addition neighbors, and a CELEX written frequency count between 1 and 17 per million. For the typing task, 24 of these nonwords were used for half the participants, and the other 24 were used for the remaining participants. For the lexical decision task, each base word was paired with a related nonword prime (banara–BANANA), or a completely unrelated non-
255
word prime that was also one-letter-different from a word (agenty–BANANA). Half of the related nonword primes had been trained, half had not. An additional set of 48 words that have a real word neighbor were included as a test of whether the conditions of the experiment were conducive to a real prime lexicality effect. These word targets were primed either by a related word prime (passive–MASSIVE), or an unrelated word prime (logical–MASSIVE) in two lists of the four, and were primes by a related NONWORD prime (cassive–MASSIVE), or an unrelated word prime (logical– MASSIVE) in the other two lists. These items were omitted on Day 2, along with a corresponding number of nonword targets. Nonword distractors for the lexical decision task consisted of 96 orthographically legal nonwords that were one-letter-different from a word, primed with either a word neighbor (marvel–MARVIL), or an unrelated nonword prime (e.g., ganger–MARVIL). Four counterbalanced lists were constructed so that each base word appeared in each condition across lists, but only once for each participant, and so that each nonword was used as a trained and as an untrained prime. Trained items were only used as related primes, unrelated primes were untrained for all participants. 2.3. Procedure The training procedure on each day was exactly the same as in the Bowers study. Altogether there were ten blocks of typing. At the end of each block the full list of the typed items appeared on the screen for participants to read through. After the typing task was completed, participants performed the lexical decision task. A typical masked priming paradigm was used consisting of a forward mask (########) presented for 500 ms, followed by the prime which was presented in lower case letters for 50 ms, which was then replaced by the upper-case target which was displayed for 500 ms. After each trial the participants were given feedback about speed and accuracy of their response. All items were presented in a different pseudorandom order for each participant. After the participant responded and the feedback was displayed, the next trial was initiated automatically. Prior to the experiment beginning, 12 practice items were used. The experiment was controlled by a Pentium PC, using the Windows DMDX software developed by J.C. Forster at the University of Arizona (Forster & Forster, 2003). Participants were assigned to lists in order of appearance at the laboratory. 3. Results Three participants did not appear for the Day 2 session, and their performance on Day 1 was excluded from the analysis. Five participants were rejected because their overall error rates exceeded 30% on at least one of the two days. In order to get an equal number of participants on each of the four lists, the data from an additional three participants were discarded, which yielded a total of 28 participants. Reaction times (RT) that were above
256
X. Qiao et al. / Cognition 113 (2009) 254–257
Table 1 Mean lexical decision times and error rates (in parentheses) for word targets that have trained nonword neighbors on the first and second day of training. Training on related primes
Related prime (banara–BANANA)
Unrelated prime (agenty–BANANA)
Priming
Day 1
Trained Untrained
588 (9.7) 612 (15.2)
602 (14.7) 626 (15.6)
14 14
Day 2
Trained Untrained
589 (8.7) 585 (10.4)
611 (12.7) 618 (15.6)
22 33
1500 ms or below 300 ms were also excluded (5.9% of the trials). Four items (one in each list) were removed from the analysis as they had an error rate exceeding 50%, yielding a total of 44 items. The excluded items were JACKAL, TARMAC, MOSAIC, and SORDID. Table 1 shows the mean lexical decision times and error rates for the word targets as a function of whether the prime was related or not, and whether it was trained or not. These means were analyzed in a 2 4 2 2 ANOVA, the factors being Days (Day 1 vs. Day 2), Groups (subject groups in the subject analysis, item groups in the item analysis), Training (trained vs. untrained prime), and Priming (related vs. unrelated prime). The overall effect of Priming (21 ms) was significant, F1(1,24) = 13.67, p < .001, F2(1,40) = 13.20, p < .001, but the interaction of Training and Priming was not (both Fs < 1), indicating that there was no difference between the priming obtained with trained and untrained primes. Nor was the three-way interaction of Day, Training, and Priming (both Fs < 1). The only other effect of interest was the main effect of Training, where training on a nonword such as banara produced faster overall RTs to a word target such as BANANA than to a word target for which there was no corresponding trained nonword, F1(1,24) = 4.89, p < .05, F2(1,40) = 10.85, p < .01. In the analysis of error rates, there was a significant effect of Priming, with fewer errors in the related condition, F1(1,24) = 9.77, p < .01, F2(1,40) = 4.81, p < .05. Also, trained primes tended to produce fewer errors, this effect being significant in the subject analysis, F1(1,24) = 4.20, p = .05, but not quite in the item analysis, F2(1,40) = 3.82, p = .06. No other effects of interest were significant. Table 2 shows the mean lexical decision times and error rates for the word targets with real word primes (passive– MASSIVE). As can be seen, there was no priming ( 1 ms) in this condition. This result establishes an important prerequisite for the success of the experiment, confirming that the conditions of this experiment were favorable for a prime lexicality effect. The effect with a nonword prime (25 ms) was significant in the subject analysis, F1(1, 24) = 7.12, p = .01, but somewhat weaker in the item analysis, F2(1,44) = 3.45, p = .07. Although the interaction between priming and lexical status of the prime was not significant, F1(1,24) = 3.56, p = .07, F2(1,88) = 1.18, p > .05,
it would not make sense to therefore conclude that the same amount of priming is observed in the two conditions, or that there is no priming in either condition, especially since there are multiple instances of priming with a nonword prime in the present experiment, and in the literature. A more likely explanation is that there is considerable variance when the prime is a word. Finally, there was no significant priming for the nonword targets on either day. For Day 1, the priming effect was 5 ms, and on Day 2, it was 2 ms (all Fs < 1). 4. Discussion The results show that the training technique used by Bowers et al. (2005) did not reduce the capacity of nonwords to prime their real word neighbors in a lexical decision task. The only hint of a reduction in priming was on Day 2, when the priming with trained primes was 11 ms less than with untrained primes. However, this difference was not at all reliable, and the priming was in fact stronger than the priming observed on Day 1, despite an extra session of training. The implication is that the trained nonwords did not compete with their real word neighbors, as they should have if they had been lexicalized. An important element in this argument is the demonstration that there was no priming for real word pairs such as passive–MASSIVE. Without this, it could have been argued that the design of the lexical decision task used in this experiment was not capable of producing a prime lexicality effect, since it is known that such an effect does not occur unless the nonword distractors are one-letter-different from words (Forster & Veres, 1998). Inspection of the results in Table 1 shows that in the trained condition on Day 1, where the targets resembled the trained nonwords, the lexical decision times are slightly faster than in the untrained condition. This is quite the opposite of what was found in the Bowers study, where training on banara interfered with the categorization response to BANANA. One possibility is that there was a bias to respond ‘‘Yes” to any word that closely resembled one of the trained nonwords, since none of the nonword targets had this feature. However, this bias would be irrelevant to priming, because it would be present for both the related and unrelated conditions. Also, it should be noted
Table 2 Mean lexical decision times and error rates (in parentheses) on Day 1 for word targets that have real word neighbor(s).
Word targets
Real word primes Nonword primes
Related prime
Unrelated prime
Priming
(passive–MASSIVE) 577 (5.4) (cassive–MASSIVE) 571 (6.5)
(logical–MASSIVE) 576 (7.0) (logical–MASSIVE) 596 (6.7)
1 25
X. Qiao et al. / Cognition 113 (2009) 254–257
that this main effect is absent on Day 2, and it is not clear why such a bias would be abandoned with further training. In any event, the evidence on Day 2 (when there was no bias) indicates that the training did not eliminate priming. This raises the question of why interference should be obtained in a semantic categorization task, as in the Bowers study. One possible reason is that in such a task, participants normally assume that all target items will be correctly spelled words, and hence the importance of a post-access spelling check is downgraded. But when many of the targets are so similar in form to a previously trained nonword (and presumably participants infer that there must be some connection between the training and test phases of the experiment), greater weight may be given to a spelling check, leading to longer RTs for those items. One might think that exactly the same argument applies to lexical decision, but if one accepts that under the conditions of this experiment, where all nonword targets were one-letter-different from words, a spelling check is always required, regardless of the similarity to a previously trained nonword. What alternative issues remain to be explored? One is whether some other form of training might be sufficient to produce an effect of prime lexicality. It could be that the typing task establishes a lexical representation at the level of form only, and that more is required to produce a PLE. In a similar study, Elgort (2007) taught her participants meanings for the new words, and obtained a clear PLE. However, this experiment used visible primes and a long prime-target interval (522 ms), which makes it difficult to rule out possible effects of anticipation and other strategic influences. If a masked PLE can be obtained by training participants to associate a meaning with the novel words, the implication would be that competition occurs at a semantic level rather than at a form level. Another possibility is that a PLE might be obtained if a semantic categorization task was used (as in the Bowers study) rather than lexical decision. However, the failure to obtain a PLE in a semantic task would not be informative, since there have been no reports of a PLE using such a task, and the reason may be that a PLE only occurs when a detailed post-access spelling check is required. Finally, we need to clarify what we mean by saying that an item has become lexicalized. Put simply, it would mean that the item is represented in a neural structure that is specialized for language. At a purely functional level, an item can be said to have become lexicalized when it has the same properties as a word. One of those properties is revealed in the PLE, and it is argued here that a failure to exhibit a PLE indicates that banara has not been lexicalized. Another property is the presence of strong masked repetition priming (nonwords do not show such an effect), and hence we could have tested to see whether repetition
257
priming could be obtained for items such as banara–BANARA. (Indeed, Qiao (2009) has actually obtained such an effect in a lexical decision task for trained pseudo-neighbors such as BALTERY, even though participants were instructed to respond ‘‘No” to these items.) However, the problem here is that it has been shown that strong repetition priming is obtained for nonwords when tested in a speeded old/ new recognition memory task (Forster, 1985), indicating that episodic memory traces can be primed just like words. So this test is not diagnostic of lexicality. The conclusion, then is that the longer categorization time observed by Bowers et al. (2005) can be explained by the post-access spelling check triggered by the episodic memory traces of the novel words. In addition, this argument also explains why a prime lexicality effect was not observed in the current experiment. Real words do have competitive effects on their word neighbors, but items represented solely in episodic memory cannot compete with a word represented in lexical memory. Therefore, we maintain that the banara effect should not be taken as unambiguous support for a competitive process. References Andrews, S. (1997). The effect of orthographic similarity on lexical retrieval: Resolving neighborhood conflicts. Psychonomic Bulletin & Review, 4(4), 439–461. Bowers, J. S., Davis, C. J., & Hanley, D. A. (2005). Interfering neighbours: The impact of novel word learning on the identification of visually similar words. Cognition, 97(3), B45–B54. Davis, C. J., & Lupker, S. J. (2006). Masked inhibitory priming in English: Evidence for lexical inhibition. Journal of Experimental Psychology: Human Perception and Performance, 32(3), 668–687. Elgort, I. (2007). The role of intentional decontextualised learning in second language vocabulary acquisition: Evidence from primed lexical decision tasks with advanced bilinguals. Unpublished thesis. Victoria University of Wellington, New Zealand. Forster, K. I. (1985). Lexical acquisition and the modular lexicon. Language and Cognitive Processes, 1(2), 87–108. Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10(4), 680–698. Forster, K. I., & Forster, J. C. (2003). DMDX: A Windows display program with millisecond accuracy. Behavior Research Methods, Instruments & Computers, 35, 116–124. Forster, K. I., & Veres, C. (1998). The prime lexicality effect: Form-priming as a function of prime awareness, lexical status, and discrimination difficulty. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24(2), 298–314. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: I. An account of basic findings.. Psychological Review, 88(5), 375–407. Nakayama, M., Sears, C. R., & Lupker, S. J. (2008). Masked priming with orthographic neighbors: A test of the lexical competition assumption. Journal of Experimental Psychology: Human Perception and Performance, 34(5), 1236–1260. Qiao, X. (2009). The representation of newly-learned words in the mental lexicon. Unpublished dissertation. University of Arizona. Segui, J., & Grainger, J. (1990). Priming word recognition with orthographic neighbors: Effects of relative prime-target frequency. Journal of Experimental Psychology: Human Perception & Performance, 16(1), 65–76.