Journal of Memory and Language 57 (2007) 195–209

Journal of Memory and Language www.elsevier.com/locate/jml

Time pressure and phonological advance planning in spoken production q Markus F. Damian *, Nicolas Dumay Department of Experimental Psychology, University of Bristol, 12a Priory Road, Bristol BS8 1TU, UK Received 20 July 2006; revision received 1 November 2006 Available online 26 December 2006

Abstract Current accounts of spoken production debate the extent to which speakers plan ahead. Here, we investigated whether the scope of phonological planning is influenced by changes in time pressure constraints. The first experiment used a picture–word interference task and showed that picture naming latencies were shorter when word distractors shared the final segments with the picture name. Experiment 2 used the same paradigm but with colored pictures to elicit determiner + adjective + noun phrases. Latencies were shorter when the distractor overlapped phonologically with the picture name. Finally, Experiment 3 demonstrated that in colored picture naming without distractors, latencies were shorter when the object noun began with the same phoneme as the color adjective. Crucially, in all experiments introduction of a response deadline accelerated latencies, but did not alter the relative magnitude of the priming effects. In sum, pressure to provide a swift response does not reduce the scope of phonological planning.  2006 Elsevier Inc. All rights reserved. Keywords: Spoken production; Phonological encoding

Introduction Speaking requires the conversion of the meaning of an utterance into a surface form that eventually can be articulated. The production of even very simple utterances is surprisingly complex and involves the interplay of a host of cognitive representational levels (semantic, syntactic, phonological, and articulatory). A key issue arising in recent research concerns the extent to which speakers plan ahead their utterances at each of these

q This research was supported by Grant BB/C508477/1 from the Biotechnology and Biological Sciences Research Council (BBSRC) to the first author. * Corresponding author. Fax: +44 117 928 8588. E-mail address: [email protected] (M.F. Damian).

levels. For instance, at the syntactic level, phrases (e.g., Ferreira, 1991; Levelt & Maassen, 1981) and clauses (e.g., Ford & Holmes, 1978; Meyer, 1996) have been suggested as ‘‘units’’ of syntactic planning. At the phonological level, ‘‘minimalist’’ theories of planning (e.g., Dell, Juliano, & Govindjee, 1993; Jordan, 1990; MacKay, 1987) stipulate that only a very small degree of advance planning is carried out. Theories of this type assume that as soon as the initial portion of an utterance (i.e., its first segment) becomes available, articulation starts. By contrast, ‘‘non-minimalist’’ accounts (e.g., Levelt, Roelofs, & Meyer, 1999; Wheeldon & Lahiri, 1997) assume that the portion of the utterance that is buffered before it is articulated is considerably larger. The unit of advance planning championed by these accounts is the ‘‘phonological word’’, which is defined as minimally a stressed foot and maximally a single

0749-596X/$ - see front matter  2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jml.2006.11.001

196

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

lexical word plus associated unstressed function words, such as auxiliaries, determiners, conjunctions, and prepositions (cf. Wheeldon, 2000). Wheeldon and Lahiri (1997) state that ‘‘when it is possible to do so, speakers preferentially initiate articulation following the phonological encoding of the initial phonological word of an utterance’’ (p. 377). Importantly, this account specifies a lower boundary on the degree of advance planning, but leaves open the possibility that speakers may opt to encode more than a single phonological word (see Alario, Costa, & Caramazza, 2002a; Alario, Costa, & Caramazza, 2002b; Levelt, 2002; for a clarifying debate on how the scope of advance planning should be defined). Speech errors such as anticipations (cup of coffee fi cuff of coffee; Fromkin, 1973) and segment exchanges (left hemisphere fi heft lemisphere; Fromkin, 1971) suggest a considerable degree of phonological advance planning because for the error to occur, the language production system must have ‘‘looked ahead’’, in the above cases across word boundaries. Recent experimental studies have also investigated the issue. Here, the central question is whether for an utterance of fixed length, speakers plan the entire utterance before articulation begins, or whether instead articulation starts as soon as the initial portion of the utterance has been retrieved (and the remainder is phonologically encoded while response execution is ongoing). To this aim, the phonological characteristics of a non-initial portion of the utterance are manipulated. If onset latencies to the entire utterance are affected, then advance planning presumably encompassed the manipulated portion. If not, then speakers likely planned only the utterance-initial portion before they started responding. For instance, Levelt and Wheeldon (1994) asked speakers to produce bisyllabic words that were varied according to both their overall word frequency and the frequency of their syllables. Not only overall frequency, but also the second-syllable frequency affected latencies, implying that the entire word, and not just its initial syllable, had been encoded prior to articulation (but see Levelt et al., 1999, p. 32, for more recent complications). Other findings similarly suggest that phonological planning comprises a minimum of one word. For instance, Meyer, Roelofs, and Levelt (2003) suggested that it takes more time to initiate responses to longer than to shorter words, a finding predicted under the assumption that speakers phonologically encode the entire word. Most likely, planning even exceeds the proposed single phonological word boundary. Using a prompt-response generation task, Roelofs (2002, Experiment 2) elicited utterances consisting of two phonological words (‘‘zoek op’’, ‘‘look up’’), and found faster response latencies for compatible (‘‘op’’) than incompatible particles. A task that has in recent years enjoyed widespread popularity in the investigation of spoken word produc-

tion is the picture–word interference (PWI) paradigm. Participants name common objects while attempting to ignore printed or spoken distractor words presented concurrently with the picture. Distractors are typically either semantically related (dog), phonologically related (cap), or unrelated (hip) to the picture (cat). Relative to the unrelated condition, a semantic relationship between word and picture slows down responses whereas a phonological one accelerates them (e.g., Damian & Martin, 1999; Glaser & Du¨ngelhoff, 1984; Lupker & Katz, 1981; Posnansky & Rayner, 1977; Rayner & Posnansky, 1978; Rosinski, Golinkoff, & Kukish, 1975; Schriefers, Meyer, & Levelt, 1990). The facilitation obtained with form-related distractors suggests that the distractor’s phonological representation contacted those of the target picture; because of the overlap between them, latencies will be accelerated compared to an unrelated condition. The effect can hence be used to investigate the properties of the spoken production system, and more specifically, the scope of phonological encoding. For instance, Meyer and Schriefers (1991) asked Dutch speakers to name objects with mono- or bisyllabic labels (e.g., ‘‘stoel’’, chair), and presented auditory distractors that shared either the initial (‘‘stoep’’, sidewalk) or the final segments (‘‘doel’’, goal) of the object name, or were unrelated. Facilitation was found for both types of overlap, suggesting that the entire word had been phonologically encoded before articulation began (see also Roelofs, 2002, Experiment 1). Concerning the production of more complex utterances, Jescheniak and Schriefers (2001) asked German speakers to produce either bare nouns, or noun phrases consisting of a determiner and a noun. Distractors were form-related to the noun, or unrelated. Substantial facilitation was obtained in bare noun production. For the noun phrases, facilitation was still present but considerably reduced. In slight contrast, Miozzo and Caramazza (1999) found similar facilitation for bare noun and determiner + noun Italian phrases. Costa and Caramazza (2002) asked English participants to name colored objects with determiner + adjective + noun phrases (e.g., ‘‘the green plane’’). Distractor words were phonologically related or unrelated to the noun. Facilitation was obtained, suggesting that speakers had encoded the phrase up to its final element before articulation was initiated. Schnur, Costa, and Caramazza (2006) demonstrated that when objects were described with sentences such as ‘‘The orange girl walks’’, visual distractors form-related to the verb (e.g., ‘‘walnut’’) speeded up responses relative to an unrelated condition. Jescheniak, Schriefers, and Hantsch (2003) obtained facilitation from related distractors when objects were simply named, and when they were named together with their determiner, but contrasting with previous findings, they found inhibition when participants named colored

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

objects and produced determiner + adjective + noun utterances.1 In combination, these findings suggest a substantial degree of phonological advance planning— even in utterances of considerable length, response initiation is affected by whether or not a distractor is related to the utterance-final element. This conclusion dovetails with the observation that, when colored objects are named with determiner + adjective + noun phrases (‘‘the blue kite’’), frequency effects of both adjective and noun are obtained on latencies (Alario et al., 2002a, 2002b). On the other hand, evidence incompatible with the claim that speakers plan ahead substantial portions of an utterance comes from a study by Schriefers and Teruel (1999) who asked German speakers to name colored objects with phrases of the form ‘‘rote Maus’’ (‘‘red mouse’’; adj + Noun). Auditory distractors consisted of either the first (‘‘ro-’’) or the second (‘‘-te’’) syllable of the adjective, or were phonologically related to the noun, or unrelated. In four experiments, no phonological facilitation was observed for the noun part of the utterance. Even more striking, reliable facilitation was restricted to the first syllable of the adjective. For the second syllable, only weak and non-significant facilitation was found that potentially reflected individual differences between speakers (see below). In view of these last results, speakers may initiate articulation on the basis of advance planning units smaller than a phonological word, namely the first syllable of the first response word. Advance planning and task demands As noted above, Schriefers and Teruel (1999) obtained only small and statistically unreliable facilitation effects for the second adjective syllable of adjective–noun utterances. They hypothesized that perhaps certain participants initiated their responses immediately upon retrieval of the initial sounds, whereas others encoded (at least) the entire adjective. The suggested individual differences in the degree of advance planning may emerge in the extent of erroneous utterances: those participants who provided a ‘‘fast’’ response should be more likely to produce restarts (‘‘li-uh-lila Sa¨ge’’; ‘‘puruh-purple saw’’) or utterances with filled or silent pauses (‘‘li-la Sa¨ge’’; ‘‘pur-ple saw’’) than those that planned the entire utterance. In a post hoc analysis, the authors subdivided speakers into two groups based on the 1 A similar inhibitory effect, albeit not in a PWI task, was demonstrated by Meyer (1996): participants named two simultaneously presented objects with either a coordinated noun phrase (‘‘the bag and the arrow’’) or a simple sentence (‘‘the bag is next to the arrow’’). Distractors that were phonologically related to the first noun yielded facilitation, but distractors related to the second noun yielded a small but significant interference effect.

197

amount of errors of this type, and indeed, only the group with less errors exhibited a sizeable priming effect. The authors suggested that the degree of phonological planning may vary between speakers. The suggestion that the extent of advance planning may not be uniform across speakers is intriguing. What is not immediately clear, however, is whether the findings reflect genuine individual differences, or whether planning is to some extent under speakers’ control and hence reflects individual adaptations to a particular situation. The latter possibility is intuitively appealing: depending on the primary goal of a particular situation (i.e., error-free vs. fast production), speakers may adopt a ‘‘conservative’’ or ‘‘liberal’’ criterion for initiating an utterance. The possibility that spoken production is subject to situational demands has also been considered elsewhere; after reviewing the evidence on advance planning, Meyer et al. (2003) conclude: ‘‘None of these studies shows that it is impossible for speakers to use planning units that are smaller than one phonological word. However, they demonstrate that speakers are not strongly inclined to use such units’’ (p. 144). Similarly, Bachoud-Levi, Dupoux, Cohen, and Mehler (1998) consider the possibility that ‘‘depending on task demands, articulation can be initiated using different levels of the hierarchy’’ (p. 343). However, so far the hypothesis that task demands could directly affect advance planning has rarely been put to the test. In a few recent studies, task demand in speaking has been manipulated, but from a different theoretical angle. In addition to measuring latencies, these studies have investigated responses’ acoustical durations. Durations provide a measure of how articulatory execution unfolds over time, and if an utterance is planned in its entirety before articulation is initiated, then response durations should not be affected by experimental variables such as relatedness in PWI. By contrast, if planning of an utterance is partially carried out while its initial portion is already being articulated, then durations could be affected by experimental manipulations. A number of studies (e.g., Schriefers, de Ruiter, & Steigerwald, 1999; Schriefers & Teruel, 1999) favor the former possibility: response durations of relatively short phrases are not affected by manipulations such as relatedness in PWI, providing additional evidence for a relatively large extent of advance planning prior to articulation. This conclusion has recently been challenged by Kello, Plaut, and MacWhinney (2000), however. In a Stroop color naming experiment, the typical congruency effects were observed in the latencies, but response durations were unaffected. Importantly, when task demand was increased using a response deadline, Stroop interference emerged in durations as well. The authors suggested that under most circumstances, articulation only begins after planning has been concluded; by contrast, under increased demands, the planning stage can

198

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

‘‘cascade’’ into response execution. This hypothesis implies that speakers typically plan an entire word prior to articulation, but when pressed for speed, they will base articulation on only the initial portion of the response. However, Damian (2003) failed to replicate the predicted duration effects in various tasks (PWI, semantic/phonological blocking, Stroop color naming), which warrants skepticism about the claim that the language production system can readily shift from a staged to a cascaded processing mode. The hypothesis that the extent of phonological advance planning may be influenced by task demands should nevertheless be put to further empirical tests. The experiments reported below combined the method of manipulating a non-initial utterance portion, previously used in for instance Costa and Caramazza (2002) and Meyer and Schriefers (1991), with the response deadline manipulation introduced by Kello et al. (2000). Based on the earlier findings, it was predicted that under normal task demands, manipulating the non-initial utterance portion should affect response latencies. The critical question was whether these effects would be reduced or eliminated when task demands were increased with the introduction of a response deadline. If so, then under these conditions, advance planning was likely to be considerably reduced, and ceased to extend to the manipulated portion. Hence, the scope of planning would be shown to be to some extent variable. On the other hand, similar effects obtained without and with a deadline would suggest that speakers always, and independent of task demands, planned ahead up to the manipulated portion. Experiment 1 investigated picture naming with bisyllabic names, and used distractors that were begin-or end-related to the picture label. Experiment 2 asked speakers to name colored pictures with determiner + adjective + noun utterances, and used distractors that were phonologically related to the noun. Experiment 3 elicited adjective + noun utterances from colored pictures in which the initial segment of the adjective and the noun matched or mismatched. In all three experiments, a group of participants operating under normal instructions, and one responding with a deadline, were compared. This enabled us to test whether the size of the phonological effect was reduced under increased task demands.

Experiment 1

remaining were tested with a response deadline. The deadline was set at 718 ms, the overall average of the ‘‘no deadline’’ group. Materials Twenty-six line drawings of common objects with bisyllabic names were selected as targets, with an average length of 4.6 phonemes, and an average frequency of 27 per million in the CELEX database of spoken English (Baayen, Piepenbrock, & Gulikers, 1995). For the begin-related condition, bisyllabic distractor words were selected that shared the initial syllable with the corresponding picture label. The CELEX frequency of the distractors was 15 per million; the length ranged between four and eight letters (mean = 5.8), and four and seven phonemes (mean = 5.1). The average overlap between target and distractor was 2.2 phonemes; The relative overlap was 48.3% of the target length in phonemes. For the end-related condition, distractors were selected that shared the final syllable with the target. The average frequency was 19 per million, and distractors ranged in length from five to eight letters (mean = 6.1), and from four to six phonemes (mean = 4.8). The average overlap between target and distractor was 2.4 phonemes; the relative overlap was 51.7% of the target length in phonemes. Begin- and end-related distractors were statistically matched in terms of overlap to the target, average frequency, and phoneme and letter length. For both types of overlap, objects and distractors were recombined to form the unrelated condition such that phonological overlap was minimal. Care was taken to avoid semantic or associative relationships. See Appendix A for a list of all stimulus combinations. On a further 52 filler trials, target objects were paired with phonologically unrelated distractor words different from the critical distractors, yielding a total of 156 trials per participant. Design The experimental design included response deadline (‘‘no deadline’’ vs. ‘‘deadline’’) as a between-participants variable, and type of overlap (begin- vs. end-related) and relatedness (related vs. unrelated) as within-participants variables. Each object was presented under each of the four conditions once, yielding a total of 104 critical trials per participant, plus 52 filler trials. Items were presented in a pseudorandom fashion such that neither the same target nor the same distractor appeared on subsequent trials.

Method Participants Thirty-six undergraduate students at the University of Bristol took part in exchange for course credit. Eighteen were tested without a response deadline first; the

Apparatus Stimuli were presented from an IBM-compatible computer on a 17 in. monitor using DMDX (Forster & Forster, 2003). The objects were digitized as line drawings to a size of approximately 8 · 8 cm, and were

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

199

In the ‘‘No Deadline’’ group, each individual trial was structured as follows: a fixation cross was presented for 500 ms. After a blank period of 500 ms, the target picture and the distractor word were shown for 500 ms. Latencies were measured from the onset of the target to the response. Following each naming response, the experimenter judged the responses to be either correct or incorrect; incorrect responses consisted of responses other than the expected ones, repairs, stuttering, or mouth clicks. An intertrial interval of 1500 ms concluded each trial. The same time structure applied to trials in the ‘‘Deadline’’ group, except that the deadline message was presented on any trial in which the calculated onset latency exceeded the overall mean response time obtained in the ‘‘No Deadline’’ group, which was tested first (718 ms). The message was presented 2000 ms after picture onset for 500 ms in the center of the screen, and coincided with the warning tone.

presented as black line drawings on light grey background. Distractor words were presented in black 18-point Arial bold font in the center of the screen. Objects and distractors were presented simultaneously (i.e., with a stimulus-onset asynchrony, or SOA, of 0 ms). A headset (Sennheiser mb40) with attached microphone was connected to the computer. The software determined the onset of the vocal responses to the nearest millisecond. Procedure Participants were tested individually. At the beginning of the experiment, they were instructed that their task would be to name objects as fast and accurately as possible. They were familiarized with the set of experimental pictures by viewing all objects on the screen in a miniaturized display, with the corresponding name printed below each image. In a first practice block, each object was presented and named once in random order; responses other than those expected were corrected. In a second practice block, the pictures were accompanied by unrelated distractor words; again, unexpected responses were corrected. Participants in the ‘‘Deadline’’ group were then additionally informed that if they initiated their response later than a particular time interval after the appearance of the target picture, the message ‘‘Too slow!’’ would be displayed, and a warning tone would be presented, indicating that they were not responding fast enough and should attempt to respond faster on subsequent trials, at the possible expense of accuracy. The warning tone consisted of a 440 Hz sine wave sound of 500 ms duration, presented at a moderate level via headphones. For these participants, an additional practice block was carried out in which they could familiarize themselves with this procedure. Subsequently, four experimental blocks of 39 trials each were carried out. Breaks were provided between the blocks. Each testing session lasted approximately 20 min.

Results Responses judged to be incorrect by the experimenter for the reasons described above were excluded from the response latency analysis (3.0%). Latencies faster than 250 ms or slower than 1500 ms were considered outliers and eliminated (1.9%). Table 1 presents the mean latencies and error percentages, varied by deadline, type of overlap, and relatedness. As can be seen, the deadline procedure led to an overall reduction in latencies. Beginand end-related facilitation of similar magnitude was found that was apparently not influenced by the deadline manipulation. Analyses of variance (ANOVAs) were conducted on both participants and items means. A main effect of deadline was obtained, F1(1, 34) = 8.99, MSE = 198,480, p = .005; F2(1, 25) = 319.32, MSE = 283,259, p = .001; min F 0 (1, 36) = 8.74, p = .006; Latencies were 74 ms, or 10.3%, faster in the ‘‘deadline’’ group than in the ‘‘no deadline’’ group. A main effect of relatedness (29 ms

Table 1 Experiment 1: Mean response latencies (RT, in ms) and mean error proportions (PE, in %), varied by deadline (No Deadline vs. Deadline), type of picture-distractor relation (Begin- vs. End-Related) and relatedness (Unrelated vs. Related) Begin-Related RT No Deadline Unrelated Related Effect

740 (86) 712 (113) +28

Deadline Unrelated Related Effect

664 (51) 634 (55) +30

Standard deviations in brackets.

End-Related PE

RT

PE

2.8 (3.2) 2.8 (3.7) 0.0

726 (97) 693 (90) +33

3.2 (4.6) 1.9 (2.7) +1.3

3.0 (2.8) 2.4 (3.5) +0.6

657 (55) 619 (49) +38

3.0 (3.6) 1.3 (2.6) +1.7

200

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

facilitation) was obtained, F1(1, 34) = 38.76, MSE = 36,795, p < .001; F2(1, 25) = 34.84, MSE = 56,642, p < .001; min F 0 (1, 57) = 18.35, p = .001. Also, some evidence for a main effect of type of overlap was found, F1(1, 34) = 10.67, MSE = 6994, p = .002; F2(1, 25) = 3.93, MSE = 8214, p = .059, min F 0 (1, 43) = 2.87, p = .097; with latencies being 14 ms slower in the begin-than in the end-related condition. None of the interactions were significant, all F1 6 0.74, ps P .396; F2 6 0.34, ps P .567; min F 0 6 0.23, p P .632. In sum, the phonological facilitation effect was comparable for begin- and end-related items, and it was not affected by the absence or presence of the response deadline. Corresponding ANOVAs conducted on the errors revealed no significant main effect or interaction, all F1 6 1.12, ps P .300; F2 6 1.26, ps P .272; min F 0 6 0.60, p P .445. Discussion The response deadline procedure increased task demand and substantially accelerated naming responses in the ‘‘deadline’’ group of participants when compared to the ‘‘no deadline’’ group, by more than 10%. Furthermore, in both the ‘‘no deadline’’ and the ‘‘deadline’’ group, comparable degrees of begin- and end-related facilitation were obtained. Assuming that end-related facilitation indicates that speakers phonologically encoded the entire word before they started speaking, these findings imply that even under increased pressure, advance planning could not be restricted to the initial portion of the utterance, but always comprised the entire response. To test further the possibility that speakers can in principle initiate their responses before they have phonologically encoded a whole word, it would be desirable to test even longer words (‘‘kangaroo’’, ‘‘alligator’’, etc.). Unfortunately, not only are pictures with such long names relatively rare in English, but it is also virtually impossible to find corresponding distractor words that overlap with the correct portion while being matched to each other on various variables such as length, etc.2 The following experiment instead turns to multiple-word responses. As outlined in Introduction, a few studies have reported phonological facilitation from distractors related to the noun in utterances such as determiner + noun, and determiner + adjective + noun phrases (e.g., Costa & Caramazza, 2002; Jescheniak & Schriefers, 2001; Miozzo & Caramazza, 1999). These findings have been taken to suggest that advance planning

2

A possible strategy would be to use fragments (‘‘ci-’’, ‘‘-ga-’’, ‘‘-rette’’), instead of words, as distractors. We conducted such an experiment with trisyllabic picture names and fragment distractors, found begin-, mid-, and end-related priming, and once again no reduction of these effects by a response deadline.

extended to the final portion of the utterance, and hence imply a planning scope larger than a single phonological word. However, it is important to assess whether such effects are still present under increased task demands. Experiment 2 used colored objects to elicit determiner + adjective + noun utterances (‘‘the green plane’’). Its first aim was to replicate the previous finding that distractors phonologically related to the noun (e.g., ‘‘plate’’) prime latencies. The next and more important goal was to investigate whether this effect was reduced or eliminated by the presence of a response deadline.

Experiment 2 Method Participants Twenty-four undergraduate students at the University of Bristol, none of whom had been in the first experiment, took part in exchange for course credit. Twelve were tested without a response deadline first; the remaining were tested with a response deadline. The deadline was set at 768 ms, the overall average of the ‘‘no deadline’’ group. Materials Thirty line drawings of common objects served as targets. They had mono- or bisyllabic names, with an average length of 4.1 phonemes, and an average spoken frequency of 37 per million in the CELEX database. These targets were presented with their lines colored in either blue, green, or red, and participants were instructed to name color and object with a determiner + adjective + noun phrase (‘‘the green clock’’). Colors were combined with targets such that word-initial segmental overlap (‘‘blue bottle’) was avoided. Care was also taken to avoid obvious associations such as ‘‘green grass’’. Phonologically related distractor words were selected for each picture that shared a minimum of the initial consonant cluster, and typically also the following vowel, with the object label. These had an average word length of 4.7 letters (4.0 phonemes), and a CELEX written frequency of 30 per million. The average overlap between target and distractor was 2.3 phonemes, or 59.6% of the target length in phonemes. Objects and distractors were then recombined to form the unrelated condition such that phonological overlap was minimal. Care was taken to avoid semantic or associative relationships. See Appendix B for a list of all stimuli. On a further 120 filler trials, target objects were paired with phonologically unrelated distractor words different from the critical distractors, yielding a total of 180 trials per participant.

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

Design The experimental design included response deadline (‘‘no deadline’’ vs. ‘‘deadline’’) as a between-participants variable, and relatedness (phonologically related vs. unrelated) as a within-participants variable. Each object was presented under each condition once, yielding a total of 60 critical trials per participant, plus 120 filler trials. Items were presented in a pseudorandom fashion such that neither the same target nor the same distractor appeared on subsequent trials. Apparatus and procedure The apparatus was identical to Experiment 1. The procedure was very similar. At the beginning of the experiment, participants were familiarized with the set of experimental pictures by viewing all objects on the screen in a miniaturized display, with the corresponding name printed below each image. In a first practice block, each object was then presented and named once in random order; responses other than the ones expected were corrected. In a second practice block, the pictures were presented in one of the three colors, and participants were instructed to name them with a determiner + adjective + noun utterance. In a third practice block, pictures were accompanied by unrelated distractor words (SOA = 0 ms), and participants were instructed to ignore the distractors. Participants in the ‘‘Deadline’’ group were then additionally informed about the deadline procedure (see Experiment 1). For this group, an additional practice block of 30 trials was carried out in which they could familiarize themselves with the procedure. Subsequently, four experimental blocks of 40 trials each were carried out. Breaks were provided between the blocks. Each testing session lasted approximately 25 min. All other procedural aspects were identical to those of Experiment 1. Results Responses judged to be incorrect by the experimenter were excluded from the response latency analysis (3.3%). Latencies faster than 250 ms or slower than 1500 ms were considered outliers and eliminated (1.3%). Table 2 presents the mean latencies and error percentages, varied by deadline and relatedness. As can be seen, the response deadline substantially reduced overall latencies. Yet, a numerically identical facilitation effect was found both without and with a response deadline. ANOVAs conducted on the latencies showed a significant effect of deadline, F1(1, 22) = 7.07, MSE = 148,655, p = .014; F2(1, 29) = 189.00, MSE = 352,049, p = .001; min F 0 (1, 24) = 6.82, p = .015. Latencies were 111 ms, or 14.5% faster in the ‘‘deadline’’ than in the ‘‘no deadline’’ group. A main effect of relatedness (28 ms) was obtained, F1(1, 22) = 42.74, MSE = 9223, p = .001; F2(1, 29) = 9.20, MSE = 22,747, p = .005;

201

Table 2 Experiment 2: Mean response latencies (RT, in ms) and mean error proportions (PE, in %), varied by deadline (No Deadline vs. Deadline) and picture-distractor relatedness (Phonologically Unrelated vs. Related) RT No Deadline Unrelated Related Effect

782 (131) 754 (131) +28

Deadline Unrelated Related Effect

671 (66) 643 (61) +28

PE 3.1 (4.1) 2.8 (4.0) +0.3 3.6 (3.6) 3.6 (3.3) 0.0

Standard deviations in brackets.

min F 0 (1, 40) = 7.57, p = .009. Importantly, no interaction between deadline and relatedness was found, F1(1, 22) < 0.01, p = .967; F2(1, 29) < 0.01, p = .986; min F 0 (1, 50) < .001, p = .944. Corresponding analyses on the errors showed no significant main effect or interaction, all F1 6 0.40, ps P .540; F2 6 0.52, ps P .475; min F 0 6 0.22, ps P .644. Discussion The results of the ‘‘no deadline’’ group nicely replicate earlier studies (e.g., Costa & Caramazza, 2002) in that in determiner + adjective + noun utterances, a distractor that is phonologically related to the noun yields a considerable facilitation effect. The novel finding is that even under increased time pressure constraints, the priming effect remained essentially unaffected. In combination, these findings suggest that phonological advance planning exceeds a single phonological word, and this relatively large planning scope is not ‘‘optional’’—even under a response deadline, participants appeared unable to restrict their planning to the utterance-initial adjective portion. As described in Introduction, the underlying assumption in the experiments reported so far, as well as previous studies such as Costa and Caramazza (2002), Meyer (1991), and Roelofs (2002) is that the phonological relatedness effect in PWI can inform us about the extent of advance planning in speaking. This assumption rests on specific claims about how phonological facilitation effects in PWI tasks come about. Most models of spoken production agree that following access to a word form, corresponding phonological segments are retrieved. The facilitatory effect in PWI could in principle emerge due to priming at either processing level, and possibly both. For instance, in the computational account of lexical access in spoken production introduced by Roelofs (1997), processing of a distractor will activate

202

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

corresponding phonemes, as well as its word form. If the word is phonologically related to the target, a facilitation effect may arise from one of two sources (or both): (a) the distractor may evoke a cohort of form-related items in the phonological output lexicon, among them the target word, (b) due to the shared segments between distractor and targets, phonological encoding of the correct target phonemes may be facilitated (see Roelofs, p. 265, for detailed computations). Importantly, the interpretation of results such as those reported here, as well as related studies such as Meyer and Schriefers (1991), rests on the assumption that it is largely priming at the segmental, but not at the word form, level which emerges in PWI. That is, it may be that the end-related distractor ‘‘glimmer’’ primed the target word form ‘‘hammer’’ directly, perhaps through a cohort in the output word form lexicon, as proposed by Roelofs (1997), but possibly also by means of cohort or neighborhood effects emerging during input processing of the distractor. In this case, end-related facilitation would imply that the target’s word form had been accessed when articulation began. However, it would not necessarily demonstrate that the subsequent stages of phonological encoding (retrieval of segments, access to motor patterns, etc.) had been completed at this point. The same criticism extends to the investigation of multiple-word utterances in this task, such as Experiment 2: if rather than priming phonological segments, the distractor primes the noun word form directly, then it is not clear whether phonological encoding really extended to the noun when articulation was initiated.3 Given the potential complications arising from the interpretation of PWI results, it would hence be advantageous to seek converging evidence from a different experimental task for the interim conclusion that phonological encoding comprises more than a single phonological word, even under considerable time pressure. The final experiment introduces a novel manipulation that abandons the use of distractors altogether. In this task, colored pictures were named, for which the initial phoneme of the color adjective and that of the picture matched (‘‘black book’’, ‘‘green gun’’) or mismatched (‘‘green book’’, ‘‘black gun’’). The basic idea was that if phonological advance planning comprised both the adjective and the noun portion (i.e., more than a single phonological word), then the repeated word-initial segment should affect utterance latencies.

3

A further possible scenario, discussed in Roelofs (2003), is that priming initially affects the segmental level, but that feedback links from the segmental to the word form level activate a set of candidates. In this way, end-related distractors could indirectly prime corresponding target word forms.

To date, only a few studies have explored the effects of repeated segments on latencies in spoken production (e.g., Sullivan & Riffel, 1999; Wheeldon, 2003). The emerging pattern is that a picture naming response, if preceded by a naming response with identical initial portion of the response (‘‘book’’, ‘‘bus’’), will exhibit slower latencies than when preceded by an unrelated word (the effect of repeated non-initial portion is more controversial). Note, however, that in this task, repeated segments occur between consecutive utterances, whereas in the experiment below, the repeated segments occur within the same phrase. Under these circumstances, it is likely that within-utterance priming, rather than inhibition, emerges, with faster latencies in the repeated than in the non-repeated condition. In either case, an effect of repeated segments would imply that speakers planned not merely the initial portion of the utterance, but the entire phrase before they initiated their response. Whether this is also the case in the presence of a response deadline which increases task demands remains to be seen. One further issue needs to be addressed. In the two experiments above, we demonstrated a substantial acceleration of overall response latencies by means of the deadline: In Experiment 1, latencies were 10.3% faster with than without a deadline, and in Experiment 2, the acceleration was 14.5%. This compares reasonably well with previous studies; in the three studies reported in Damian (2003), the acceleration was 7.1, 9.0, and 19.5%; in Kello et al.’s (2000) Stroop task, the acceleration was approximately 20%. Clearly it is the case that participants can increase their speed of responses to such an extent with relative ease. However, a curious aspect is that at present in none of the studies with the deadline procedure adopted from Kello et al., did error rates rise substantially. As far as we can tell, this pattern—substantial acceleration in response latencies, but only a minor corresponding rise in errors—reflects speakers’ general tendency to avoid errors. However, it could be argued that in order to demonstrate a genuine shift in response criterion, it is not sufficient to show accelerated response times, but additionally error rates should also increase, i.e., a speed–accuracy trade-off. On this argument, the deadline procedure in the form introduced by Kello et al. does not succeed. On the other hand, a number of speech production studies have explicitly attempted to elevate error rates with response deadlines (e.g., Vitkovitch & Humphreys, 1991; Vitkovitch, Humphreys, & Lloyd-Jones, 1993). Specifically, Starreveld and La Heij (1999) used a picture–word interference task in combination with a response deadline in order to investigate potential word substitution errors, and reported an overall error rate of 21.2%. This high error rate was accomplished mainly by explicitly instructing participants to respond as quickly as possibly, and encouraging them to make errors. These

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

results suggest that it is in principle possible to elevate error rates with a response deadline, but that the procedure introduced by Kello et al. and adapted in Damian (2003) and the experiments reported above is not well suited to this aim. For Experiment 3, we therefore modified our deadline procedure in the following way: (i) we accelerated the overall time structure of each trial, including the intertrial interval, (ii) we adopted a severe response deadline of 600 ms. In Kello et al.’s procedure, the overall mean of the ‘‘no deadline’’ group provides the limit for the response deadline. From pilot experiments, we know that speakers initiate a response to colored objects within roughly 800 ms. The 600 ms deadline that we adopted instead is obviously much more stringent, and we expected it to be relatively difficult for speakers to beat it, (iii) we changed instructions to encourage participants to respond as quickly as possible, and to tolerate errors. In accordance with Starreveld and La Heij (1999) participants were instructed that if they did not produce occasional errors, they were presumably not responding quickly enough. This modified procedure should substantially accelerate response latencies (likely to a larger degree than in the experiments reported above), but crucially it may now also induce a cost in accuracy. The central question remains whether the facilitation effect induced by the repeated word-initial phonemes is modified or eliminated by the deadline procedure.

Experiment 3 Method Participants Twenty-four undergraduate students at the University of Bristol, none of whom had been in the experiments above, took part in exchange for course credit. Twelve were tested without a response deadline first; the remaining were tested with a response deadline. The deadline was set at 600 ms. Materials Twenty line drawings of common objects with monosyllabic names were selected as targets, with an average length of 3.5 phonemes, and a spoken CELEX frequency of 18 per million. These targets were presented with their lines colored in either blue, green, pink, or red, and participants were instructed to name color and object with an adjective + noun phrase (‘‘blue bed’’). For the phonologically related condition, pictures and colors were paired such that the initial phoneme of adjective and noun (and in two cases, also the following vowel) coincided (e.g., ‘‘blue bed’’; the average overlap was 1.2 phonemes, or 34.8% of target length in phonemes). To form phonologically unrelated

203

adjective–noun combinations, the colors and nouns were recombined such that the initial phoneme differed. In both conditions, care was taken to avoid obvious associations such as ‘‘green grass’’. Appendix C shows a complete list of all combinations. A further 80 filler trials were generated on which each target picture was presented and named four times, paired with phonologically unrelated color adjectives. Design The experimental design included response deadline (‘‘no deadline’’ vs. ‘‘deadline’’) as a between-participants variable, and adjective–noun relatedness (related vs. unrelated) as a within-participants variable. Each object was presented under each of the two conditions once, yielding a total of 40 critical trials per participant, plus 80 filler trials. Items were presented in a pseudorandom fashion such that neither the same color adjective nor the same target appeared on subsequent trials. Apparatus and procedure Participants were first familiarized with the entire picture set, followed by a first practice block of 20 trials in which each object was named once. Then, the color manipulation was introduced, followed by a second practice block in which objects and colors were named with an adjective + noun phrase. Participants in the ‘‘Deadline’’ group were then additionally informed about the deadline procedure. For this group, an additional practice block of 20 trials was included in which they could familiarize themselves with the procedure. Subsequently, four experimental blocks of 30 trials each were carried out, with short breaks in between them. A new pseudorandom trial sequence was generated for each participant such that neither the same color nor the same object appeared on subsequent trials. Each testing session lasted approximately 20 min. Each individual trial was structured as follows: a fixation cross was presented for 500 ms, followed by the target display which remained on the screen for 1800 ms. Latencies were measured relative to the onset of the target. An intertrial interval of 500 ms concluded each trial. For the ‘‘Deadline’’ group, if responses did not occur within 600 ms, the message ‘‘Too slow!’’ was displayed after target offset, accompanied by a 440 Hz square wave warning tone of 200 ms presented at a moderate level via the headphones. Results The same criteria for data exclusion as in the other experiments were applied, resulting in the deletion of 5.1% of latencies due to errors, and 0.4% to outliers. Inspection of the data revealed that the item ‘‘broom’’

204

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

had a mean latency and error rate 2.5 standard deviations above the corresponding overall mean (latency: 804 ms, compared to 712 ms across all items; error rate: 12.0%, 5.1% for all items). Therefore the item was excluded from further analysis. An initial analysis, conducted on latencies both from experimental and filler items, demonstrated a clear speed–accuracy trade-off (i.e., an overall acceleration of response latencies through the deadline manipulation, accompanied by a rise in error rates). An ANOVA showed a significant difference in latencies, F1(1, 30) = 52.91, MSE = 468,827, p = .001, F2(1, 18) = 933.71, MSE = 553,359, p = .001; min F 0 (1, 33) = 50.07, p < .001, as well as in error rates, F1(1, 30) = 21.03, MSE = 236, p = .001; F2(1, 18) = 46.62, MSE = 280, p < .001; min F 0 (1, 47) = 14.49, p = .001. As opposed to the first two experiments, the modified deadline procedure hence produced the desired effect. Further analyses were restricted to the experimental items Table 3 presents mean latencies and error percentages, varied by deadline and relatedness. Phonologically related adjectives and nouns exhibited a substantial facilitation effect over unrelated combinations. The size of the priming effect was numerically larger in the ‘‘no deadline’’ than in the ‘‘deadline’’ group, but was still considerable in the latter case. ANOVAs conducted on the latencies showed a significant effect of deadline, F1(1, 30) = 47.96, MSE = 895,651, p < .001; F2(1, 18) = 644.95, MSE = 1,049,822, p = .001; min F 0 (1, 34) = 44.64, p = .001. Latencies were 237 ms (28.8%) faster in the ‘‘deadline’’ than in the ‘‘no deadline’’ group (as reported below, error rates were dramatically increased in the deadline condition, thereby implying a clear speed–accuracy trade-off on the test materials). A main effect of relatedness (47 ms priming) was obtained, F1(1, 30) = 33.27, MSE = 35,573, p = .001; F2(1, 18) = 17.79, MSE = 43,843, p =< .001; min F 0 (1, 36) = 11.59, p = .002. The interaction between deadline and relatedness was marginally significant, F1(1, 30) = 2.93, Table 3 Experiment 3: Mean response latencies (RT, in ms) and mean error proportions (PE, in %), varied by deadline (No Deadline vs. Deadline) and adjective–noun relation (Related vs. Unrelated) RT

PE

No Deadline Unrelated Related Effect

853 (137) 792 (121) +61

3.0 (3.0) 0.3 (1.8) +2.7

Deadline Unrelated Related Effect

603 (60) 570 (55) +33

12.5 (7.9) 5.6 (5.2) +6.9

Standard deviatons in brackets.

p = .097; F2(1, 18) = 3.15, p = .093, although not in the combined analysis, min F 0 (1, 46) = 1.52, p = .224. Note that overall responses in the ‘‘deadline’’ condition were by almost a third faster than in the ‘‘no deadline’’ condition; typically, reduced latencies entail compressed priming effects. The interpretation of group x treatment interactions under conditions where substantial group differences exist is known to be problematic (e.g., Faust, Balota, Spieler, & Ferraro, 1999). A straightforward way of dealing with this issue, often used in research on effects of ageing (e.g., Spieler, Balota, & Faust, 1996), is to perform proportion transformations on the treatment effects, which filters out group differences. To this aim, we divided each observed conditional mean by the average latency for that participant, or item. An ANOVA performed on these transformed scores (‘‘no deadline’’: related = 0.96, unrelated = 1.04; ‘‘deadline’’: related = 0.97, unrelated = 1.03) showed a significant effect of relatedness, F1(1, 30) = 37.28, MSE = 0.066, p < .001; F2(1, 18) = 17.04, MSE = 0.084, p = .001; min F 0 (1, 34) = 11.69, p = .002, but crucially no interaction between deadline and relatedness, F1(1, 30) < 1, p = .414; F2(1, 18) < 1, p = .478; min F 0 (1, 45) < 1, p = .944. Hence, when overall response speed differences between the two groups were taken into account, there was no evidence that the size of the relatedness effect, and hence the scope of advance planning, was affected by the deadline manipulation. An alternative method of corroborating this point is to compute, for each item, correlations between the relatedness effect under the ‘‘no deadline’’ and the ‘‘deadline’’ condition. If similar processing mechanisms underlie the facilitation effects in both groups, the correlation should be substantial; by contrast, if the presence of a deadline induces a change in planning strategies, then the effects obtained in the ‘‘no deadline’’ and the ‘‘deadline’’ condition should not correlate. We observed a correlation of .553 (p = .011), suggesting that the very same processing mechanisms supported priming in both groups. Similar analyses conducted on the errors showed a highly significant effect of deadline, F1(1, 30) = 20.26, MSE = 876, p = .001; F2(1, 18) = 47.46, MSE = 1041, p = .001; min F 0 (1, 47) = 14.20, p = .001; participants made 7.4% more errors in the ‘‘deadline’’ than in the ‘‘no deadline’’ condition. A main effect of relatedness was obtained, F1(1, 30) = 23.58, MSE = 364, p = .001; F2(1, 18) = 10.12, MSE = 432, p = .005; min F 0 (1, 36) = 8.74, p = .006, with 4.7% less errors in the related than in the unrelated condition. The interaction between deadline and relatedness was significant in the analysis by participants, F1(1, 30) = 4.74, MSE = 73, p = .038, but not by items, F2(1, 18) = 1.78, p = .199, nor in the combined analysis, min F 0 (1, 31) = 1.29, p = .264. The findings hence parallel those in the latencies such that

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

the facilitatory effect was essentially unaffected by increased task demands. Discussion In this experiment we observed a facilitatory effect of repeated word-initial segment in the production of adjective + noun phrases, both in response latencies and in error rates. This suggests that speakers phonologically encoded the entire utterance, and not merely its initial portion. The response deadline induced a speed–accuracy trade-off which had not been found in the studies using the deadline procedure pioneered by Kello et al. (2000). Crucially, the facilitatory effect of phoneme repetition was statistically unaffected by the presence of a response deadline. Even under conditions of dramatically heightened time pressure, it seems that phonological encoding is not reduced to the minimal planning unit of a phonological word, as advocated by, for instance, Levelt et al. (1999) and Wheeldon and Lahiri (1997). The observation of a facilitation effect for related adjectives and nouns poses an intriguing contrast to previous studies (e.g., Sullivan & Riffel, 1999; Wheeldon, 2003) that had shown a trial-to-trial inhibition effect of repeated word-initial portion. The discrepancy suggests different mechanisms for the two types of effects. The between-trial inhibitory effect is at present not entirely understood (see Wheeldon, 2003; for a detailed discussion). One possibility is that individual repeated phonemes may be in competition with each other and are hence more difficult to retrieve than non-repeated ones. This claim fails to account for Wheeldon’s finding that (i) inhibition is found with begin-overlapping word pairs, but not when words overlap in their end portion, (ii) even with begin-overlapping word pairs, inhibition is only found if there is a mismatching portion (bloed-bloem), but not when this is not the case (paar-paard). The account championed by Wheeldon is a modified version of the phoneme competition model introduced by Sevald and Dell (1994) in which phonological activation from the prime trial miscues processing on the target trial. Even this explanation remains speculative at present, however. On the other hand, if the repeated phonemes occur within the same utterance, as in the experiment above, then concurrent planning may have a facilitatory effect—a phoneme may prime itself. Again, the issue is not well understood, not the least because repeated token instantiation of one and the same type, in this case a phoneme, poses considerable computational difficulties (see Dell & O’Seaghdha, 1994; for a discussion). Experiments are currently underway to explore this phenomenon further. The exact underlying

205

mechanism is less relevant, however, for the main point derived from Experiment 3: the phonological properties of the second word influence response latencies, implying that advance planning encompassed the entire phrase. Crucially, this was still the case under increased task demands. The results hence converge with those from Experiment 2 in implying a relative large and incompressible scope of advance planning.

General discussion The three reported experiments set out to investigate the extent of phonological advance planning in spoken production, and specifically the possibility that the scope of planning may be reduced under increased task demands. In all studies we found that the facilitation induced by phonological overlap remained statistically unchanged whether or not a response deadline was used. These results dovetail with the conclusions derived from recent findings reported by Damian (2003): contrary to what is proposed by ‘‘minimalist’’ accounts of advance planning, the data give no reason to believe that speakers are ever able to initiate a response before they have phonologically encoded at least a single word. Instead, our findings suggest that advance planning encompasses at least two phonological words, even under time pressure. The obtained pattern contradicts ‘‘minimalist’’ accounts of advance planning, according to which articulation starts as soon as the initial portion of the utterance becomes available. But it is also at odds with the ‘‘phonological word hypothesis’’ endorsed by, for instance, Wheeldon and Lahiri (1997) and Levelt et al. (1999) according to which the lower boundary (and the preferred unit) of advance planning is a single phonological word. As outlined in Introduction, previous findings (e.g., Alario et al., 2002a, 2002b; Costa & Caramazza, 2002; Jescheniak et al., 2003; Schnur et al., 2006) have already suggested that this account underestimates the true amount of planning that speakers engage in. Our findings support this conclusion, and additionally show that even under intense task demand, speakers plan the entire utterance which in this case comprised two phonological words. It remains for future research to determine whether this is the case even for larger utterances, such as ‘‘the orange girl walks’’ investigated in Schnur et al. (2006). The reported studies suggest that advance planning always exceeds a single phonological word, but they do not illuminate exactly which variables constrain planning in the spoken production system. Indeed, at this point the mechanisms underlying advance planning, as well as the critical planning

206

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

units, remain largely hypothetical. The phonological word hypothesis of advance planning (e.g., Wheeldon, 2000) was designed with a perspective on prosodic properties of the utterance. For instance, according to Selkirk’s (1981) theory of prosody, an utterance is divided into intonational phrases, which themselves subdivide into phonological phrases, and finally into phonological words. The latter ones, according to the phonological word hypothesis, are the preferred unit of advance planning. Alternatively, it could be that the advance planning mechanism ‘‘looks ahead’’ by a number of morpho-syntactic words, possibly in a graded fashion. A model along these lines was recently proposed by Jescheniak et al. (2003): here, the phonological codes of successive words in a multi-word utterance receive some degree of activation, with the amount linearly decreasing as a monotonic function of the word’s position in the utterance. For instance, in the utterance ‘‘the green table’’, at the moment at which articulation starts, the determiner will have received the most activation, the adjective relatively less, and the noun the least amount of activation. In this framework, no explicit reference is made to prosodic properties of the utterance; the degree of advance planning is mainly constrained by the linear distance between an element’s position within an utterance, and the beginning of the utterance. Wheeldon and Lahiri (1997, 2002) recently presented evidence for the prosodic, and against the morpho-syntactic, account of speech planning. They asked participants to prepare utterances that varied in length, either according to morphosynactic units, or to the number of phonological words. Latencies to initiate the response was shown to be function of the latter, and specifically, of planning of phrasal-level phonological words that are blind to word-internal structure. Future research using non-prepared production tasks is required to corroborate these findings. Given that all three experiments manipulated phonological relatedness, but did not include a condition which could potentially slow down responses (such as semantically related distractors in the PWI task), could some or all of the effects demonstrated in this article be due to participants using the relatedness manipulations to their advantage? This is unlikely, based on the fact that Meyer and Schriefers (1991) manipulated the proportion of form-related items in a PWI task, and found that the size of the phonological facilitation effect was unaffected. In the coloredobject naming task introduced in Experiment 3, the overall relatedness proportion was relatively low (16.7%), and latencies in the ‘‘deadline’’ group were very fast (at or below 600 ms). Additionally, although post-experimental interviews revealed that participants were aware of the phonological relation between

adjective and noun, all of them stated that they had not tried to use this information strategically. Hence, we think it is unlikely that participants could have exploited conscious strategies, such as predicting the name of the object based on the color name, which could compromise the validity of our conclusions. One earlier finding that continues to evade explanation is Schriefers and Teruel’s (1999) failure to find significant facilitation for the second syllable of an adjective in an adjective–noun phrase, such as ‘‘rote Maus’’ (red mouse). This finding is certainly anomalous in that a number of other, similar studies did find facilitation on the noun portion, which logically would appear to imply priming on the adjective as well. As described in Introduction, the authors obtained some tentative evidence for individual differences in the degree of advance planning: participants who produced relatively few errors tended to show second-syllable facilitation, whereas those with more errors showed no such priming. The results of Experiments 2 and 3 do not suggest a similar division of participants into those with a larger or smaller extent of planning. They were conducted with relatively few participants (N = 24 in each study) yet the factor relatedness was highly significant, suggesting a high degree of consistency of the effect across participants and hence a similar degree of advance planning. Further research will have to attempt to replicate the original finding, and delineate under what circumstances the described pattern occurs. At present, the available evidence converges on the conclusion that the degree of advance planning is consistently larger than a single phonological word, even under pressure. In the reported experiment, along with the previous studies that had pursued a similar approach, speakers produced very short utterances with a fixed format. An issue that arises in this context pertains to the generalizability of the conclusions derived from such experiments to speech occurring in a natural context: Is it possible that speakers use a planning scope different from one used in natural speech? For instance, because speakers can largely ignore several other aspects of planning that are normally required (e.g., pragmatic, conceptual, syntactic), they may be able to initiate their responses at a relatively earlier preparation stage than they would be able to in a comparable ‘‘natural’’ setting. Conversely it may be argued that reduced planning demands at higher processing levels bring about a larger degree of advance planning at the phonological level. At present it is impossible to dismiss these possibilities, mainly because empirical evidence from alternative, perhaps more naturalistic, tasks, is yet to be brought forward.

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

Appendix A Stimuli used in Experiment 1 Target barrel beaver beetle bottle camel candle chisel cradle dagger hammer kettle letter lorry monkey pillow rabbit rattle rocket ruler saddle sandal spider table tiger turkey window

Begin-related Unrelated distractor distractor basil beacon beeper bottom cannon canvas chicken crater daddy habit kennel lettuce locker monday pistol racket raven roster ruby salad sandwich spiral tapir title turban winter

turban pistol canvas crater bottom salad raven beacon monday title daddy habit cannon kennel basil winter spiral sandwich chicken ruby roster lettuce locker racket beeper tapir

End-related distractor squirrel driver mantle nettle pommel bundle weasel handle trigger glimmer battle butter berry jockey fellow ambit shuttle bracket butler fiddle pedal leader bubble finger hockey shadow

Unrelated distractor shadow fiddle trigger leader butter bracket butler ambit jockey squirrel hockey shuttle bundle bubble berry pedal title handle weasel fellow driver battle finger nettle pommel glimmer

Appendix B Stimuli used in Experiment 2 Color

Target

Related distractor

Unrelated distractor

blue blue blue blue blue blue blue blue blue blue green green green green green green green

carrot duck guitar horse monkey rabbit saw spoon table tiger clock flute hand hat lamp leg sheep

caviar dust guilt horn monk rack salt spider tail tile clown floor hammer ham latch lemon shield

shield hammer lemon treat snort monk carp clown guilt frost doll caviar salt pill form rack bend

207

Color

Target

Related distractor

Unrelated distractor

green green green red red red red red red red red red red

snake sock sun bed bottle camel cannon car dolphin fork frog pig tree

snort sob supper bend boss camera candle carp doll form frost pill treat

tail latch tile horn supper sob floor dust camera candle ham boss spider

Appendix C Stimuli used in Experiment 3 Phonologically related color blue blue blue blue blue green green green green green pink pink pink pink pink red red red red red

Phonologically unrelated color green green pink red red blue blue pink red pink blue green green red red blue green pink blue pink

Target ball bed boot broom bus glass globe glove goat gun pen pill pin pipe purse rake rat ring rope rug

References Alario, F.-X., Costa, A., & Caramazza, A. (2002a). Frequency effects in noun phrase production: implications for models of lexical access. Language and Cognitive Processes, 17, 299–319. Alario, F.-X., Costa, A., & Caramazza, A. (2002b). Hedging one’s bets too much? A reply to Levelt (2002). Language and Cognitive Processes, 17, 673–682. Bachoud-Levi, A.-C., Dupoux, E., Cohen, L., & Mehler, J. (1998). Where is the length effect? A cross-linguistic study of

208

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209

speech production. Journal of Memory and Language, 39, 331–346. Costa, A., & Caramazza, A. (2002). The production of noun phrases in English and Spanish: implications for the scope of phonological encoding in speech production. Journal of Memory and Language, 46, 178–198. Damian, M., & Martin, R. C. (1999). Semantic and phonological codes interact in single word production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25, 345–361. Damian, M. F. (2003). Articulatory duration in single word speech production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 416–431. Dell, G. S., Juliano, C., & Govindjee, A. (1993). Structure and content in language production: a theory of frame constraints in phonological speech errors. Cognitive Science, 17, 149–195. Dell, G. S., & O’Seaghdha, P. G. (1994). Inhibition in interactive activation models of linguistic selection and sequencing. In D. Dagenbach & T. H. Carr (Eds.), Inhibitory processes in attention, memory, and language (pp. 409–453). New York: Academic Press. Faust, M. E., Balota, D. A., Spieler, D. H., & Ferraro, F. R. (1999). Individual differences in information-processing rate and amount: implications for group differences in response latency. Psychological Bulletin, 125, 777–799. Ferreira, F. (1991). Effects of length and syntactic complexity on initiation times for prepared utterances. Journal of Memory and Language, 30, 210–233. Ford, M., & Holmes, V. M. (1978). Planning units and syntax in sentence production. Cognition, 35–53. Forster, K. I., & Forster, J. C. (2003). DMDX: A Windows display program with millisecond accuracy. Behavior Research Methods, Instruments, and Computers, 35, 116–124. Fromkin, V. A. (1971). The nonanomalous nature of anomalous utterances. Language, 47, 27–52. Fromkin, V. A. (1973). Introduction. In V. A. Fromkin (Ed.), Speech errors as linguistic evidence (pp. 11–45). The Hague: Mouton. Glaser, W. R., & Du¨ngelhoff, F. J. (1984). The time course of picture–word interference. Journal of Experimental Psychology: Human Perception and Performance, 10, 640–654. Jescheniak, D., Schriefers, H., & Hantsch, A. (2003). Utterance format affects phonological priming in the picture–word task: implications for models of phonological encoding in speech production. Journal of Experimental Psychology: Human Perception and Performance, 29, 441–454. Jescheniak, J. D., & Schriefers, H. S. (2001). Priming effects from phonologically related distractors in picture–word interference. The Quarterly Journal of Experimental Psychology, 34A, 371–382. Jordan, M. I. (1990). Motor learning and the degrees of freedom problem. In M. Jeannerod (Ed.), Attention and performance, XIII (pp. 796–836). Hillsdale, NJ: Lawrence Erlbaum. Kello, C. T., Plaut, D. C., & MacWhinney, B. (2000). The taskdependence of staged vs. cascaded processing: an empirical and computational study of Stroop interference in speech production. Journal of Experimental Psychology: General, 129, 340–361.

Levelt, W. J. M. (2002). Picture naming and word frequency: comments on Alario, Costa and Caramazza, Language and Cognitive Processes, 17, 299–319. Language and Cognitive Processes, 17, 663–671. Levelt, W. J. M., & Maassen, B. (1981). Lexical search and order of mention in sentence production. In W. Klein & W. J. M. Levelt (Eds.), Crossing the boundaries in linguistics (pp. 221–252). Dordrecht: D. Reidel. Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22, 1–75. Levelt, W. J. M., & Wheeldon, L. (1994). Do speakers have access to a mental syllabary? Cognition, 50, 122–142. Lupker, S. J., & Katz, A. N. (1981). Input, decision, and response factors in picture–word interference. Journal of Experimental Psychology: Human Learning and Memory, 7, 269–282. MacKay, D. G. (1987). The organization of perception and action: A theory for language and other cognitive skills. New York: Springer. Meyer, A. S. (1991). The time course of phonological encoding in language production: phonological encoding inside a syllable. Journal of Memory and Language, 30, 69–89. Meyer, A. S. (1996). Lexical access in phrase and sentence production: results from picture–word interference experiments. Journal of Memory and Language, 35, 477–496. Meyer, A. S., Roelofs, A., & Levelt, W. J. M. (2003). Word length effects in object naming: the role of a response criterion. Journal of Memory and Language, 48, 131–147. Meyer, A. S., & Schriefers, H. (1991). Phonological facilitation in picture–word interference experiments: effects of stimulus onset asynchrony and types of interfering stimuli. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 1146–1160. Miozzo, M., & Caramazza, A. (1999). The selection of determiners in noun phrase production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25, 907–922. Posnansky, C. J., & Rayner, K. (1977). Visual-feature and response components in a picture–word interference task with beginning and skilled readers. Journal of Experimental Child Psychology, 24, 440–449. Rayner, K., & Posnansky, C. (1978). Stages of processing in word identification. Journal of Experimental Psychology: General, 107, 64–80. Roelofs, A. (1997). The WEAVER model of word-form encoding in speech production. Cognition, 64, 249–284. Roelofs, A. (2002). Spoken language planning and the initiation of articulation. The Quarterly Journal of Experimental Psychology, 55A, 465–483. Roelofs, A. (2003). Goal-referenced selection of verbal action: modeling attentional control in the Stroop task. Psychological Review, 110, 88–125. Rosinski, R. R., Golinkoff, R. M., & Kukish, K. S. (1975). Automatic semantic processing in a picture–word interference task. Child Development, 46, 247–253. Schnur, T. T., Costa, A., & Caramazza, A. (2006). Planning at the phonological level during sentence production. Journal of Psycholinguistic Research, 35, 189–213. Schriefers, H., de Ruiter, J., & Steigerwald, M. (1999). Parallelism in the production of noun phrases: experiments

M.F. Damian, N. Dumay / Journal of Memory and Language 57 (2007) 195–209 and reaction time models. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25, 702–720. Schriefers, H., Meyer, A. S., & Levelt, W. J. M. (1990). Exploring the time course of lexical access in language production: picture–word interference studies. Journal of Memory and Language, 29, 86–102. Schriefers, H., & Teruel, E. (1999). Phonological facilitation in the production of two-word utterances. European Journal of Cognitive Psychology, 11, 17–50. Selkirk, E. O. (1981). On prosodic structure and its relation to syntactic structure. In T. Fretheim (Ed.), Nordic prosody II (pp. 111–140). Trondheim, Norway: Tapir. Sevald, C. A., & Dell, G. S. (1994). The sequential cuing effect in speech production. Cognition, 53, 91–127. Spieler, D. H., Balota, D. A., & Faust, M. E. (1996). Stroop performance in healthy younger and older adults and in individuals with dementia of the Alzheimer’s type. Journal of Experimental Psychology: Human Perception and Performance, 22, 461–479. Starreveld, P. A., & La Heij, W. (1999). Word substitution errors in a speeded picture–word task. American Journal of Psychology, 112, 521–553.

209

Sullivan, M. P., & Riffel, B. (1999). The nature of phonological encoding during spoken word retrieval. Language and Cognitive Processes, 14, 15–45. Vitkovitch, M., & Humphreys, G. W. (1991). Perseverant responding in speeded naming of pictures: It’s in the links. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 664–680. Vitkovitch, M., Humphreys, G. W., & Lloyd-Jones, T. J. (1993). On naming a giraffe a zebra: Picture naming errors across different object categories. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 243–259. Wheeldon, L. (2000). Generating prosodic structure. In L. R. Wheeldon (Ed.), Aspects of language production (pp. 249–274). Hove: Psychology Press. Wheeldon, L. (2003). Inhibitory form priming of spoken word production. Language and Cognitive Processes, 18, 81–109. Wheeldon, L., & Lahiri, A. (1997). Prosodic units in speech production. Journal of Memory and Language, 37, 356–381. Wheeldon, L. R., & Lahiri, A. (2002). The minimal unit of phonological encoding: prosodic or lexical word. Cognition, 85, B31–B41.