You are on page 1of 21

Journal of Memory and Language 48 (2003) 468488

Journal of Memory and Language


www.elsevier.com/locate/jml

Semantic facilitation and semantic interference in word translation: Implications for models of lexical access in language production
Ineke Bloem* and Wido La Heij
Faculty of Social Sciences, Unit of Cognitive Psychology, Leiden University, P.O. Box 9555, 2300 RB Leiden, The Netherlands Received 31 October 2001; revision received 5 August 2002

Abstract We rst show that in a word-translation task, context words induce semantic interference whereas context pictures induce semantic facilitation. Experiments 2 and 3 show that this nding is not due to dierences between context words and context pictures in terms of (a) relative speed of lexical activation or (b) the category level of the activated concepts. To account for our ndings, we propose that conceptually-driven lexical access is conned to the selected target concept (or preverbal message). A version of Starreveld and La Heijs (1996) connectionist model in which this proposal was implemented successfully simulated the polarity and the time course of the semantic context eects observed. In Experiment 4 the prediction that context pictures do not induce lexical context eects was tested and conrmed. 2002 Elsevier Science (USA). All rights reserved.
Keywords: Language production; Stroop; Semantic priming; Semantic interference; Selective attention

One of the empirical enigmas in experimental psychology is that the presence of a semantic relation between a target stimulus and a context has a positive eect on performance in some paradigms, but a clear negative eect in others. This discrepancy, that Neumann (1986) termed the semantic relatedness paradox, is most evident when one compares the results obtained with traditional semantic priming tasks with those obtained with variants of the Stroop task. For example, in comparison with an unrelated word context, the reading of the word CHAIR is facilitated when it is preceded by the word TABLE (e.g., Warren, 1977), but the naming of the color red is hampered by the simultaneous presentation of the word BLUE (Klein, 1964; see MacLeod, 1991, for an overview). However, as will be discussed shortly, even within Stroop-like paradigms in which target and context are presented in close tem* Corresponding author. Fax: +31-715273783. E-mail address: bloem@fsw.leidenuniv.nl (I. Bloem).

poral proximity, semantic interference as well as semantic facilitation can be obtained. The semantic relatedness paradox is mentioned at several places in the literature (see, e.g., Levelt, Roelofs, & Meyer, 1999; Underwood, 1986; Vitkovitch & Humphreys, 1991; Vitkovitch & Tyrrell, 1999) and a number of attempts were made to systematically investigate its cause (see, e.g., La Heij, 1988; La Heij, Dirkx, & Kramer, 1990b; La Heij, Van der Heijden, & Schreuder, 1985). A conclusion that can be derived from these and other studies is that semantic interference will always be obtained when two conditions are fullled. First, the context is a word presented in close temporal proximity to the target. Second, the task is a speechproduction task. That is, the response word has to be retrieved based on conceptual information, as is the case in color naming (Klein, 1964), picture naming (Rosinski, 1977; Underwood, 1976), denition naming (La Heij, Starreveld, & Steehouwer, 1993b) and word translation (La Heij et al., 1990a).

0749-596X/02/$ - see front matter 2002 Elsevier Science (USA). All rights reserved. doi:10.1016/S0749-596X(02)00503-X

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

469

Fig. 1. The relevant part of Roelofs (1992) and Starreveld and La Heijs (1996) models of lexical access. Ct conceptual representation of the target, Cc conceptual representation of the context stimulus, Lt lexical representation of the targets name, Lc lexical representation of the name of the context stimulus. The link between Ct and Cc has a zero weight when target and context are semantically unrelated.

To the best of our knowledge, Glaser and Glaser (1989) were the rst to publish a processing model in which this semantic interference eect was accounted for in terms of spreading activation between related concepts in semantic memory (Collins & Loftus, 1975). A neural-network version of Glaser and Glasers account, proposed by Roelofs (1992) and Starreveld and La Heij (1996), is presented in Fig. 1. In the network in Fig. 1, Ct and Cc represent the conceptual representations of the target and the context stimulus, respectively. Lt and Lc represent the lexical representations of the names of the target and the context, respectively.1 In the models of Roelofs (1992) and Starreveld and La Heij (1996), the presentation of a pictureword stimulus is simulated by activating the nodes Ct and Lc . The instruction to name the target picture is implemented as an extra activation input to Ct (task input in Fig. 1). If target and context are semantically related, spreading activation between Ct and Cc results in an increase in activation of both nodes, which in turn results in an increase of activation of both lexical representations (note that the models assume that during lexical access all activated concepts activate their lexical representations). The crucial point is that there is an asymmetry in the network: the targets conceptual representation gets more activation than the conceptual representation of the context word. The reasons are that (a) Ct is activated directly, whereas Cc is activated via its lexical representation Lc (as a consequence, per time unit

only a fraction of Lc s activation will spread to Cc ) and (b) Ct receives additional task activation. Because Ct is stronger activated than Cc , relatively more activation will spread from Ct to Cc than vice versa. Consequently, spreading activation between related concepts results in a relatively stronger increase in activation of the Lc node than of the Lt node. Because Lt and Lc compete for selection (selection takes place when the activation of Lt exceeds the activation of Lc by a critical amount), spreading activation ultimately results in a larger delay in the selection of the correct lexical representation than when target and context are unrelated: the semantic interference eect. Within research on language production, this account of semantic interference is widely accepted by now (see e.g., Humphreys, Lloyd-Jones, & Fias, 1995; Levelt, Roelofs, & Meyer, 1999; Starreveld & La Heij, 1995, 1996). Moreover, computer simulations (Roelofs, 1992; Starreveld & La Heij, 1996) have conrmed that under the conditions that prevail in pictureword interference tasks, spreading activation indeed results in semantic interference. The mechanism of spreading activation is also widely used to account for semantic facilitation eects in, for instance, priming studies. It is therefore somewhat surprising that little attempt has been made to investigate whether the model depicted in Fig. 1 is capable of simulating semantic facilitation in speech-production tasks.2 One reason might be that it is not completely clear which factors are responsible for a reversal from semantic interference into semantic facilitation. Factors that may play a role are the inter stimulus interval (ISI), SOA, type of relation between target and context (categorical or associative; see La Heij et al., 1990b) and the instruction with respect to the processing of the context stimulus. An accidental observation that we made in two of our studies on word translation may be very helpful in a further understanding of the processes that underlie semantic interference and semantic facilitation in speech production. In a rst study, La Heij et al. (1990a) used a word-translation variant of the Stroop task in which an English target word (e.g., SPOON) had to be translated into Dutch (lepel). In the relevant experimental conditions the target word was accompanied by a semantically related context word (e.g., VORK, the Dutch translation equivalent of the word fork) or by an
2 One notable exception is Roelofss (1992) attempt to account for semantic facilitation by assuming that context words that are not part of the response set do not compete for selection at the lexical level. However, this account was severely challenged by ndings that these context words do induce semantic interference (Caramazza & Costa, 2000, 2001; Starreveld & La Heij, 1999; see Levelt et al., 1999; Roelofs, 2001, for a reply).

1 The issue whether these representations are abstract lemmas, as in Roelofss (1992) model, or contain phonological information, as in Starreveld and La Heijs (1996) model, is irrelevant for the present issue.

470

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

unrelated Dutch context word. Completely in accordance with the results of the pictureword interference paradigm, this experiment showed a semantic interference eect. This result was taken as support for the hypothesis that semantic interference is obtained in all tasks in which a response word has to be selected based on a conceptual representation. In a second study, La Heij, Hooglander, Kerling, and Van der Velden (1996) addressed the question whether translation from a foreign language into the rst language (backward translation) is conceptually mediated. In these experiments, a context picture accompanied the English target word (e.g., BEACH); the picture was either semantically related (e.g., a parasol) or unrelated (e.g., a house) with the target. The results showed a clear eect of semantic relatedness, which was taken as evidence that backward translation is indeed achieved via the target words conceptual representation. For our present purposes, the interesting observation was that in these experiments a semantically related context picture facilitated word translation in comparison with an unrelated picture. Probably because the studies of La Heij et al. (1990a) and La Heij et al. (1996) examined dierent issues, the striking dierence in polarity of the semantic context eect has not received any attention in the literature. However, if it could be shown that in word-translation tasks a mere change in the modality of the context stimulus (a word or a picture) results in a complete reversal from semantic interference into semantic facilitation, we have an excellent empirical basis for a further development of language production models (e.g., Levelt et al., 1999; Roelofs, 1992; Starreveld & La Heij, 1996). To elucidate this point, let us examine the network in Fig. 1 again. If we assume that backward translation is conceptually mediated (we return to that assumption in the General discussion), the presentation of a target word in a foreign language ultimately leads to the activation of the target concept Ct . Semantic interference by context words as observed by La Heij et al. (1990a) can be explained in the usual way: activation spreads from Ct to Cc and from there to Lc . Lc also gets activation from the context word and becomes a strong competitor in the selection of the correct response Lt . When instead of a context word a context picture is presented, as in the study of La Heij et al. (1996), the situation is somewhat dierent, but the model still predicts semantic interference. Due to the presentation of the context picture, Cc receives a lot of activation. If target and context are semantically related, part of this activation spreads to the target node Ct . This spread of activation is often assumed to underlie semantic facilitation. However, the activation of Ct (including task activation) spreads back to Cc and from there to Lc . Ultimately, Lc will prot more than Lt from the spread of activation at the conceptual level and this will hamper the selection of Lt , just

as in the situation in which a context word was presented. Indeed, Roelofs (1992) reported that in his model of lexical access context pictures, although to a lesser degree than context words, induce semantic interference.3 Simulations performed with the computer implementation of the Starreveld and La Heij (1996) model, conrmed our theoretical analysis of the models behavior. In these simulations a number of parameters were systematically varied up to the point where the model was unable to select a response (no critical difference at the lexical level could be reached; see Appendix D for details). Manipulation of the following parameters seemed most important. First, to simulate the fact that the L2 target words used by La Heij et al. (1996) probably activate their conceptual representations less strongly than target pictures in a pictureword interference task, the input activation Ain was systematically reduced. Second, the amount of spreading activation (parameter wcc ) was varied to investigate whether a semantically related context picture induces facilitation when enough of its activation spreads to the target concept. Third, the amount of task activation (At ) added to the target concept was varied. Completely in line with our theoretical analysis of the behavior of the model, none of these manipulations (alone or in combination) made the model produce a semantic facilitation eect. Before discussing possible modications of language production models, it is necessary to determine whether the remarkable dierence in direction of the semantic context eects reported by La Heij et al. (1990a) and La Heij et al. (1996) is indeed due to the modality of the context stimuli. In fact, the stimulus materials and the display conditions used in the two studies diered in many respects, which may have contributed to or perhaps even caused the strong dissimilarity in results. To examine this issue, in Experiment 1 a to-be-translated English target word was accompanied by a word or a picture that was either semantically related or unrelated to the target. This experiment indeed showed a full reversal from semantic interference with word context into semantic facilitation with picture context. In Experiment 2 SOA was manipulated to test a horse-race account (or relative speed hypothesis) of the semantic facilitation eect obtained with picture context. This account was rejected based on the nding that context pictures still induced semantic facilitation
3 Because Glaser and Glaser (1989) obtained semantic interference in a picture-naming task with picture context, Roelofs (1992) considered this result as a successful simulation of an empirical nding. However, Damian and Bowers (in press) and La Heij, Heikoop, Akerboom, and Bloem (submitted) recently showed that Glaser and Glasers nding is irregular and probably due to selection problems in their sequential discrimination task (see the General discussion).

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

471

when pre-exposed by 250 ms. Experiment 3 examined whether the semantic facilitation eect induced by context pictures could be due to the (implicit) naming of the pictures at a superordinate-category level. In the Discussion of Experiment 3, we propose a modication of current language production models to account for our ndings. Finally, Experiments 4a and 4b tested and conrmed a prediction derived from the revised model.

Experiment 1 Method Participants. Sixteen Leiden University students served as paid participants. They were native Dutch speakers and highly procient in the English language (all received more than ve years English education in high school). They all had normal or corrected-to-normal vision. Apparatus. The Experiment was programmed using MEL Professional software (version 2.0d; Schneider, 1988). Presentation of the stimuli and collection of the data were performed using a fast IBM compatible PC. Two monitors were connected to the PC. The participants were seated in front of an Iiyama 17-in. monitor. The experimenter was shown the correct response and the subjects response latency on a VISA FM 1200 black/white monitor. Response latency was measured by means of a voice-key with an accuracy of 1 ms. Stimuli. Thirty two high frequency English words were selected (Appendix A) that were familiar to Dutch university students and did not show a clear phonological or orthographic relation with their Dutch translation equivalent (that is, no cognates were used). For each of the target words a related concept was selected that could be presented both as a word and as a picture context. For the unrelated condition, the context concepts were re-paired with the targets, such that the context and the target were not semantically related. For example, the target word TROUSERS (to be translated into the Dutch word broek) was accompanied by the picture of a coat or by the word JAS (the Dutch translation equivalent of coat); in the unrelated condition TROUSERS was accompanied either by the picture of a cow or by the word KOE (the Dutch translation equivalent of a cow). The Dutch translation equivalents of the English target words and the Dutch context words were of similar mean language frequency (log frequencies of 1.87 and 1.73, respectively; CELEX database, Burnage, 1990). The pictures were selected from the line drawings provided by Snodgrass and Vanderwart (1980). These pictures had a high name agreement, which indicates that (for the English participants in that study) they were easy to identify.

The to-be-translated English target word was always presented in black lower-case letters against a white background. The height and width of the target and context words were 1:4 2 of visual angle for 3-letter words up to 1:4 5 of visual angle for 8-letter words. The target words were positioned in such a way that the second letter appeared at the point of xation. In the picture-context condition the target word and context picture were superimposed. The maximum size of the pictures was 7 7 degree of visual angle. To assure a high legibility of the target word the black letters that made up the word were surrounded by a white border (approximately 2 mm) and the context picture was presented in gray. The pictures were positioned in such a way that the target word did not cover their essential features. In the word-context condition, the context word was presented in red letters, 1.4 degree of visual angle (center-to-center) below the target word. Viewing distance was approximately 80 cm. Procedure. The participants were tested individually in a dimly illuminated room. In a written instruction, they were asked to translate the English target words as fast as possible while maintaining accuracy and to ignore the context (word or picture). Next, each participant received a series of 32 trials in which the English target words were presented in isolation. If the participant did not know the translation, produced an incorrect translation or hesitated, the trial was repeated at the end of the series. Finally, two blocks of 64 trials were presented, corresponding to the two context-type conditions. Half of the participants started with picture context, the other half with the word context. Each of these experimental series was preceded by a practice series in which each of the 32 English target words was presented with a related or unrelated context stimulus. Half of the target-context stimuli were semantically related, the other half unrelated. The context stimuli used in these practice series were of the same modality, but taken from a dierent set than the context stimuli in the experimental series. Each trial involved the following sequence. First, a xation point appeared for 500 ms in the center of the display. Next, the stimulus (target and context stimulus) appeared and remained on the display until response. If no response was registered after 2000 ms, the stimulus was removed from the display and the next trial was started. The experimenter judged the responses and entered a code into the computer to indicate whether the response was correct or false. Voice key malfunction could also be registered. Results. Response latencies (RTs) of incorrect responses were excluded from the analyses. In addition, RTs of trials in which the voice-key malfunctioned and RTs under 300 ms (most probably reecting voice key malfunctioning) were excluded (2.1% of the data). The remaining RTs were used in the calculation of the means

472

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

Table 1 Mean RTs (in ms) and percentages of errors in the various experimental conditions of Experiment 1 Related RT Word context Picture context 793 769 %Error 2.0 1.8 RT 765 797 Unrelated %Error 1.6 2.2 Relatedness eect (RT unrelated ) RT related) )28 +28

per participant and per item for each of the four experimental conditions. The participant means per condition are shown in Table 1. Analyses of variance (ANOVAs) were performed on the participant (F1 ) and item means (F2 ) with relatedness (related versus unrelated target-context pairs) and context modality (picture versus word) as within-participant factors. These analyses showed a signicant interaction between the factors context modality and semantic relatedness, F1 1; 15 4:18; p < :005; MSe 868:93, and F2 1; 31 16:13; p < :001; MSe 1382:99, indicating that the semantic relatedness eect was dierent for word and picture contexts. With a word context, a semantic interference eect of 28 ms was obtained and with a picture context, a 28 ms semantic facilitation effect was obtained. To further investigate this interaction, paired-samples t tests were performed on both the participant means (t1 ) and the item means (t2 ). The semantic interference eect induced by a word context were signicant, t1 15 2:58; p < :025; t2 31 2:38; p < :025. The same was true for the semantic facilitation eect induced by a picture context, t1 15 3:55; p < :005; t2 31 3:33; p < :005. Inspection of Table 1 reveals that the error percentages paralleled the latency data. They were considered too low to allow a meaningful analysis. Discussion. The results of this Experiment completely conrm the ndings reported by La Heij et al. (1990a) and La Heij et al. (1996). In accordance with the results of the former study, backward translation was hampered by the presence of a semantically related context word in comparison to an unrelated word. In accordance with the La Heij et al. (1996) study, backward translation was facilitated by the presence of a semantically related context picture in comparison to an unrelated picture. Apparently, the strong dierence in the direction of the semantic context eects in our previous studies is completely due to a dierence in context modality. To our knowledge, the present experiment is the rst to isolate one single variable that is capable of inducing a full reversal from semantic interference into semantic facilitation. One other observation is worth discussing. The word and picture context conditions did not dier in mean response latency (779 and 783 ms, respectively), suggesting that context words and pictures induce a similar overall eect on word translation latencies. However,

this nding should be interpreted with caution: in our experiments, the context words were presented below the central xation point whereas the context pictures were centered on the point of xation. Therefore, a somewhat reduced legibility of the context words and a possible larger visual masking eect induced by the context pictures may have contributed to this nding.

Experiment 2 A factor that might have contributed to the semantic facilitation eect observed with picture contexts in Experiment 1, is that pictures take more time to activate their lexical representations than words do (Cattell, 1886; Fraisse, 1969). Consequently, context pictures may induce facilitation at the conceptual level before their names can interfere at the lexical level. The traditional way to investigate this relative speed hypothesis or horse race account is to give the nonverbal context stimulus a head start (see, e.g., Glaser & Dngelho, u 1984; Glaser & Glaser, 1982, for a similar approach in the Stroop task and the pictureword interference task, respectively). That is, the nonverbal context is presented before the target word to compensate for a possible dierence in the speed with which the corresponding lexical representations become activated. In the present experiment, this objective was combined with the need to obtain information about the time course of both semantic interference and semantic facilitation in the word-translation task. In Experiment 2 SOA values of )250, 0, and +150 ms were used. The SOA of )250 seems large enough to compensate for a dierence in lexical access by pictures and words (Glaser & Dngelho, 1984) and small enu ough to prevent anticipation strategies (Neely, 1977). In addition, a neutral condition was added that might allow for a further interpretation of the semantic eects observed. In the word-context condition, the neutral stimulus consisted of a series of ve Xs. In the picturecontext condition the neutral stimulus consisted of six concentric rectangles. Method Participants. Thirty-six Leiden University students served as paid participants. They were native Dutch

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

473

speakers and highly procient in the English language. Eighteen participants were randomly assigned to the word-context condition, the others to the picture-context condition. Stimuli. The stimuli were identical to the ones in Experiment 1, with the only dierence that a neutral condition was added. In the word-context condition, the neutral stimuli consisted of a series of 5 Xs. In the picture-context condition the neutral stimulus was a symmetrical geometric gure consisting of 6 concentric rectangles with largest dimensions of 5.4 (width) 3.4 (height) of visual angle. Procedure. The procedure was very similar to the one in Experiment 1. The participants rst examined a list containing the 32 English target words and their Dutch translations. Next, they received a practice series in which they were required to translate the 32 English target words as fast as possible. Finally, they received three experimental series corresponding to the three SOA conditions. The order of presentation of these conditions was completely balanced across participants. Each series consisted of 96 trials each (32 English target words 3 experimental conditions) and was preceded by eight practice trials that were randomly chosen from the experimental materials (Experiment 1 learned that eight trials suced to get used to the presence of context words). Results. The data were treated in the same way as in Experiment 1. RTs of incorrect responses and RTs of trials in which the voice-key malfunctioned or under 300 ms were excluded from the analyses (4.67% of the data). Table 2 shows the participant means in the various experimental conditions. ANOVAs were performed with target-context relation (semantically related, unrelated and neutral) and SOA as within-participant factors and with context modality (words versus pictures) as between-participants or within-items factor. These analyses showed a signicant main eect of context modality, F1 1; 34 14:33; p < :001; MSe 85,195.27, and F2 1; 31 430:38; p < :001; MSe 4912:12, reecting that mean response

latencies were smaller with a word (730 ms) than with a picture context (853 ms). In addition, the main eect of the factor target-context relation was signicant, F1 2; 68 20:54; p < :001; MSe 1443:39, and F2 2; 62 33:40; p < :001; MSe 1734:86. The interaction between context modality and target-context relation was signicant, F1 2; 68 9:02; p < :001; MSe 1443:39, and F2 2; 62 15:12; p < :001; MSe 2113:35. The interaction between SOA and context modality did not reach signicance in the participants analysis, but did reach signicance in the item analysis, F1 2; 68 0:69; p :504; MSe 8459:90; F2 2; 62 6:12; p < :01; MSe 1477:39. In addition, the threeterm interaction between semantic relatedness, context modality and SOA reached signicance, F1 4; 136 3:05; p < :05; MSe 919:53, and F2 4; 124 3:35; p < :05; MSe 1469:41. Averaged across SOAs, context words induced a semantic interference eect of 10 ms and context pictures a semantic facilitation eect of 34 ms. Having found that the eect of a semantic relation between target and context diered for a word and picture context, we performed paired-samples t tests on the data of the word context condition and of the picture context condition per SOA. For the word context condition, t tests showed a signicant semantic interference eect at SOA 0 ms, t1 17 2:31, p < :05, t2 (31) 3.05, p < :005 and SOA +150 ms, t1 17 2:37; p < :05; t2 31 2:41; p < :05. For the picture context condition, t tests showed a signicant semantic facilitation eect at SOA )250 ms, t1 17 3:22; p < :005; t2 31 2:89; p < :01, at SOA 0 ms, t1 17 2:51; p < :05; t2 31 3:64; p < :001 and at SOA 150 ms; t1 17 3:07; p < :01; t2 31 2:94; p < :01. Inspection of Table 2 reveals that the error percentages in the various conditions mirrored the latency data. These percentages were considered too small to allow a meaningful analysis. Discussion. The main nding in this experiment is a signicant interaction between context modality and

Table 2 Mean RTs (in ms) and error percentages in the various experimental conditions of Experiment 2, the relatedness eects and the simulated relatedness eects SOA RT Word context )250 719 0 751 150 761 Picture context )250 847 0 849 150 836 Related %Error 1.6 2.3 2.8 1.2 1.6 2.8 RT 726 732 744 885 888 862 Unrelated %Error 0.9 1.6 1.9 0.5 1.6 2.6 RT 712 714 709 832 836 837 Neutral %Error 1.4 1.2 1.7 1.6 1.0 1.7 Relatedness eect Simulated relatedness eect

+7 )19 )17 +38 +39 +26

0 )20 )20 +40 +20 +20

474

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

semantic relatedness, again showing that the eect of a semantic relation between target and context diers for word and picture contexts. At SOA 0, words induced a semantic interference eect of 19 ms (28 ms in Experiment 1) and pictures induced a semantic facilitation effect of 39 ms (28 ms in Experiment 1). One reason for manipulating SOA in the present experiment was to examine whether a pre-exposure of a context picture would provide time for the pictures name to become activated, which should result in semantic interference. The results of the SOA )250 ms condition clearly proved this conjecture incorrect. At this SOA, instead of semantic interference, context pictures induced a semantic facilitation eect of 38 ms. So, we can safely conclude that the semantic facilitation eect observed with picture context is not due to a delayed activation of the corresponding names at a lexical level. The fact that context pictures induced semantic facilitation even when presented 150 ms after target onset supports the view that pictures have a very fast access to their conceptual representations (see, e.g., Glaser & Glaser, 1989). Whereas pictures induced semantic facilitation in the full SOA range examined, the semantic interference effect induced by context words appears to be conned to the SOA values of 0 and +150 ms. At SOA )250 ms, semantically related and unrelated context words induced very similar mean response latencies (719 and 726 ms, respectively). This nding is in line with the lack of a semantic interference eect at an SOA of )400 ms in the word-translation study by La Heij et al. (1990a) and in line with the time courses in pictureword interference studies (Glaser & Dngelho, 1984; Starreveld & La u Heij, 1996). We return to this lack of semantic interference at SOA )250 ms in the General discussion. The neutral context condition (a series of Xs in the word-context condition and a set of concentric rectangles in the picture-context condition) was intended to serve as a baseline condition to get an indication of real facilitation and interference eects. The results show that for context words the mean response latencies obtained in the neutral condition were always smaller than those obtained in the semantically related and unrelated conditions. This nding is often attributed to interference of context words at a phonological level. For context pictures, the mean response latencies obtained in the neutral condition were not signicantly dierent from those obtained in the related condition. At rst sight, one might interpret the results of this condition as evidence that related context pictures do not induce real semantic facilitation. However, as noted by Jonides and Mack (1984), it could also be argued that our neutral condition was not neutral in all respects. For instance, it may be that the neutral picture chosen in our experiment (a set of concentric rectangles) induced less visual masking than the related and unrelated con-

text pictures. However, it is equally possible that other factors contributed to this nding, like the visual complexity and the meaningfulness of the related and unrelated pictures in comparison with the concentric rectangles. In addition, the large number of repetitions of the neutral context picture may have played a role. Therefore, unfortunately, the results of the neutral condition do not provide much insight in the nature of the context eects in terms of real facilitation or inhibition. The mean response latencies obtained were larger in the picture-context condition than in the word-context condition. In the Discussion of Experiment 1, we mentioned a number of factors that may contribute to a dierence between these conditions, like retinal locus and visual masking. Given the fact that the two conditions did not dier in Experiment 1, in which a within-participants design was used, we do not venture a further interpretation of this dierence in the present experiment. As discussed above, current models of language production cannot account for the semantic facilitation eect observed with context pictures. The reason is that these models assume that context pictures automatically activate their names at the lexical level. For example, when the English target word PIGEON is presented in combination with the context picture of a swan, the name of the picture (swan) will hamper the retrieval of the Dutch response word duif (pigeon). Our data clearly showed that this prediction is incorrect: the picture of a swan facilitates the translation of the word PIGEON into the Dutch word duif. Before concluding that language production models have to be modied to accommodate this nding, we discuss two alternative accounts of semantic facilitation eects with picture context. First, it could be argued that the words that were activated by our context pictures may have diered from the context words that we used in the word-context condition. For example, the picture of a swan may not have activated the correct basic-level concept (swan), but a dierent basic-level concept (e.g., goose) instead. The response to such an objection is simple: when PIGEON has to be translated, activation of the concept goose should induce as much interference as the activation of the intended concept swan. In addition, as noted above, the context pictures that we used had a high name agreement (Snodgrass & Vanderwart, 1980), which makes it unlikely that many of these misidentications occurred. A second and more serious concern is the following.4 It could be argued that some, or perhaps all of our

4 We thank Dr. R. Treiman for suggesting this possible account of our ndings.

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

475

context pictures (e.g., of a swan) did not activate their basic-level concepts, but the superordinate level concept instead (e.g., bird). Recently, Vitkovitch and Tyrrell (1999) reported that the naming of pictures at a subordinate level (e.g., naming a picture of a poodle as poodle instead of dog) is facilitated by the corresponding basic-category word (e.g., dog) in comparison to an unrelated word. Given this nding, it is conceivable that the naming of a picture of a pigeon is facilitated when the context picture of a swan activates the superordinate name bird. In Experiment 3, we tested this superordinate-category hypothesis.

Experiment 3 As noted above, Vitkovitch and Tyrrell (1999) showed that naming pictures at a subordinate-category level (e.g., poodle) is facilitated by correct basic-level names (e.g., dog) in comparison to unrelated basiclevel names (e.g., pear). Given this nding, it is conceivable that names at a superordinate level (e.g. animal) facilitate basic-level naming (e.g., dog). If so, the semantic facilitation eect observed in Experiments 1 and 2 may have been due to participants retrieving (in part of the trials) the superordinate-category name instead of the basic-level name of the context picture. In this experiment, we test whether superordinate-category names indeed induce semantic facilitation in basic-level word production. To that end, we used context words at the basiccategory level and at the superordinate-category level. For example, the English target word (e.g., LETTUCE) was accompanied by the basic-level words ERWT (pea; semantically related) and BEEN (leg; semantically unrelated) and with the superordinate-category context words GROENTE (vegetable; correct category label) and DIER (animal; incorrect category label). Method Participants. Eighteen Leiden University students served as paid participants. They were native Dutch speakers and highly procient in the English language. Stimuli. Thirty English target words were selected from ten dierent semantic categories (vehicles, vegetables, furniture, fruits, clothes, colors, animals, occupa-

tions, body parts and buildings). In the semantically related, basic-level condition, these target words were accompanied by basic-level names from the same semantic category. In the related category-level condition, the target words were accompanied by their category labels. In the corresponding unrelated conditions, targets and context words were re-paired to form unrelated pairs (see Appendix B, for a full list of stimulus materials). Procedure. The procedure was very similar to the one in Experiment 2. The participants were rst shown the list of 30 English target words and their translations. Next, they received a practice series in which the 30 target words were presented in isolation. The participants were instructed to translate these words as fast as possible while maintaining accuracy. Finally, they received an experimental series, consisting of 120 trials each (30 English target words 4 experimental conditions), preceded by eight practice trials that were randomly chosen from the experimental materials. Results. The data were treated in the same way as in Experiment 2. RTs of incorrect responses and RTs of trials in which the voice-key malfunctioned or RTs under 300 ms were excluded from the analyses (4.64% of the data). Table 3 shows the participant means in the various experimental conditions. ANOVAs were performed with relatedness (semantically related and unrelated target-context pairs) and level of context word (basic and super ordinate category level) as within-participant factors. This analysis showed a signicant main eect of relatedness, F1 1; 17 8:62; p < :01; MSe 2229:147, and F2 1; 29 10:20; p < :01; MSe 3330:485. The factor categorization level of the context word approached signicance in the participant analysis and was signicant in the items analysis, F1 1; 17 4:34; p < :055; MSe 3010:732, and F2 1; 29 7:033; p < :025; MSe 2676:506. The interaction was far from signicant. The error percentages mirrored the latency data and were too low to allow a meaningful analysis. Discussion. The results of this experiment clearly showed that categorization level of the context words had no eect on the direction or size of the semantic context eect. That is, context words at a superordinate-category level induced semantic interference, just as basic-level context words. In addition, the sizes of the two eects were similar: basic-category and superordinate-category context words induced semantic

Table 3 Mean RTs (in ms) and error percentages in the various experimental conditions of Experiment 3 Related RT Basic-level distractors Category-level distractors 898 874 %Error 5.6 5.0 RT 868 838 Unrelated %Error 4.1 3.3 Semantic interference RT 30 36

476

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

interference eects of 30 and 36 ms, respectively. Given this nding, we can safely conclude that naming (part of) the context pictures in Experiments 1 and 2 at the superordinate-category level cannot be responsible for the observed semantic facilitation eect with picture context. How then to account for the observation that context pictures induce semantic facilitation in word translation? In our view, all empirical evidence points in one direction: context pictures activate their conceptual representations, but do not automatically activate their names. In this way, semantically related context pictures may facilitate the identication of the target word at a conceptual level (e.g., due to spreading activation), without hampering the selection of the response word at the lexical level. It should be noted that the idea that nonverbal stimuli do not automatically activate their names is not new. A similar conclusion was reached in research on Stroop interference and selective attention. We will briey discuss a number of relevant experimental ndings. In research on Stroop color-word interference, much attention has been given to the fact that reading a color word is not delayed when it is presented in an incongruent color (see MacLeod, 1991). The rst explanation given for this lack of a reversed Stroop eect was the horse-race account, or relative speed hypothesis in which it is assumed that colors are slower in activating their names than word are (Cattell, 1886; Fraisse, 1969). As discussed above, this account was tested by introducing a time interval between the presentation of the color and the presentation of the target word (Glaser & Glaser, 1982; Neumann, 1980). The conclusion of these studies was clear. Pre-exposure of the color did not induce a reversed Stroop eect. Referring to the horse race account, Neumann aptly summarized his ndings as follows: Das zweite Pferd startet. . . gar nicht zum Wettlauf (the second horse. . . . does not even start in the race; p. 60). That is, incongruent colors do not interfere with word reading because they are slower in activating their names, but because they do not activate their names at all. Glaser and Glaser (1982) reached the same conclusion and provided additional evidence for the controlled character of color naming: when in the negative SOA conditions the context color correctly predicted the target word in 80% of the trials (so it became advantageous for the participants to retrieve the colors name), a reversed Stroop eect did emerge. A possible counterargument against the conclusion of Neumann (1980) and Glaser and Glaser (1982) is that color words may be read via a sublexical (graphemephoneme conversion) route and for that reason are not prone to interference by the name of the context color at a lexical level (see, e.g., Roelofs, 1992). However, this argument cannot be used for ndings obtained with a

colorcolor variant of the Stroop task (in which color naming is hampered by an incongruent color patch) reported by La Heij, Helaha, and Van Den Hof (1993a) and La Heij et al. (1995; see also La Heij, Van der Heijden, & Plooij, 2001). These authors argued that colorcolor interference eects reported in the literature may have been due to the incorrect selection of the context color for naming; an error that would lead to a strong activation of the name of the context color. This activation may lead either to an incorrect response or to interference in retrieving the correct color name, just as in the orthodox color-word Stroop task. In accordance with this account, La Heij et al. 1993a, 19931993b and La Heij, Kaptein, Kal, and de Lange (1995) showed that colorcolor interference could be eliminated by facilitating the selection of the target color. This facilitation was achieved by increasing the discriminability of target and context in terms of position (presentation of the target at a xed position), form (using dierent shapes for target and context) and exposure duration (using dierent exposure durations for target and context stimulus). Based on their ndings, these authors concluded that nonverbal context stimuli (e.g., colors) are processed up to the level of identication but do not activate their names, provided that they are not erroneously selected for naming (see Driver & Tipper, 1989, for a similar conclusion with respect to context pictures in the negative priming paradigm). Finally, our proposal that nonverbal stimuli do not automatically activate their names at the lexical level is in line with the results and interpretations of the postcue naming task reported by Humphreys et al. (1995). In this task, two pictures were presented and the participants were asked to retrieve both names. A post-cue indicated which picture had to be named aloud. The authors reported that a semantic relation between the two pictures resulted in interference, a nding that is in accordance with the idea that the two names compete for selection at the lexical level. The interesting nding was that when one of the pictures was pre-cued, no semantic interference eect was obtained. This nding suggests that in that situation, as in our experiments, the name of the context picture was not activated. The reason why Humphreys et al. (1995) did not observe semantic facilitation in their pre-cue condition (as we did in our Experiments 1 and 2) is most likely that target identication (concept activation) is faster for pictures than for L2 words. Consequently, a semantically related context will induce only a small facilitating eect on picture recognition. We return to this issue in the General discussion. We conclude that these ndings, in combination with the results obtained in Experiments 13, provide substantial empirical evidence in favor of the view that nonverbal context stimuli activate their conceptual representations, but do not automatically activate their

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

477

names. The studies discussed above suggest that the names of these context stimuli only become activated when the participants erroneously select the context picture for naming or when they are instructed to retrieve the name of the context picture, as in the Humphreys et al.s (1995) post-cue naming task. Another way of putting this conclusion is to say that name retrieval is a controlled process that only uses the target concept to retrieve the appropriate word in the mental lexicon; a conclusion that seems in line with Levelts (1989) original blueprint of the speaker in which lexicalization is conned to the content of the preverbal message. In the following section, we present a possible implementation of this idea in Starreveld and La Heijs (1996) model of lexical access.

Models of lexical access As discussed above, current models of language production are able to account for semantic interference induced by a context word, but not for semantic facilitation induced by a context picture. Because alternative accounts of this semantic facilitation eect could be rejected based on the results of Experiments 13, a modication of the current models seems required. The proposed modication concerns the assumption in language production models that during lexical access all activated concepts activate their lexical representations automatically and in parallel. The consequence of this assumption is that spreading activation at the conceptual level inevitably leads to a relatively strong increase in activation of the name of a semantically related context stimulus. To account for the semantic facilitation eect observed with context pictures, we propose that activated concepts do not automatically activate their names. Instead, lexical access in language production is viewed as a controlled process that is only applied to the targets conceptual representation. As discussed above, this assumption is not new; various researchers have advanced the same idea in order to account for, for instance, the lack of a reversed Stroop eect. Before presenting the modications made to the implemented version of the Starreveld and La Heij (1996) model, one issue has to be discussed. As mentioned in the introduction, we assume that word translation resembles picture naming in the fact that it is conceptually mediated (we return to that assumption in the General discussion). However, it is clear that L2 words and pictures dier in the processing stages that lead to conceptual activation. The most important dierence is that an L2 target word will strongly activate its representations at the lexical level before activating the corresponding concept. Therefore, in theory, in our experiments the lexical representation of the L2 target

word (e.g., HORSE) could compete for selection with the correct response word (e.g., PAARD; see Green, 1998). For the following two reasons, we decided not to model this rst stage of lexical activation in the word translation task. First, within the literature on bilingualism it is generally assumed that in word translation, bilinguals are able to ignore or suppress words in the nonresponse language. For instance, Costa, Miozzo, and Caramazza (1999) proposed that the lexical-selection mechanism considers words from the response language only. If that were true, activation of the target word would have no eect on response-retrieval times. Another possibility is that the response selection mechanism is not language selective, but that the L2 target word has suciently decayed by the time the correct L1 word has to be selected. The idea that the L2 target word does not strongly interfere with the selection of the correct response word is supported by our experimental ndings. If the lexical representation of the L2 target word would be strongly activated at the time of response selection (due to its visual presentation and its high conceptual similarity with the correct response word), its massive interference at the lexical level should render eects induced by the context words, like semantic interference and phonological facilitation, invisible. The very fact that the translation of the English word HORSE into the Dutch word PAARD takes longer when it is presented with the context word KOE (cow) than when presented with the context word SOK (sock) indicates that the lexical activation of the L2 target word HORSE does not play a substantial role in lexical selection. Given these considerations, we decided to simulate the presentation of an L2 target word in the same way as Starreveld and La Heij (1996) simulated the presentation of a target picture: by adding activation to the corresponding conceptual representation. How to modify the Starreveld and La Heij (1996) model to implement our assumption that only the target concept is lexicalized? A very simple way is to assume a threshold at the conceptual level.5 Only the concept that is selected for naming, that is, the concept that receives task activation, exceeds threshold (is selected) and activates its lexical representation. With this assumption, the model easily accounts for semantic facilitation with picture context. Spreading activation at the conceptual level between the two activated, semantically related concepts cause the target concept (in our experiments

5 It should be noted that a threshold is only one way to implement this proposal in a network model. Another possibility is a disinhibition (elevation of the inhibition) of the link between concept and lexical representations (see, e.g., Van der Velde & De Kamps, 2001).

478

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

Fig. 2. The modication of Starreveld and La Heijs (1996) model of lexical access, used in the simulations. A conceptual representation only activates lexical representations when its activation exceeds a threshold value.

the meaning of the L2 word) to reach the threshold earlier, resulting in semantic facilitation. The spread of activation from the target concept to other, related, concepts has no consequence for activation levels at the lexical level, as long as the activation levels of these semantically related conceptual representations remain below the assumed threshold level.6 However, to account for semantic interference an additional modication has to be made: once above threshold, the target concept does not only activate its own name but also - to a small degree - semantically related words. This assumption is in line with earlier conceptualizations of lexical access, in which the activation level of a word was thought to be proportional to the amount of overlap of its semantic characteristics and the semantic characteristics of the concept to-be-expressed (e.g., Levelt, 1989; Morton, 1969). The resulting modications are illustrated in Fig. 2. Simulations were performed with the relevant part of the Starreveld and La Heij (1996) model, using parameters very similar to those of Starreveld and La Heij (1996; see Appendix D). The main purpose of the simulations was to check whether the revised model was able to simulate the change in polarity of the semantic context eect with a change in context modality. In addition, an attempt was made to simulate the time courses of semantic facilitation and semantic interference. The rightmost column in Table 2 shows the simulated dierences between the semantically related and unrelated context conditions in the various conditions of Experiment 2. As is evident from the simulation results, the model successfully simulated the reversal from semantic inter6 In the simulations, the assumption was made that the threshold does not apply to the spreading of activation in the reverse direction: from lexical representations to conceptual representations (as it occurs in word perception). This assumption, however, was not crucial in simulating the reversal from semantic interference into facilitation.

ference with word context into semantic facilitation with picture context. Moreover, the model appeared quite successful in simulating the time course of semantic interference and facilitation. The correlation between the observed data and the simulated data was high (r :96; p < :005). Most importantly, the model simulated the lack of a semantic interference eect with word context at an SOA of )250 ms. This result appeared to be due to the decay of the activation of the lexical representation of the context word by the time the target was presented. Finally, it should be noted that the modied model was still capable of simulating the word-context eects reported by Starreveld and La Heij (1996): phonological facilitation, semantic interference and the interaction between these two factors. In the modied model, the assumption is made that context pictures do not activate their lexical representations. Therefore, a prediction that can be derived from our model is that context pictures should not induce any eect at the lexical level, including phonological facilitation (La Heij et al., 1990a; Posnansky & Rayner, 1977; Schriefers, Meyer, & Levelt, 1990; Starreveld & La Heij, 1995). That is, the translation of an English target word (e.g., WITCH) into the Dutch translation equivalent (e.g., HEKS) should not be facilitated by a context picture (e.g., a fence; Dutch name HEK), whose name has a strong phonological similarity to the correct response word. In Experiment 4, we put this prediction to the test.

Experiment 4 In the Experiments 4a and 4b, the prediction was tested that, in contrast to context words, context pictures do not induce phonological facilitation. As in the previous experiments, the target stimuli were to-betranslated English words (e.g., the target WITCH that should be translated into the Dutch word heks). In the word-context condition, these target words were accompanied by phonologically related context words (e.g., the word HEK, the Dutch translation equivalent of fence) or unrelated context words (e.g., the word KANON, the Dutch translation equivalent of canon). In the picture context condition, the corresponding pictures replaced the context words. For example, the target WITCH (Dutch: heks) was accompanied by the picture of a fence (hek) or the picture of a canon (kanon). In Experiment 4a, the SOA values )250 and 0 ms were used. Participants did not receive any training in naming the context picture to prevent them from noticing the phonological similarity between the names of the context pictures and the correct responses and from using this similarity strategically. In Experiment 4b, only one SOA was used (0 ms) andas a stronger test of the predictionparticipants did receive prior training in naming the context pictures.

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

479

Experiment 4a It takes more time to name a picture than to name a word (see Cattell, 1886; Fraisse, 1969). Previous research has shown that with the highly familiar pictures used in our experiments this dierence typically amounts to 200300 ms (see, e.g., Glaser & Dngelho, 1984). In u combination with Starreveld and La Heijs (1996) nding that context words produce phonological facilitation in the SOA range of )200 up to +100 ms, this leads to the expectation that if context pictures induce phonological facilitation this eect will be maximal with a short pre-exposure of the picture. Therefore, in Experiment 4a, in addition to the simultaneous presentation of target and context, also an SOA of )250 ms was used. In this experiment, in contrast to Experiment 4b, participants did not receive practice in naming the context pictures to prevent them from adopting a strategy of using the initial phonemes of the name of the context picture in the translation task. After completion of the experiment, the participants were asked to name all context pictures. Data of trials in the word translation task that involved context pictures that were named incorrectly in this test (that is, context pictures that were given a name dierent from the one in Appendix C) were removed from the analyses on a subject-by-subject basis. Method Participants. Thirty-two Leiden University students served as paid participants. They were native Dutch speakers and highly procient in the English language. Sixteen participants were randomly assigned to the word-context condition, the others to the picture-context condition. Stimuli. The targets were 32 to-be-translated English words (Appendix C). The targets had no phonological relation with the Dutch translation equivalent, that is, no cognates were used. For each of the response words a phonologically related Dutch context word and the corresponding picture were selected. For example, in the phonologically related condition, the picture of a book or the word BOEK (the Dutch translation equivalent of book) accompanied the target word FARMER (to be translated into the Dutch word boer). The unrelated stimuli were created by re-pairing these sets of words. The phonologically related and unrelated context stimuli did not have an obvious semantic or associative relation with the accompanying target word. The Dutch translation equivalents of the English target words and the Dutch context words had similar mean language frequencies (log frequencies of 1.76 and 1.63, respectively; CELEX database, Burnage, 1990). The pictures were selected from the line drawings provided by Snodgrass and Vanderwart (1980). The same display layout was used as in the previous experiments.

Procedure. The procedure was similar to the one in Experiments 2 and 3. Two SOA values were used: )250 ms (context presented rst) and 0 ms (simultaneous presentation). The participants were rst shown a list with the 32 English target words and their translations. Next, they received a practice series in which the 32 target words were presented in isolation. Finally, they received two experimental series corresponding to the two SOA conditions. The order of presentation of these series was balanced across participants. Each series consisted of 64 trials each (32 English target words 2 context conditions) and was preceded by eight practice trials that were randomly chosen from the experimental materials. At the end of the experimental session, participants were shown a list of the context pictures and they were asked to name them. Results. The data were treated in the same way as in the other experiments (3.61% of the data were removed). In addition, RTs of trials involving context pictures that were incorrectly named in the follow-up test series were excluded from the analyses on a subject-by-subject basis (10.55% on average). Table 4 shows the mean response latencies in the various experimental conditions. ANOVAs were performed with target-context relation (phonologically related and unrelated) and SOA as within-participant factors and with context modality (words versus pictures) as between-participants and within-items factor. These analyses showed a signicant main eect of context modality, F1 1; 30 6:07; p < :05; MSe 18,898.52, and F2 1; 31 200:73; p < :001; MSe 3582:68, reecting that the mean response latencies were smaller with a word (707 ms) than with a picture context (767 ms). Also the main eects of the factors target-context relation, F1 1; 30 9:26; p < :005; MSe 1079:43, and F2 1; 31 12:92; p < :001; MSe 2376:29, and SOA were signicant, F1 1; 30 11:26; p < :01; MSe 3903:87 and F2 1; 31 46:24; p < :001; MSe 1662:19. The interaction between context modality and target-context relation was signicant, F1 1; 30 4:69; p < :05; MSe 1079:43 F2 1; 31 2:17; p < :151; MSe 2166:02, indicating that the eect of a phonological relation between target and context diered for word and picture contexts. The t tests showed that, in the word context condition, the mean response latencies were signicantly smaller with phonologically related (692 ms) than with unrelated context stimuli (723 ms), t1 (15) 3.39, p < :01; t2 (31) 3.25, p < :01. In the picture context condition, this dierence was not signicant. A signicant interaction was obtained between SOA and targetcontext relation, F1 1; 30 5:55; p < :05; MSe 496:36 (F2 1; 31 3:84; p < :059; MSe 1376:81). The interaction between SOA and context modality only reached signicance in the item analysis, F1 1; 30 1:40; p < :246; MSe 3861:31; F2 1; 31 8:17; p < :01; MSe 1376:20.

480

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

Table 4 Mean RTs (in ms) and error percentages in the various experimental conditions of Experiments 4a and 4b Phonologically Related RT Experiment 4a. Word context SOA )250 0 Picture context SOA )250 0 Experiment 4b Word context Picture context %Error RT Unrelated %Error Phonological Relatedness eect (unrelated ) related)

721 664

3.3 3.5

745 700

2.9 3.5

+24 +36

782 747

3.1 1.8

775 764

4.5 4.1

)7 +17

770 825

4.9 7.0

825 811

6.3 7.8

+55 )14

Having found that the eect of a phonological relation between target and context diered for a word and picture context, we performed separate ANOVAs on the data of the word- and picture context condition. For the word-context condition signicant main eects were obtained for the factors target-context relation (related, unrelated), F1 1; 15 11:50; p < :01; MSe 1272:98, and F2 1; 31 10:55; p < :01; MSe 2700:77, and SOA, F1 1; 15 17:42; p < :001; MSe 2372:71, and F2 1; 31 100:14; p < :001; MSe 733:56. For the picture context condition, the main eect of SOA was signicant in the item analysis, F1 1; 15 1:59; p < :226; MSe 5419:30; F2 1; 31 6:36; p < :05; MSe 2304:83. The interaction between target-context relatedness and SOA reached signicance in the subject analysis, F1 1; 16 5:84; p < :05; MSe 410:78 F2 1; 31 2:36; p < :135; MSe 1830:51. Separate t tests performed on the data of the two SOA conditions showed that neither the )7 ms dierence at SOA )250 ms (p > :30) nor the 17 ms dierence at SOA 0 ms (p > :10) reached signicance. Inspection of Table 4 reveals that the error percentages paralleled the latency data.. The error percentages were too low to allow a meaningful analysis. Discussion. In accordance with the results of many studies on picture naming (e.g., Posnansky & Rayner, 1977; Schriefers et al., 1990) and with the results of a study on word translation (La Heij et al., 1990a), context words induced a phonological facilitation eect (24 ms facilitation at SOA )250 and 36 ms facilitation at SOA 0 ms). In accordance with the prediction of our model, context pictures did not induce a signicant phonological facilitation eect. In the introduction we argued that because of the time that it takes for pictures to activate their lexical representations, such an eect

should be maximal with a short pre-exposure of the context picture. This was clearly not what we obtained: at SOA )250 ms, the phonological facilitation eect was virtually zero. The 17 ms dierence that was obtained at SOA 0 did not reach signicance. However, given the size and direction of this dierence between the phonologically related and unrelated conditions, we decided to reexamine this condition in Experiment 4b.

Experiment 4b In this experiment only the SOA 0 condition was examined. In addition, to test the prediction of our modied model more rigorously, participants in this experiment were trained in using the correct names for the context picture in advance of the experimental series. Again, the prediction was that if the context pictures do not automatically activate their names, no phonological facilitation eect should be observed. Participants. Thirty-two Leiden University students served as paid participants. They were native Dutch speakers and highly procient in the English language. Sixteen participants were randomly assigned to the word-context condition, the others to the picture-context condition. Stimuli. The same targets were used as in Experiment 4a. Procedure. The procedure was similar to the one in Experiments 2 and 3. The participants were rst shown a list containing the 32 English target words and their translations. Next, the 32 target words were presented without context and the participants were required to translate them as fast as possible while maintaining accuracy. Finally, they received one experimental series of

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

481

64 trials (32 target words 2 context conditions), preceded by eight practice trials that were randomly chosen from the experimental materials. Results. The data were treated in the same way as in Experiments 13 (6.49% of the data were removed). Table 4 shows the mean response latencies in the various experimental conditions. ANOVAs were performed with target-context relation (phonologically related and unrelated) as withinparticipant factor and with context modality (words versus pictures) as between-participants and withinitems factor. In the item analysis a signicant main eect was obtained for context modality, F1 1; 30 0:22; p < :642; MSe 30,018.25, F2 1; 31 7:01; p < :05; MSe 2097:040. In addition, the interaction between context modality and target-context relation were signicant, F1 1; 30 9:02; p < :01; MSe 2099:13; F2 1; 31 9:28; p < :01; MSe 4388:74, indicating that the eect of a phonological relation between target and context diered for word and picture contexts. The t tests showed that, in the word context condition, the mean response latencies were signicantly smaller with phonologically related than with unrelated context stimuli (a phonological facilitation eect of 55 ms), t1 15 3:77; p < :01; t2 31 3:34; p < :01. In the picture context condition, this dierence was not signicant. The error percentages in this experiment were somewhat higher than in the previous experiments, but were still judged too small to allow a meaningful analysis. The pattern of error percentages paralleled the latency data. Discussion. The results again show that the phonological similarity between the context words and the response words was strong enough to induce a substantial amount of facilitation (55 ms). In contrast to this nding with context words, but in accordance with the prediction of our model, context pictures did not induce a signicant phonological facilitation eect ()13 ms). In addition, an ANOVA performed on the combined data of Experiment 4a (picture context, SOA 0 condition) and Experiment 4b, with target-context relation (phonological related versus unrelated) as within-participant factors and experiment (4a and 4b) as betweenparticipants factor revealed no signicant phonological facilitation eect. The mean response latencies in the phonological related and unrelated conditions were 786 and 788 ms, respectively (F < 1:0). Therefore, Experiments 4a and 4b clearly show that context words induce phonological facilitation, but context pictures do not. This nding is completely in accordance with the prediction of the modied Starreveld and La Heij (1996) model. It should be noted that Experiments 4a and 4b provide a test of our proposed modication within models of language production that assume cascaded processing (e.g., Caramazza, 1997;

Dell, 1986; Starreveld & La Heij, 1996). Serial, two-stage models (e.g., Levelt et al., 1999), in which only the selected lemma is phonologically encoded, do not predict phonological eects of context pictures, even if these pictures activate their lexical representations (lemmas). Of course, these models still face the problem of accounting for the semantic facilitation eects observed in our Experiments 1 and 2. Very recently, Morsella and Miozzo (2002) published experimental results that seem to be at variance with our present ndings. These authors presented two superimposed linedrawings, one in green and one in red and asked their participants to name the green picture and to ignore the red one. The main experimental variable was the presence or absence of a phonological relation between the names of the two pictures. In the single experiment reported, such a phonological context eect was obtained. A possible cause of the discrepancy between this nding and the results of our Experiments 4a and 4b is that the presentation of two superimposed pictures may have induced selection problems, whereas target selection in our experiments was probably easy because of the large physical dierence between the target (a word) and the context (a line drawing). Possible selection problems in part of the trials of Morsella and Miozzos experiment may have resulted in the activation of the context pictures name at a lexical level. Clearly, this interpretation of the discrepancy between the two studies is in need of further investigation. It is important to note that Morsella and Miozzos (2002) observation of phonological facilitation in a picturepicture task is not only at variance with our present ndings, but is also hard to reconcile with the lack of semantic interference in the picture-picture tasks reported by Humphreys et al. (1995) and Humphreys et al. (1995). If a context picture activates its name, as suggested by the results of Morsella and Miozzo, all language production models predict that also semantic interference should be obtained. Finally, it should be mentioned that in Experiments 4a and 4b response latencies were again somewhat larger when the target words were accompanied by context pictures than when accompanied by context words. As argued in Discussion of Experiment 1, this nding is hard to interpret, given the dierences in display characteristics between the two context conditions and given the fact that the two conditions did not dier in Experiment 1, in which a within-participants design was used.

General discussion A semantic relation between a target stimulus and a context stimulus facilitates performance in some language production tasks, but hampers performance in

482

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

others. Results obtained in word-translation tasks reported by La Heij et al. (1990a) and La Heij et al. (1996) suggest that one factor that determines the polarity of the semantic context eect is the modality of the context stimulus: a word or a picture. In Experiment 1, we rst determined whether context modality is indeed a crucial factor underlying the reversal from semantic interference with word context into semantic facilitation with picture context. This proved to be the case. The nding of semantic facilitation with picture context is particularly interesting, because current models of language production predict that context pictures, like context words, induce semantic interference. As discussed in the introduction, this prediction is a consequence of the models assumption that concepts corresponding to the context pictures automatically activate their names at the lexical level and that more activation spreads from the target concept to the context concept than vice versa. Before concluding that this aspect of current language production models needs modication, two alternative accounts of the semantic facilitation eect induced by context pictures were evaluated. First, Experiment 2 showed that context pictures induce semantic facilitation even when presented 250 ms in advance of the target; a nding that eliminates the hypothesis that the semantic facilitation eect at SOA 0 is due to the fact that context pictures take more time to activate their names than context words do (the relative speed hypothesis; see Glaser & Dngelho, 1984). Next, Exu periment 3 eliminated the hypothesis that the semantic facilitation eect was due to the naming of context pictures at a superordinate-category level (e.g., bird) instead of the basic-category level (e.g., swan). In the Discussion of Experiment 3, we concluded that our results indicate that the assumption of current models of language production that activated concepts automatically activate the corresponding lexical representations is incorrect. Interestingly, this conclusion is not new: within the area of Stroop interference research on the lack of a reversed Stroop eect (color-word reading is not hampered by an incongruent color) is attributed to the fact that colors do not automatically activate their names. In addition, we argued that our conclusion is in line with results obtained in colorcolor variants of the Stroop task reported by La Heij et al. (1993a) and in a picture-naming task with picture context, reported by Humphreys et al. (1995). Finally, our conclusion also seems in line with Levelts (1989) original proposal that only the conceptual information that the speaker wants to express (the preverbal message) is lexicalized. The assumption that only the target concept is used in the process of lexical access was implemented in Starreveld and La Heijs (1996) connectionist model as a threshold at the conceptual level. Only the concept that is selected (that is, the concept that receives task acti-

vation) reaches threshold and activates lexical representations. To account for the semantic interference eect obtained with context words, the additional assumption had to be made that the target concept activates a cohort of semantically related words. Another way to achieve this cohort activation is to assume semantically decomposed conceptual representations (semantic features). However, besides arguments that have been raised against decompositionality (see, e.g., Roelofs, 1992), it is questionable whether feature-based models will be able to account for the reversal from semantic interference into semantic facilitation. Given the general lack of knowledge about the representation of word meaning, our proposal has the virtue of being a rather straightforward extension of current language production models. To summarize, we propose that conceptually-driven lexical access starts with the selection of a single conceptual representation (or preverbal message). Only this conceptual representation (e.g., horse) is used in the process of lexical access. Other concepts, activated by nontarget objects in the visual eld (e.g., the concepts barn, grass, and fence) do not take part in that process. To account for semantic errors (e.g., saying cow instead of horse) and semantic interference eects in Strooplike naming tasks, we assume that the process of lexical access leads to the activation of a cohort of semantically related words. The modied version of Starreveld and La Heijs (1996) model successfully simulated the reversal from semantic interference with word context into semantic facilitation with picture context. Moreover, the model simulated the lack of semantic interference when a context word was pre-exposed by 250 ms. Importantly, the model was still capable of simulating the main ndings reported by Starreveld and La Heij: phonological facilitation, semantic interference and the interaction between these factors. In the remainder of this discussion, two issues will be examined. First, we examine the generalizability of our ndings, obtained with the word translation paradigm, to other word-production tasks like picture naming. In that discussion, we will also examine the results of Glaser and Glasers (1989) picture-naming task with picture context that seem to contradict our present proposal. Second, we will discuss in what way our model can be extended to account for results of semantic priming studies in which target pictures are preceded by word primes. In this study we assumed that word translation, like picture naming, is conceptually mediated andfor that reasoncan be used to study language production (see Potter, So, von Eckardt, & Feldman, 1984; Snodgrass, 1993; see Jescheniak & Levelt, 1994, for a similar use of the word-translation task). However, this assumption is not agreed upon by all researchers in the area of bilin-

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

483

gualism. Notably, Kroll and Stewart (1994) proposed that translation from L1 to L2 is conceptually mediated, but that translation from L2 to L1 (used in our experiments) is based on wordword associations at a lexical level. Kroll and Stewarts proposal was largely based on their nding that forward translation was aected by a semantic manipulation (the presentation of words in randomized or semantically categorized lists), whereas backward translation was not. However, since then, many studies have shown that translation from L2 to L1 is aected by semantic factors. Semantic context eects were reported by La Heij et al. (1990a) and La Heij et al. (1996) and, in addition, De Groot, Dannenburg, and Van Hell (1994) reported evidence in favor of semantic involvement in backward translation. For now, we conclude that if the presence of semantic context eects is indicative for conceptual mediation, as argued by Kroll and Stewart (1994), backward translation in our experiments was clearly conceptually mediated. This is not to say that there are no dierences between picture naming and word translation. In the Discussion of Experiment 3 we mentioned the fact that pictures dier from words in the speed with which they activate conceptual representations. This can be inferred from the observation that word categorization is slower than picture categorization (e.g., Glaser & Dngelho, u 1984; Potter & Faulconer, 1975). Consequently, it is likely that the identication (concept activation) of an L2 target word will prot more from the presence of a semantically related context picture than the identication of a target picture. Indeed, a comparison between data reported by Glaser and Dngelho (1984) and u Glaser and Glaser (1989) suggests that picture context has a larger eect on word categorizing than on picture categorizing. Given this conclusion, it seems likely that a semantically related context picture will have a smaller facilitating eect on picture naming than on word translation. This is exactly what was found: La Heij et al. (submitted) reported a small semantic facilitation eect in a picture-naming task with picture context and Humphreys et al. (1995) and Damian and Bowers (in press) reported no semantic context eect. The only study in which a semantic interference eect was deserves discussion. Glaser and Glasers (1989, Experiment 6) reported a semantic interference eect in a picture-naming task with picture context. Inspection of this experiment, however, reveals a number of unusual characteristics that may have been responsible for this deviant nding. Damian and Bowers (in press) and La Heij et al. (submitted) discuss a number of these characteristics. First, in Glaser and Glasers experiment only nine target pictures were used that were selected from only three semantic categories. It seems likely that reducing the number of semantic categories will decrease the size of a

semantic facilitation eect (see La Heij et al., 1985). Second, these nine target pictures were also used as context pictures. Therefore, a picture that had to be named in one trial could be identical to the picture that had to be ignored in the next trial. Third, the target picture was presented above or below a central xation point, and the context picture appeared in the opposite position. The target picture was dened as either the rst or the second picture that appeared on the display (the sequential discrimination task). With SOAs as small as 50100 ms this task is dicult and it is conceivable that in a number of trials instead of the target, the context picture was selected for naming, resulting in an error or a time-consuming correction. In an experiment in which these characteristics of Glaser and Glasers (1989) experiment were eliminated (for example, the context picture was embedded in the target picture which rendered target selection much easier), Damian and Bowers (in press) failed to obtain semantic interference. La Heij et al. (submitted), used in their Experiment 1 the same experimental stimuli and display characteristics as in Glaser and Glasers study and replicated their semantic interference eect. They also showed, however, that the semantic interference eect was accompanied by a relatively large number of incorrect selections (the context picture was named instead of the target). In a second experiment, they showed that the semantic interference eect reversed into a small semantic facilitation eect when the number of pictures was increased, the context pictures were not part of the target set and the selection of the target picture was facilitated. La Heij et al. concluded, as Damian and Bowers did, that Glaser and Glasers semantic interference eect was most probably due to the unusual characteristics of their experiment. For that reason, Glaser and Glasers semantic interference eect does not pose a challenge for the modied model we proposed. Finally, the relation between our present experiments and semantic priming studies deserves some discussion. In Experiment 2, we manipulated SOA and found that when a context word preceded the target by 250 ms, no semantic context eect was obtained. It is known from the priming literature that a larger pre-exposure of a prime word may even lead to semantic facilitation in picture naming (e.g., Bajo, 1988; Carr, McCauley, Sperber, & Parmelee, 1982). To the best of our knowledge, there is no pictureword study in which manipulation of SOA resulted in a reversal from semantic interference into semantic facilitation. If such a result could be obtained in a word translation task, it would be as problematic for current models of language production as our present eect of context modality. Again, the problem is in the models assumption that all activated concepts activate their names at the lexical level. Because of this assumption, the conceptual representation

484

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

activated by a prime word continuously reactivates its lexical representation. Due to spreading activation between related concepts, this reactivation will be stronger in the case of semantically related prime-target pairs than in the case of unrelated pairs, resulting in semantic interference instead of semantic facilitation. The modication we proposed, that only the target concept is lexicalized, resolves at least part of this problem: the prime word activates its conceptual representation but due to the thresholdthis activation does not feed back to the lexical level. When the lexical representation of the prime word is allowed to decay (by introducing a time interval between prime and target), semantic interference should disappear. Indeed, our simulations showed that at SOA )250 ms context words do not induce a semantic context eect. In our current research we address the issue whether in a word-translation task with prime word, manipulation of SOA induces a reversal from semantic interference into semantic

facilitation and whether our modied model is capable of accounting for such a change in polarity of the semantic context eect. In conclusion, we have shown that in a word translation task, context words induce semantic interference whereas context pictures induce semantic facilitation. We have argued that this nding can most easily be accounted for by assuming that in language production only the target concept (Levelts (1989), preverbal message) is lexicalized.

Acknowledgments The authors thank Steve Lupker, Judith Kroll, Rob Hartsuiker and an anonymous reviewer for very helpful comments on an earlier version of this article.

Appendix A
The target words and context stimuli used in Experiments 1 and 2. English translations of the Dutch context words are given in parentheses. Target word Pig Horse Duck Donkey Dog Deer Pigeon Frog Ant Shark Plane Garlic Lemon Rabbit Cherry Potato Watch Spoon Bottle Saw Knife Trousers Dress Skirt Boot Leg Eye Suitcase Window Chair Glove Bike Translation Varken Paard Eend Ezel Hond Hert Duif Kikker Mier Haai Vliegtuig Knoook Citroen Konijn Kers Aardappel Horloge Lepel Fles Zaag Mes Broek Jurk Rok Laars Been Oog Koer Raam Stoel Handschoen Fiets Related context Geit Koe Kip Zebra Kat Buel Zwaan Slak Spin Doljn Trein Ui Aardbei Eekhoorn Appel Sla Klok Vork Glas Hamer Bijl Jas Trui Blouse Schoen Arm Neus Mand Deur Bank Sok Auto (goat) (cow) (chicken) (zebra) (cat) (bualo) (swan) (snail) (spider) (dolphin) (train) (onion) (strawberry) (squirrel) (apple) (lettuce) (clock) (fork) (glass) (hammer) (ax) (coat) (sweater) (blouse) (shoe) (arm) (nose) (basket) (door) (bench) (sock) (car) Unrelated context Glas Sok Mand Auto Sla Blouse Klok Arm Deur Aardbei Neus Jas Doljn Schoen Zebra Kat Slak Eekhoorn Zwaan Appel Bank Koe Geit Buel Spin Ui Bijl Vork Trui Trein Kip Hamer (glass) (sock) (basket) (car) (lettuce) (blouse) (clock) (arm) (door) (strawberry) (nose) (coat) (dolphin) (shoe) (zebra) (cat) (snail) (squirrel) (swan) (apple) (bench) (cow) (goat) (bualo) (spider) (onion) (ax) (fork) (sweater) (train) (chicken) (hammer)

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

485

Appendix B
The target words and context stimuli used in Experiment 3. English translations of the Dutch context words are given in parentheses. Target word (English) Car Bike Plane Onion Carrot Lettuce Couch Closet Chair Grape Lemon Cherry Dress Trousers Coat Yellow Target word (English) Black Purple Horse Lion Pig Lawyer Teacher Dentist Eye Chest Face Church Factory Farm Translation (Dutch) Auto Fiets Vliegtuig Ui Wortel Sla Bank Kast Stoel Druif Citroen Kers Jurk Broek Jas Geel Translation (Dutch) Zwart Paars Paard Leeuw Varken Advocaat Leraar Tandarts Oog Borst Gezicht Kerk Fabriek Boerderij Related basic level Motor (motor) Trein (train) Boot (boat) Tomaat (tomato) Boon (bean) Erwt (pea) Tafel (table) Plank (shelf) Bed (bed) Perzik (peach) Banaan (banana) Appel (apple) Trui (sweater) Hemd (shirt) Sok (sock) Rood (red) Related Basic Level Groen (green) Blauw (blue) Kat (cat) Konijn (rabbit) Hond (dog) Dokter (doctor) Bakker (baker) Slager (butcher) Been (leg) Enkel (ankle) Arm (arm) Flat (at) Museum (museum) Huis (house) Unrelated basic level Trui(sweater) Perzik (peach) Tafel (table) Hemd (shirt) Sok (sock) Been (leg) Hond (dog) Banaan (banana) Rood (red) Huis (house) Arm (arm) Blauw (blue) Erwt (pea) Dokter (doctor) Flat (at) Konijn (rabbit) Unrelated Basic Level Plank (shelf) Boon (bean) Museum (museum) Enkel (ankle) Bed (bed) Appel (apple) Groen (green) Boot (boat) Kat (cat) Motor (motor) Slager (butcher) Bakker (baker) Tomaat (tomato) Trein (train) Related super level Voertuig (vehicle) Voertuig (vehicle) Voertuig (vehicles) Groente (vegetable) Groente (vegetable) Groente (vegetable) Meubel (furniture) Meubel (furniture) Meubel (furniture) Fruit (fruit) Fruit (fruit) Fruit (fruit) Kleding (clothes) Kleding (clothes) Kleding (clothes) Kleur (color) Related Super Level Kleur (color) Kleur (color) Dier (animal) Dier (animal) Dier (animal) Beroep (occupation) Beroep (occupation) Beroep (occupation) Lichaamsdeel (body part) Lichaamsdeel (body part) Lichaamsdeel (body part) Gebouw (building) Gebouw (building) Gebouw (building) Unrelated super level Groente (vegetable) Meubel (furniture) Fruit (fruit) Kleding (clothes) Kleur (color) Dier (animal) Beroep (occupation) Lichaamsdeel (body part) Gebouw (building) Voertuig (vehicle) Meubel (furniture) Dier (animal) Fruit (fruit) Beroep (occupation) Gebouw (building) Voertuig (vehicle) Unrelated Super Level Beroep (occupation) Meubel (furniture) Lichaamsdeel (body part) Fruit (fruit) Kleding (clothes) Groente (vegetable) Lichaamsdeel (body part) Kleur (color) Voertuig (vehicle) Kleur (color) Gebouw (building) Groente (vegetable) Dier (animal) Kleding (clothes)

486

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488

Appendix C
The target words and context stimuli used in Experiment 4. English translations of the Dutch context words are given in parentheses. Target word Reply Can Farmer Fur Neighbour Curtain Shark Trade Witch Oce Throat Suitcase Bullet Grain Herb Army Ribbon Fate War Horse Suit Arrow Mail Council Juice Candy Sign Looks Union Cattle Paint Cradle Translation Antwoord Blikje Boer Bont Buurman Gordijn Haai Handel Heks Kantoor Keel Koer Kogel Korrel Kruid Leger Lint Lot Oorlog Paard Pak Pijl Post Raad Sap Snoepje Teken Uiterlijk Vakbond Vee Verf Wieg Related context Anker Bliksem Boek Borstel Bureau Golf Hamer Hand Hek Kanon Ketel Kompas Konijn Kopje Kruis Lepel Lipstick Longen Oor Paprika Palm Pijp Potlood Raam Saxofoon Snoer Teen Uil Varken Veer Vest Wiel (anchor) (lightning) (book) (brush) (oce) (wave) (hammer) (hand) (fence) (canon) (kettle) (compass) (rabbit) (cup) (cross) (spoon) (lipstick) (lungs) (ear) (paprika) (palm) (pipe) (pencil) (window) (saxophone) (wire) (toe) (owl) (pig) (feather) (cardigan) (wheel) Unrelated context Ketel Vest Lipstick Wiel Potlood Longen Kruis Konijn Kanon Sax Hamer Paprika Anker Palm Veer Boek Kompas Hek Kopje Bliksem Oor Lepel Teen Borstel Raam Bureau Uil Pijp Hand Snoer Golf Varken (kettle) (cardigan) (lipstick) (wheel) (pencil) (lungs) (cross) (rabbit) (canon) (saxophone) (hammer) (paprika) (anchor) (palm) (feather) (book) (compass) (fence) (cup) (lightning) (ear) (spoon) (toe) (brush) (window) (oce) (owl) (pipe) (hand) (wire) (wave) (pig)

Appendix D
Simulations with Starreveld and La Heijs (1996) model of lexical access. Starreveld and La Heijs (1996) model consists of two major layers, a layer of concept nodes and a layer of phonological nodes. For our present purposes, these phonological nodes can also be interpreted as lexical representations or lemmas (Kempen & Huijbers, 1983; Levelt et al., 1999). As indicated in Fig. 1, at the layer of concept nodes, semantic (categorical) relations between concepts are represented by connections between the corresponding concept nodes. In the model, these connections are bi-directional and have the same weights (wcc ). All connections between the nodes at the conceptual layer and the corresponding word nodes at the lexical layer are bi-directional and have the same weights (wcp ). A third layer (not shown in Fig. 1) consists of input nodes that represents the input from the visual system to the concept layer (in the case of pictures or to-be-translated words) and to the lexical layer (in the case of context words). The connections to these layers have a weight wi . A node representing the task of picture naming (or word translation) represents the result of an attentional selection process and connects with weight wt to the target concept. In each iteration, the activation of each node i(Ai ) at time t 1 is calculated using the CALM-activation rule (Murre, 1992)

Ai t 1 1 kAi t with ei X
j

ei 1 1 kAi t 1 ei

wij Aj t:

Using this rule, the minimum activation value of each node is 0, and the maximum value is 1. The rst component of the rule, 1 kAi t, represents the autonomous decay of a node. The presentation of a target (picture or to-be-translated word) and of a context (picture or word) can be simulated by presenting input activation Ain to the corresponding input nodes. Selection of a response word takes place when the activation of a lexical node exceeds that of all other lexical nodes by a critical amount c (see Roelofs, 1992, for a selection mechanism that

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488 incorporates a similar parameter). If the selection process is more dicult, the selection will take more iterations, and the simulated RT will be longer (each iteration represented a time step of Dt, which was set at 10 ms). The parameter values used in the original simulations were as follows (for further details, see Starreveld & La Heij, 1996):

487

c 0:071; wcc 0:01;

wcp 0:025; k 0:03;

wi 0:14; Ain 0:025;

wt 1:6; Dt 10:

In the simulations of the model depicted in Fig. 2, a threshold was introduced at the conceptual level with a value of 0.4 and concepts activated semantically related lexical representations (cohort activation) with a weight of .003 (12% of the weight of the link between a concept and its correct name, wcp ). To obtain semantic facilitation with picture context and semantic interference with word context, only the amount of spreading activation within the conceptual system (wcc ) had to be increased (from 0.01 to 0.03). With this adjusted value, the model was able to simulate the pattern of results obtained by Starreveld and La Heij (1996; semantic interference, phonological facilitation & an interaction between semantic & phonological relatedness, all with word context) and our present nding of semantic facilitation with picture context (see the simulation results in Table 2). A satisfactory quantitative t with all empirical observations was obtained when the value of Dt was increased from 10 to 20 ms. The context eects that were obtained with these modications are shown in Table 2.

References
Bajo, M.-T. (1988). Semantic facilitation with picture and words. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 579589. Burnage, G. (1990). CELEX: A guide for users. Nijmegen, The Netherlands: CELEX. Caramazza, A. (1997). How many levels of processing are there in lexical access? Cognitive Neuropsychology, 14, 177208. Caramazza, A., & Costa, A. (2000). The semantic interference eect in the pictureword interference paradigm: Does the response set matter? Cognition, 75, 5164. Caramazza, A., & Costa, A. (2001). Set size and repetition in the pictureword interference paradigm: Implications for models of naming. Cognition, 80, 291298. Carr, T. H., McCauley, C., Sperber, R. D., & Parmelee, C. M. (1982). Words, pictures, and priming: On semantic activation, conscious identication, and the automaticity of information processing. Journal of Experimental Psychology: Human Perception and Performance, 8, 757777. Cattell, J. M. (1886). The time it takes to see and name objects. Mind, 11, 6365. Collins, A. M., & Loftus, E. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82, 407428. Costa, A., Miozzo, M., & Caramazza, A. (1999). Lexical selection in bilinguals: Do words in the Bilinguals two lexicons compete for selection? Journal of Memory and Language, 41, 365397.

Damian, M. F., & Bowers, J. S. (in press). Locus of semantic interference in pictureword interference tasks. Psychonomic Bulletin & Review. De Groot, A. M. B., Dannenburg, L., & Van Hell, J. G. (1994). Forward and backward translation by bilinguals. Journal of Memory and Language, 33, 600629. Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93, 283321. Driver, J., & Tipper, S. P. (1989). On the nonselectivity of selective seeing: Contrasts between interference and priming in selective attention. Journal of Experimental Psychology: Human Perception and Performance, 15, 304 314. Fraisse, P. (1969). Why is naming longer than reading? Acta Psychologica, 30, 96103. Glaser, W. R., & Dngelho, F.-J. (1984). The time course of u pictureword interference. Journal of Experimental Psychology: Human Perception and Performance, 10, 640654. Glaser, W. R., & Glaser, M. O. (1982). Time course analysis of the Stroop phenomenon. Journal of Experimental Psychology: Human Perception and Performance, 8, 875894. Glaser, W. R., & Glaser, M. O. (1989). Context eects on Stroop-like word and picture processing. Journal of Experimental Psychology: General, 118, 1342. Green, D. W. (1998). Mental control of the bilingual lexicosemantic system. Bilingualism: Language and Cognition, 1, 6781. Humphreys, G. W., Lloyd-Jones, T. J., & Fias, W. (1995). Semantic interference eects on naming using a postcue procedure: Tapping the links between semantics and phonology with pictures and words. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 961980. Jescheniak, J. D., & Levelt, W. J. M. (1994). Word frequency eects in speech production: Retrieval of syntactic information and of phonological form. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 824843. Jonides, J., & Mack, R. (1984). On the cost and benet of cost and benet. Psychological-Bulletin, 96, 2944. Kempen, G., & Huijbers, P. (1983). The lexicalization process in sentence production and naming: Indirect election of words. Cognition, 14, 185209. Klein, G. S. (1964). Semantic power measured through the interference of words with colour-naming. American Journal of Psychology, 93, 245248. Kroll, J. K., & Stewart, E. (1994). Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language, 33, 149174. La Heij, W. (1988). Components of Stroop-like interference in picture naming. Memory & Cognition, 16, 400410. La Heij, W., de Bruyn, E., Elens, E., Hartsuiker, R., Helaha, D., & Van Schelven, L. (1990a). Orthographic facilitation and categorical interference in a word-translation variant of the Stroop task. Canadian Journal of Experimental Psychology, 44, 7683. La Heij, W., Dirkx, J., & Kramer, P. (1990b). Categorical interference and associative priming in picture naming. British Journal of Psychology, 81, 511525. La Heij, W., Heikoop, K. W., Akerboom, S. P., & Bloem, I. (submitted). Picture naming in picture context: Semantic interference or semantic facilitation?

488

I. Bloem, W. La Heij / Journal of Memory and Language 48 (2003) 468488 Potter, M. C., So, K.-F., von Eckardt, B., & Feldman (1984). Lexical and conceptual representation in beginning and procient bilinguals. Journal of Verbal Learning and Verbal Behavior, 23, 2338. Roelofs, A. (1992). A spreading-activation theory of lemma retrieval in speaking. Cognition, 42, 107142. Roelofs, A. (2001). Set size and repetition matter: Comment on Caramazza and Costa (2000). Cognition, 80, 283290. Rosinski, R. R. (1977). Pictureword interference is semantically based. Child Development, 48, 643647. Schneider, W. (1988). Micro Experimental Laboratory: An integrated system for IBM-PC compatibles. Behavior Research Methods, Instruments, and Computers, 20, 206 217. Schriefers, H., Meyer, A. S., & Levelt, W. J. M. (1990). Exploring the time course of lexical access in language production: Pictureword interference studies. Journal of Memory and Language, 29, 86102. Snodgrass, J. G., & Vanderwart, M. (1980). A standardized set of 260 pictures: Norms for name agreement, image agreement, familiarity, and visual complexity. Journal of Experimental Psychology: Human Learning and Memory, 6, 174 215. Snodgrass, J. G. (1993). Translating versus picture naming: Similarities and dierences. In R. Schreuder & B. Weltens (Eds.), The bilingual lexicon. Amsterdam: John Benjamins. Starreveld, P. A., & La Heij, W. (1995). Semantic interference, orthographic facilitation, and their interaction in naming tasks. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 686698. Starreveld, P. A., & La Heij, W. (1996). Time-course analysis of semantic and orthographic context eects in picture naming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 896918. Starreveld, P. A., & La Heij, W. (1999). What about phonological facilitation, response-set membership, and phonological co-activation? Behavioral and Brain Sciences, 22, 56 58. Underwood, G. (1976). Semantic interference from unattended printed words. British Journal of Psychology, 67, 327338. Underwood, G. (1986). Facilitation or inhibition from parafoveal words? The Behavioral and Brain Sciences, 9, 48. Van der Velde, F., & De Kamps, M. (2001). From knowing what to knowing where: Modeling object-based attention with feedback disinhibition of activation. Journal of Cognitive Neuroscience, 13(4), 479491. Vitkovitch, M., & Humphreys, G. W. (1991). Perseverant responding in speeded naming of pictures: Its in the links. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 664680. Vitkovitch, M., & Tyrrell, L. (1999). The eects of distractor words on naming pictures at the subordinate level. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 52A, 905926. Warren, R. E. (1977). Time and the spread of activation in memory. Journal of Experimental Psychology: Human Learning and Memory, 3, 458466.

La Heij, W., Helaha, D., & Van Den Hof, E. (1993a). Why does blue hamper the naming of red? Colourcolour interference and the role of locational (un)certainty. Acta Psychologica, 83, 159177. La Heij, W., Hooglander, A., Kerling, R., & Van der Velden, E. (1996). Nonverbal context eects in forward and backward word translation: Evidence for concept mediation. Journal of Memory and Language, 35, 648665. La Heij, W., Kaptein, N. A., Kal, A. C., & de Lange, L. (1995). Reducing colourcolour interference by optimizing selection for action. Psychological Research, 57, 119130. La Heij, W., Starreveld, P. A., & Steehouwer, L. C. (1993b). Semantic interference and orthographic facilitation in definition naming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 352368. La Heij, W., Van der Heijden, A. H. C., & Plooij, P. (2001). A paradoxical exposure-duration eect in the Stroop task: Temporal segregation between stimulus attributes facilitates selection. Journal of Experimental Psychology: Human Perception and Performance, 27, 622632. La Heij, W., Van der Heijden, A. H. C., & Schreuder, R. (1985). Semantic priming and Stroop-like interference in wordnaming tasks. Journal of Experimental Psychology: Human Perception and Performance, 11, 6280. Levelt, (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press. Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22, 175. Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). Multiple perspectives on word production. Behavioral and Brain Sciences, 22, 6169. MacLeod, C. M. (1991). Half a century of research on the Stroop eect: An integrative review. Psychological Bulletin, 109, 163203. Morsella, E., & Miozzo, M. (2002). Evidence for a cascade model of lexical access. Speech Production, 28, 555563. Morton, J. (1969). The interaction of information in word recognition. Psychological Review, 76, 165178. Murre, J. M. J. (1992). Categorizing and learning in neural networks. New York: Harvester Wheatsheaf. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106, 226254. Neumann, O. (1980). Informationsselektion und Handlungssteuerung [Information selection and action control]. Unpublished Doctoral Dissertation, University of Bochum, Federal Republic of Germany. Neumann, O. (1986). Facilitative and inhibitory eects of semantic relatedness (Report No. 111/1986). University of Bielefeld, Bielefeld, Federal Republic of Germany. Posnansky, C. J., & Rayner, K. (1977). Visual-feature and response components in a pictureword interference task with beginning and skilled readers. Journal of Experimental Child Psychology, 24, 440460. Potter, M. C., & Faulconer, B. A. (1975). Time to understand pictures and words. Nature, 253, 437438.

You might also like