1 University of Liege
2 Fund of Scientific Research Belgium (F.R.S.-FNRS)
Address for correspondence:
Steve Majerus, PhD
Center for Cognitive and Behavioral Neuroscience
Department of Cognitive Sciences, University of Liege
Boulevard du Rectorat, B33
4000 Liège, Belgium
tel: 0032 43664656
fax: 0032 43662808
Despite extensive research, the role of phonological short-term memory (STM) during oral sentence comprehension remains unclear. We tested the hypothesis that phonological STM is involved in phonological analysis stages of the incoming words, but not in sentence comprehension per se. We compared phonological STM capacity and processing times for natural sentences and sentences containing phonetically ambiguous words. The sentences were presented for an auditory sentence anomaly judgment task and processing times for each word were measured. STM was measured via nonword and word immediate serial recall tasks, indexing phonological and lexico-semantic STM capacity, respectively. Significantly increased processing times were observed for phonetically ambiguous words, relative to natural stimuli in same sentence positions. Phonological STM capacity correlated with the size of this phonetic ambiguity effect. However, phonological STM capacity did not correlate with measures of later semantic integration processes while lexico-semantic STM did. This study suggests that phonological STM is associated with phonological analysis processes during sentence processing.
Despite a vast amount of experimental and neuropsychological studies, the relation between oral sentence processing and verbal short-term memory (STM) continues to puzzle researchers. Intuitively, it seems obvious that some form of temporary verbal storage capacity is necessary for maintaining and integrating the sound forms, meanings and syntactic functions of the different words a sentence is made of. Yet, in the light of past scientific studies, the precise role of verbal STM during sentence comprehension remains elusive. In this study, we explore the hypothesis following which the involvement of phonological STM would be limited to phonological analysis stages during sentence comprehension. In other words, we may need phonological STM to analyze the sound patterns of speech occurring in a sentence context, but not to process sentence meaning per se.
Early neuropsychological studies in brain injured patients with impaired verbal STM capacity suggested that poor verbal STM capacity is necessary for maintaining the representations of function words and inflections, as well as for maintaining the order of words, especially in long sentences (Caramazza, Basili, Koller, & Berndt, 1981; Vallar & Baddeley, 1984). However, later neuropsychological studies and studies in healthy adults, relating inter-individual differences in verbal STM capacity to different measures of sentence comprehension ability, failed to observe a consistent link between performance on verbal STM tasks (e.g., digit span, word span) and various measures of sentence comprehension (Butterworth, Campbell, & Howard, 1986; Caplan & Waters, 1999; Just & Carpenter, 1992; McElree, Foraker, & Dyer, 2003; Waters, & Caplan, 2004; Waters, Caplan, & Hildebrandt, 1991). Furthermore, more recent neuropsychological studies have shown dissociations between two types of verbal STM capacity: some brain injured patients display preserved phonological STM (for storing speech sounds) but impaired semantic STM (for storing the meanings), while other patients present a reverse profile (Freedman & R. Martin, 2001; Majerus, Van der Linden, Poncelet, & Metz-Lutz, 2004; R. Martin & Romani, 1994). When investigating sentence processing in these patients, researchers observed that patients with poor phonological STM capacity (as evidenced by poor performance on nonword immediate serial recall tasks or rhyme probe span tasks) showed impaired performance for verbatim sentence repetition (which requires the precise maintenance of the phonological codes of the words), while sentence comprehension was less impaired (R. Martin, 2006; R. Martin & He, 2004; R. Martin, Lesch, & Bartha, 1999; R. Martin, Shelton, & Yaffee, 1994). On the other hand, patients with poor semantic STM capacity (as evidenced by poor performance on tasks probing the retention of semantic information such as category probe span tasks or word immediate serial recall) showed a reverse profile, with impairments in sentence comprehension, especially when the semantic integration between two or more different sentence constituents was differed, while sentence repetition was preserved (R. Martin et al., 1994).
Lastly, a third STM capacity, syntactic STM, has been proposed to be specifically involved in syntactic analysis and integration processes during sentence comprehension (Caplan & Waters, 1999). This is based on studies showing that classic verbal STM tasks such as word span and digit span, but also more complex verbal working memory tasks (requiring both storage and processing of the information held in STM), are generally very poor predictors of online sentence comprehension processes while the integration of syntactic and semantic information nevertheless needs some form of short-term storage (Caplan & Waters, 1999; Waters & Caplan, 2004). However, according to these studies, verbal working memory capacity is involved in so-called off-line sentence comprehension processes when, at the end of a sentence, the product of online syntactic interpretation is consciously reviewed to perform a subsequent task, such as sentence anomaly decision (Waters & Caplan, 2004).
In sum, there is very little evidence for an involvement of phonological STM capacity in sentence comprehension. Yet, sentence processing requires sound-based processing and short-term maintenance of incoming information, the time initial identification of the incoming words is completed and semantic meaning is accessed. One explanation for the absence of a consistent correlation between phonological STM capacity and sentence comprehension may be that in typical listening conditions, semantic access is very fast (this is in the 300-400msec range or less; see for example Kotz, Von Cramon, & Friederici, 2005; Marslen-Wilson & Welsh, 1978), and hence phonological storage requirements will be minimal. Only patients with very severe phonological STM limitations may thus experience difficulties in sentence comprehension. This is supported by studies showing that patients with very severe phonological STM impairments (digit spans of two or less) display sentence comprehension difficulties; it should be noted that in these patients, the STM deficit is so severe that even single word identification and repetition are impaired (Majerus, Lekeu, Van der Linden, & Salmon, 2001; N. Martin & Saffran, 1992).
Similarly, if access to semantic knowledge is delayed, either due to slowed phonological processing (e.g., if the acoustic input is ambiguous or noisy) or to slowed semantic processing, then the role of phonological STM maintenance in sentence comprehension may become more predominant. Indirect evidence for this hypothesis can be derived from studies in elderly people: their speech identification abilities are often suboptimal (due to reduced auditory processing abilities) and semantic access is slowed (Pfutze, Sommer, & Schweinberger, 2002; Roberts & Lister, 2004; Sommers, 2005). Some studies observed a tendency towards a stronger association between measures of verbal short-term / working memory and sentence comprehension in elderly people than in younger participants (DeDe, Caplan, Kemtes, & Waters, 2004; Van der Linden, Hupet, Feyereisen, Schelstraete, Bestgen, Bruyer et al., 2001). Somewhat more direct evidence stems from recent neuropsychological case studies showing that patients with phonological STM impairments have difficulty in making phonological judgments for words presented in a sentence context, but not for the same words when presented in isolation, suggesting that phonological STM may be involved in phonological processing and maintenance during sentence processing (Jacquemot, Dupoux, Decouche, & Bachoud-Levi, 2006).
In the present study, we investigate the hypothesis that phonological STM is involved in sentence processing, but that its role in sentence processing is restricted to the stage of phonological analysis for individual constituent words, by enabling the listener to maintain the perceived sequence of sounds the time necessary to access its meaning and sentence context function. In order to test this hypothesis, we compared healthy adults’ performances for processing sentences including target words that were either natural or phonetically ambiguous. We expected the phonetically ambiguous words to lead to slower identification times relative to the natural variants of the target words, both variants being embedded in the same sentence types and same sentence positions. If phonological STM supports phonological analysis stages during sentence processing, then it should be specifically associated with the increase of processing times for the phonetically ambiguous words, but there should be no association with processing times for other, phonetically non-ambiguous sentence positions or for end of sentence processing times. We further distinguished online and offline sentence processes, and did so by using the auditory moving technique (Ferreira & Anes, 1994; Ferreira, Henderson, Weeks, & McFarlane, 1996). This technique is an auditory adaptation of the widely used self-paced reading tasks. In this technique, an auditory recording of a sentence is segmented in different parts, and the participants must activate a response key in order to move from one segment of the sentence to the next. The time the participants take to activate the response key after hearing a word is supposed to reflect online processing times for the given word in a sentence context (Ferreira & Anes, 1994). By using this technique, we were able to obtain online processing times for natural words and for phonetically ambiguous words occurring in a sentence context. Offline sentence processing is reflected by the time taken to make a sentence decision at the end of the sentence, when all sentence information has been presented (Waters & Caplan, 2004).
With respect to the specific directionality of the expected association between phonological STM capacity, as measured by a nonword immediate serial recall task in the present study, and increased processing times for phonetically ambiguous words, different outcomes are possible. (1) A positive relation between phonological STM capacity and increased processing times for online measures of phonetically ambiguous words, meaning the higher performances on the phonological STM task, the more time spent on phonetically ambiguous words. Two interpretations are possible here: either participants with higher phonological STM capacity are able to hold phonetically ambiguous words for a longer time, in order to allow for more time for phonological analysis, or participants with higher phonological STM capacity have more detailed phonological processing abilities in general, and hence will be more proficient in processing the nonwords used for the phonological STM task as well as in detecting subtle phonetic irregularities; the latter interpretation is based on a number of studies showing that nonword processing in STM tasks is highly dependent upon the quality and level of segmentation of phonological representations in the language system (e.g., Gathercole, Frankish, Pickering & Peaker, 1999; Majerus, Van der Linden, Mulder, Meulemans & Peters, 2004; Metsala, 1999). In order to be able to distinguish between these two alternative interpretations, we included a second STM task having less extensive STM load and blocking phonological loop related STM maintenance processes such as articulatory rehearsal while having the same phonological processing requirements. If the association observed between increased processing times for phonetically ambiguous words and performance on phonological STM tasks is merely due to the intervention of a common phonological processing factor, then both the STM task with high STM load and the STM task with low STM load but similar phonological processing requirements should correlate with increased processing times for phonetically ambiguous words; otherwise, if phonological maintenance capacities drive the association, then the STM task with the highest phonological STM load should correlate most strongly with the processing times for phonetically ambiguous words. (2) We could also expect a negative correlation between performance on phonological STM tasks and increased processing times for phonetically ambiguous words in a sentence context, meaning the higher phonological STM performance, the less time spent on phonetically ambiguous words. The most plausible interpretation for this potential outcome would be in terms of a phonological processing factor in conjunction with a semantic strategy: participants with highly developed phonological abilities will encode nonwords more efficiently in STM, leading to higher performance levels, and will detect more quickly the phonetic ambiguity of the target word in the sentence ; given that the word is phonologically ambiguous and a final decision cannot be reached only based on phonological analysis, sentence analysis processes will quickly move to subsequent words in order to disambiguate the word based on a more complete semantic representation of the sentence. This kind of semantic strategy would nevertheless imply, on behalf of the participants, some (conscious or unconscious) detection of the structure of the manipulated variables and the way they are implemented in the sentences.
Finally, in order to test the specificity of the hypothesized association between phonological STM capacity and phonetic/phonological analysis of words embedded in a sentence context, we included a third STM measure probing lexico-semantic retention capacities. This measure was a word immediate serial recall task, which, relative to nonword immediate serial recall, has been shown to be sensitive to lexico-semantic retention capacities (e.g., Hulme, Maughan & Brown, 1991; R. Martin et al., 1994, 1999). If the hypothesized relation between increased phonetic/phonological processing times and phonological STM performance is specific to phonological retention capacities, then no or at least a diminished correlation should be observed between online measures of phonetic/phonological processing times and the lexico-semantic STM task. On the other hand, this task should correlate with later stages of lexico-semantic sentence integration processes, as suggested by the studies of R. Martin and colleagues.
For the sentence processing task, phonetic ambiguity was implemented by exploiting the characteristics of categorical perception for consonant perception. Many studies have shown that speech perception processes categorize phonemes on the basis of a number of distinctive acoustic features such as voice onset time and first and second formant transitions of the acoustic frequency spectrum (Delattre, Liberman, & Cooper, 1955; Raphael, 2005). For example, in many languages, stop consonants can be distinguished by the duration of voice onset time (onset of vibration of vocal cords) relative to the plosive noise (e.g., /d/ versus /t/); in French, stimuli with a negative voice onset time of -30msec are categorized as /d/ while stimuli with a voice onset time of +30msec are categorized as /t/ (Serniclaes & Wajskop, 1979). The zone between -30msec and +30msec is a zone of categorical uncertainty where stimuli are increasingly perceived as ambiguous. In the present study, we exploited the existence of this zone of phonetic ambiguity, and manipulated the voice onset time of consonants of target words in such a way that the voice onset time fell within this zone of phonetic ambiguity. In accordance with previous studies that explored this type of phonetic manipulation on word identification in sentence or lexical processing tasks, we expected slowed identification processes for the manipulated words (e.g., Borsky, Tuller, & Shapiro, 1998; Connine & Clifton, 1987). These manipulated words were inserted in object-relativized or subject-relativized sentences, and the participants were instructed to listen to the sentences by activating each successive word by a button press response, allowing us to record online processing times for both natural and phonetically ambiguous words. At the end of the sentence, the participants had to judge the sentence according to its perceived semantic correctness/anomaly.
The STM tasks we correlated with online and offline sentence processing measures were a nonword immediate serial recall task, a single delayed nonword repetition task and a word immediate serial recall task. Nonword stimuli were used in the first two tasks in order to tap phonological STM processes as directly as possible. As already discussed, the nonword immediate serial recall task had the greatest phonological STM load as nonword lists of increasing length had to be recalled. The delayed nonword repetition task entailed the presentation of a single nonword which had to be recalled after a filled delay during which a backwards counting task was presented. This second task required similar phonological analysis and processing abilities as the nonword immediate serial recall task as the same type of nonword stimuli had to be processed; however STM load was reduced as a single nonword had to be repeated and rehearsal-based maintenance processes were reduced. At the same time, this task had greater working memory requirements given that the task combined storage requirements and the performance of a concurrent task (Peterson & Peterson, 1959). Hence this second task allowed us to control for general phonological processing abilities, shared with the nonword immediate serial recall task, as well as for elementary working memory processes. The last STM task, a word immediate serial recall task, assessed more directly lexico-semantic retention capacities, as compared to the nonword immediate serial recall task, in line with the studies by R. Martin and colleagues, using word and nonword span tasks in order to dissociate phonological and semantic retention capacities in brain injured patients (e.g., Freedman & Martin, 2001). In order to make this measure most sensitive to lexico-semantic item retention capacities, we determined the total number of items correctly recalled, independently of serial position. Previous studies have shown that lexico-semantic retention processes, relative to phonological processes, are most directly expressed via item recall rather than order recall (e.g., Poirier & Saint-Aubin, 1996; Nairne & Kelley, 2004).
Fifty-one1 young adults (age range: 18-30 years) volunteered for participation in this study. They were all native monolingual French speakers (30 female).
Sentence processing task – construction of phonetically ambiguous words. Eleven nouns containing one voiceless target consonant were selected in such a way that there existed another noun containing the voiced equivalent of the target consonant, all other phonemes being identical (see Table 1); the two sets of words were matched for lexical frequency, according to the Lexique database www.lexique.org (lexical frequency range for voiceless and voiced stimuli: 0.2 – 119.39 and 1.15 – 90.74, respectively). These target consonants were mostly plosives (/t/, /p/, /k/), except for one fricative (/s/), and represented the first phoneme of the first or second syllable of each word. The target nouns were spoken by a female voice and digitally recorded on computer disk (recording sampling rate: 44100 Hz, down-sampled to 22050 Hz for editing). In order to create phonetically ambiguous stimuli, we edited the waveforms (using Praat software, www.praat.org) of each noun by locating the onset of plosive or fricative noise and the onset of voicing for vowels (or voicing for the liquid in the stimulus /klas/) on the spectrograms and by shortening this period by 33%, 50% or 66% (portion nearest to the onset of voicing), in order to reduce voice onset times. In a pilot study, the manipulated stimuli, the natural voiceless stimulus as well as their natural voiced counterpart were presented to 16 adult participants for an identification task (each type of stimulus was presented 8 times for identification); using a two-alternative forced choice paradigm, the participants decided for each stimulus whether they heard the voiceless or the voiced stimulus (e.g., /boule/ vs /poule/). Based on the results of the pilot study, we selected for each of the eleven nouns, the version of the manipulated stimulus that yielded the most ambiguous identification responses, averaged over the 16 participants. The manipulated versions we retained and their characteristics in terms of voice onset time (VOT) and mean identification responses are reported in Table 1. The mean identification score, by collapsing over the 11 manipulated stimuli we retained, was .54 (.15), showing that identification for these stimuli was indeed highly ambiguous (although we should note that for a minority of stimuli, identification scores were slightly less ambiguous, the scores ranging from .27 to .73 for these stimuli).
< INSERT TABLE 1 ABOUT HERE >
Sentence processing task – construction of sentences. For each of the eleven target words and their voiced counterpart, two types of sentences were constructed. The two sentence types either contained subject-relativized (SR) or object-relativized (OR) clauses, allowing us to manipulate the location of the ambiguous target words in the sentence (early versus late positions), while using exactly the same words in both sentence types and the same number of words (e.g., “Ce1pain2 que3 prépare4 le5 boulanger6 plait7 beaucoup8 à9 ma10 soeur11” [This bread that the baker is preparing pleases a lot to my sister] versus “Le1 boulanger2 qui3 prépare4 ce5pain6 plait7 beaucoup8 à9 ma10 soeur11” [This baker who prepares this bread pleases a lot to my sister]). Hence in OR sentences, the target word always occurred in the second position while in SR sentences, the target words appeared in positions 6, 7 or 8, depending on verb phrase syntactical requirements. Implausible filler sentences were constructed by exchanging word positions of the plausible target sentences (e.g., “Ce pain qui plait beaucoup à ma soeur prépare le boulanger” [This bread that pleases a lot to my sister is preparing the baker] versus “Ma soeur qui plait beaucoup à ce pain prépare le boulanger” [My sister who pleases a lot to this bread prepares the baker]). Finally, each sentence occurred twice, once using the natural target word, once using the manipulated target word, resulting in a total of 176 sentences (half of them being implausible filler sentences)
The words of the different sentences were recorded into individual sound files by the same female voice as for the target nouns (recording sampling rate: 44100 Hz, down-sampled to 22050 Hz for presentation). Using E-Prime software (Psychology Software Tools, Pittsburgh, USA) the different sentences were presented in random order via high quality headphones connected to the soundcard of a PC. The participants were comfortably installed in front of a computer screen and the task started by a text display informing the participants that they would hear a sentence, presented word-by-word, and that they would have to press the space bar of the keyboard to activate the presentation of the different words. The screen then changed by displaying the text “new sentence” and the participants started listening to the different words at their own listening speed. At the end of the sentence, a new text display appeared, asking the participants whether the sentence they heard was plausible or not. The participants answered by pressing yes/no response buttons. The different sentence conditions (phonetically ambiguous versus natural) and sentence types (SR or OR) were presented in random order. Latency times for each of the eleven word positions in each sentence and response accuracy were automatically recorded from space bar and response key presses.
Nonword immediate serial recall task. This task was comprised of 60 nonword stimuli with a consonant-vowel-consonant structure; all nonwords contained phonotactically legal phoneme combinations relative to French phonology (Tubach & Boë, 1990). The stimuli were recorded by a female voice and stored as sound files. Using speech editing software, the sound files were then combined to form lists containing 2, 3, 4 or 5 nonwords, separated by a 500ms interstimulus interval. Each nonword occurred only once during the task. There were 4 trials for each list length. The lists were presented with increasing length, via high quality headphones connected to the sound card of a PC. The first four trials were practice trials and contained a single nonword. At the end of each list, the participants were instructed to recall all the nonwords in the order of presentation. The participants’ responses were digitally recorded for later transcription and scoring. We computed the number of items recalled in correct serial position, by pooling over the different sequence lengths.
Single nonword delayed repetition task. Thirty-four nonword stimuli with a consonant-vowel-consonant structure were constructed, following the same procedure as for the previous task. Each trial included the presentation of a single nonword, followed by an interfering task during which the participants had to start counting backwards aloud, by starting at 96 and by subtracting each time three units. After 10 seconds, the experimenter asked the participants to stop counting and to recall the nonword stimulus. The first four trials were practice trials. The participants’ responses were digitally recorded for later transcription and scoring. The number of correct responses was computed.
Word immediate serial recall task. This task was comprised of 108 word stimuli of moderate to high lexical frequency (larger than 500, according to the Brulex database; Content et al., 1990), with the same consonant-vowel-consonant structure as the nonword stimuli used in the two other short-term retention tasks. As for these tasks, the words were digitally recorded by a female voice and transformed to PC-compatible audio files, by arranging the words in lists containing two to six words, with four lists per list length. Within each list, the words were separated by a 500ms interstimulus interval. The words were presented for immediate serial recall via the same apparatus as used in the previous tasks. The participants’ responses were digitally recorded for later transcription and scoring. We computed the number of items recalled independently of correct serial position, by pooling over the different sequence lengths.
Vocabulary task. In order to control for the influence of general verbal abilities on verbal STM and sentence processing performance, we administered a standardized vocabulary task (EVIP, Dunn, Thériault-Whalen, & Dunn, 1993). This task is the French adaptation of the Peabody Picture Vocabulary Test (Dunn & Dunn, 1981) and requires the participant to match a spoken word to its picture (among a choice of four pictures). We determined the raw vocabulary score (maximum score: 170).
The impact of phonetic ambiguity on online and offline sentence processing measures A first set of analyses aimed to determine those online and offline sentence processing measures that are affected by the presence of phonetically ambiguous words, and which will be the target for the correlation analyses with the STM measures. All analyses reported here are restricted to plausible sentences only. A first analysis assessed the effect of phonetic ambiguity on online sentence processing for OR sentences, by examining response latency differences for natural and phonetically ambiguous sentences as a function of word position, excluding the final position which reflects offline sentence processing. A 10 (positions) by 2 (phonetically ambiguous / natural) repeated measures ANOVA showed a main effect of position, F(9, 450) = 25.93, MSE=11252, p<.0001, latencies being generally larger for initial sentence positions, and a main effect of phonetic ambiguity, F(1,50) = 8.52, MSE=9257, p <.01, latencies being globally larger for phonetically ambiguous sentence types, as well as a significant interaction, F(9, 450) = 3.45, MSE=4261, p<.001 (see Figure 1a). Planned comparisons showed that the effect of phonetic ambiguity was significant for the second position (where the ambiguous word occurred), F(1,50) = 15.35, MSE=7568, p <.001; further significant ambiguity effects were also observed for the fifth, F(1.54) = 8.61, MSE=3913, p<.01, and sixth positions, F(1,50) = 9.36, MSE=2415, p <.01, the latter two positions corresponding to the moment of occurrence of the subject of the OR clause whose meaning has to be integrated with that of the (phonetically ambiguous) subject of the main clause (see Figure 1a). The same analysis was conducted for SR sentences, where the ambiguous word occurred mainly in the seventh, but also sometimes in the sixth and eighth positions. This analysis yielded very comparable results, showing main effects of sentence position, F(9,450)=20.75, MSE=9273, p<.0001, and phonetic ambiguity, F(1,50)=11.39, MSE=7886, p<.001, as well as a significant interaction, F(9,450)=7.45, MSE=4820, p<.0001 (see Figure 1b). Planned comparisons showed a very large effect of phonetic ambiguity for the seventh position, F(1,50)=44.35, MSE=6477, p<.0001, and further significant effects in the sixth, F(1,50)=4.20, MSE=3913, p<.05 and eighth positions, F(1,50)=12.74, MSE=7794, p<.001; no other position showed effects of phonetic ambiguity.
< INSERT FIGURE 1 ABOUT HERE >
A final set of analyses addressed offline sentence processing by assessing the effect of phonetic ambiguity on response decision times and response accuracy. For both SR and OR sentence types, response times were longer when the sentence included a phonetically ambiguous word (although this was significant only for SR sentences) (reaction times: 1184+436.33 vs. 1292+437.21, t(50)=1.65, n.s. for OR sentences; 1344+407.74 vs. 1518+488.74, t(50)=2.45, p<.05 for SR sentences) and response accuracy decreased (accuracy: .94+.05 vs .75+.17, t(50)=7.88, p<.0001, for OR sentences; .92+.08 vs. .75+.15, t(50)=6.87, p<.0001, for SR sentences). The fact the response accuracy decreases shows that the participants are actually misidentifying the phonetically ambiguous word on a number of occasions and thus building up an ‘erroneous’ sentence interpretation. However, the word ‘erroneous’ should not be interpreted in its absolute meaning here given that for phonetically ambiguous sentence conditions, there are in fact two possible ‘correct’ sentence interpretations, depending on the outcome of the identification process for the phonetically ambiguous word. For example, if for the phonetically ambiguous sentence “Ce *pain que prépare le boulanger plait beaucoup à ma soeur”, ‘*pain’ is identified as ‘pain’, the sentence is semantically plausible; however, if ‘*pain’ is identified as ‘bain’, the sentence will be semantically implausible. Hence, the fact that accuracy decreases for the phonetically ambiguous sentence conditions is a natural consequence of the phonetic manipulations we had implemented. At the same time, this shows that the phonetic manipulations were effective in producing measurable phonetic/phonological uncertainty effects.
The analyses of the sentence processing measures show a clear and significant effect of phonetic ambiguity on both online and offline sentence processing. Increased phonetic ambiguity leads to longer processing times as predicted, reproducing earlier results by Borsky et al. (1998) but using a different paradigm to establish these effects. This increase is most marked for those words which are actually phonetically ambiguous, with smaller carry over effects to later sentence positions (only observed for OR sentence types). These carry over effects probably reflect more prolonged semantic interpretation processes rather than purely phonological processes; when the noun of the relative clause appears, the partial semantic sentence representation that has started to build up can be completed by including the meaning of the relative clause, which might be in contradiction with the sentence meaning implied by the word in the ambiguous sentence position, depending on the way the latter has been identified given its phonetic ambiguity.
In the light of these results, we selected the following positions for the correlation analyses with the different STM tasks: (1) Those positions where the phonetically ambiguous word had actually occurred: position 2 for OR sentences and positions 6, 7 or 8 for SR sentences. In order to obtain for the latter sentence type a single variable containing latency times for exclusively ambiguous words, we created a new variable by selecting, for each sentence, the precise position where the phonetically ambiguous word had actually occurred. If phonological STM capacity is related to phonological analyses stages during sentence processing, then an association should be observed between the nonword immediate serial recall task and these positions. (2) In order to investigate the specificity of these correlations, we also selected later sentence positions supposed to reflect semantic integration processes, which should correlate with the word immediate serial recall task, estimating lexico-semantic STM capacity, but not with the nonword recall tasks, estimating phonological short-term retention capacity ; for sentence processing measures reflecting semantic integration processes, we selected position 5 in OR sentences which had shown larger latency times in a phonetically ambiguous sentence context, presumably reflecting delayed semantic integration processes between the possible meanings generated by the earlier occurring phonetically ambiguous word and the meaning of the words introduced in the relative clause; we also selected final sentence positions reflecting offline semantic integration processes.
In order to reduce the number of variables to be entered in the correlation analyses as well as to obtain the most direct measure of phonetic ambiguity and its impact on phonological analysis and semantic integration processes, we created, for each position of interest, a processing cost variable by subtracting the latency time for the natural sentence condition from the latency time for the phonetically ambiguous sentence condition. This procedure yielded measures reflecting the processing cost for sentences containing a phonetic ambiguity, with some of these measures reflecting directly the impact of phonetic ambiguity (position 2 for OR sentences, positions 6, 7 and 8 for SR sentences) while other reflect delayed semantic integration processes indirectly induced by the phonetically ambiguous word that had occurred earlier in the sentence (position 5 for OR sentences; final position for both SR and OR sentence types).
Correlations between STM measures and selected sentence processing measures First, we checked the reliability of the different measures to be entered in the correlation analyses. For the three STM measures, recall accuracy was comparable, ranging between .62 and .70, suggesting that task difficulty was similar in all tasks, with no floor or ceiling effects (see Table 2). For the sentence processing cost measures, processing cost ranged between 36 ms (for position 6, 7 and 8 in SR sentences) and 181 ms (for position 5 in OR sentences) and were associated with a relatively important variability, especially for the final positions in both SR and OR sentences. The relatively large variance in the latter measures was partly due to some participants showing reversed processing costs (i.e., they responded faster for sentences containing a phonetically ambiguous word). The overall shape of the distribution of scores was also examined, showing that for all STM and sentence processing measures, skewness and kurtosis estimates remained within the recommended two standard error range (Tabachnick & Fidell, 1996). Finally, computation of Cronbach’s alpha showed satisfactory reliability estimates for all measures (all Cronbach’s alpha > .70; Cronbach, 1951).
< INSERT TABLE 2 ABOUT HERE >
Table 3 presents raw correlation matrixes between the different STM measures. Significant correlations but of moderate size only were obtained between all three measures. This suggests that the different measures did not measure entirely identical STM processes, consistent with the theoretical distinctions adopted in the present study, distinguishing between phonological and semantic retention capacities as well as storage and processing capacities. Table 4 presents the raw correlation matrix between the different sentence processing cost measures. A significant correlation was observed between processing cost measures in position 2 in OR sentences and positions 6, 7 and 8 in SR sentences, the positions where the phonetically ambiguous word actually occurred; these positions did not correlate with later or final sentence positions. A marginally significant correlation was also observed between sentence position 5 in OR sentences, presumably reflecting semantic integration processes, and the final position in OR sentences, consistent with the hypothesis that these two positions reflect the impact of phonetic ambiguity on later semantic levels of processing.
< INSERT TABLES 3 AND 4 ABOUT HERE >
The next set of analyses assessed the critical correlations between the STM tasks and processing cost variables for the positions where the phonetically ambiguous word had actually occurred. We observed significant positive correlations of moderate size between the nonword immediate serial recall task and processing cost variables in both SR and OR sentence types (see Table 5). These correlations also remained significant after control of general verbal ability as estimated by a receptive vocabulary task (partial correlation=.35, p<.01, for OR sentences; partial correlation=.29, p<.05, for SR sentences). A scatterplot of these correlations is presented in Figure 2, showing the higher performance on the nonword immediate serial recall task, the larger the processing cost for phonetically ambiguous sentence positions. It is also interesting to note that the participants with the poorest performance on the nonword immediate serial recall task had in fact a tendency towards presenting negative processing cost values: they spent less time for processing phonetically ambiguous words than they spent for processing natural stimuli, this tendency being most marked for OR sentences where the phonetically ambiguous word had occurred at the beginning of the sentence (15 participants presented negative processing costs for OR sentences, versus 6 participants for SR sentences). On the other hand, the single nonword delayed repetition task, having a lower phonological STM load, and the word immediate serial recall task, tapping lexico-semantic retention abilities, did not significantly correlate with processing cost variables for phonetically ambiguous words in any sentence type.
< INSERT TABLE 5 AND FIGURE 2 ABOUT HERE >
A last set of correlation analyses assessed the relations between the STM tasks and processing cost variables presumably reflecting delayed semantic integration and decision stages induced by the earlier occurring phonetically ambiguous word (see Table 6). This time, a significant negative correlation was observed between the word immediate serial recall task and position 5 in OR sentences (partial correlation, after control of receptive vocabulary knowledge: R=-.31, p<.05). All other correlations were non-significant, although there was a small correlation between the word immediate serial recall task and the final position in SR sentences, as well as between the single nonword delayed repetition task in and the same final position in SR sentences. The largest of these correlations, R=.-21, would have been significant for a sample greater than 85 participants, at a significance threshold of p=.05.
< INSERT TABLE 6 ABOUT HERE >
Before discussing the meaning and theoretical implications of the present study, we will first provide a brief summary of the main results. Nonword immediate serial recall, but not single nonword delayed repetition or word immediate serial recall correlated significantly with processing times for sentence positions where a phonetically ambiguous word had occurred. This correlation was positive: participants with higher performance on the STM task spent more time on phonetically ambiguous word positions in the sentence processing task. The word immediate serial recall task, but not the other STM tasks, correlated with a later sentence position in OR sentences, presumably reflecting delayed semantic integration processes as a result of the earlier occurrence of the phonetically ambiguous word. This correlation was negative: the higher performance on the word immediate serial recall task, the less time the participants spent on this position. Finally, none of the three STM tasks correlated significantly with end-of-sentence semantic anomaly decision times.
The relation between phonological STM capacity and online sentence processing
Assuming a deterministic relation between phonological STM and phonological analysis during sentence processing, we had hypothesized that a negative correlation should be observed between estimates of phonological STM and online processing times for phonetically ambiguous words occurring in a sentence context. The rationale behind this was that participants with higher phonological STM capacity will try to rely on their better preserved phonological STM traces of the initially presented information, in order to pursue ongoing phonological identification processes. Assuming a deterministic relation, we did not expect a negative correlation since this would mean that participants with better phonological STM capacity spend less time on phonetically ambiguous information, suggesting that they are able to identify phonological information faster, to detect and dismiss phonetic irregularities more quickly and to process phonological information more accurately in STM. In the latter case, the association between STM performance and online ambiguous word processing times could be most parsimoniously subsumed by a general phonological processing factor favouring phonological analysis and processing in both sentence and nonword STM tasks (Gupta, 2003; Majerus et al., 2004; Thorn & Frankish, 2005).
The positive correlation observed between phonological STM capacity and online phonetically ambiguous word processing times hence argues in favour of the hypothesized deterministic relation between phonological STM capacity and phonological analysis processes during sentence processing. Nevertheless, as we had noted in the Introduction, this positive relation could also mean that participants with better phonological processing capacity (and resulting higher phonological STM capacity) are more likely to identify phonetic irregularities, but contrary to the situation where a negative correlation would be observed, they will not quickly dismiss these irregularities but instead try to resolve the phonetic ambiguity, spending more time on the phonetically ambiguous information. Several additional results lead us to think that the latter interpretation is not the most likely. First, two phonological STM tasks sharing equal phonological processing requirements but differing in STM load had been used; if the association between phonological STM tasks and processing times for phonetically ambiguous words is to be related to a common factor such as greater phonological abilities and sensitivity, then both STM tasks, sharing similar phonological identification and processing requirements, should have correlated with increased online processing times for phonetically ambiguous words. This was not the case. Only the nonword immediate serial recall task, having the greatest phonological STM load, correlated with these online sentence processing measures. Second, an analysis of the processing cost variables for phonetically ambiguous words showed that the participants with the poorest phonological STM capacity had in fact a tendency towards presenting negative processing cost values: they spent less time for processing phonetically ambiguous words than they spent for processing natural stimuli. This means that participants with the lowest phonological STM capacity were equally sensitive to phonetic ambiguity, relative to the participants with the highest phonological STM capacity, but they reacted differently to the phonetic ambiguity. A possible explanation for the negative processing costs in participants with low phonological STM capacity is that, due to their less precise phonological STM traces, further prolonged phonological analysis of the presented stimulus becomes impossible since the phonological trace will have faded away; these participants might then try to rely to a greater extent on semantic contextual factors when encountering phonetically ambiguous material during sentence processing, moving forward as quickly as possible to the next words of the sentence in order to reconstruct the meaning of the preceding ambiguous word by using sentence-level semantic information. In sum, the overall pattern of results suggests that phonological STM is involved in online phonological analysis processes during sentence processing, enabling the acoustic/phonological form of a phonetically ambiguous word to be maintained longer, giving the phonological analysis processes additional time to reach the most accurate interpretation. This interpretation is also in line with computational models of language processing, considering that the speed of short-term trace decay for phonological information is an important factor underlying phonological analysis processes, and that an abnormally increased decay rate of phonological activations will lead to receptive language processing impairments (N. Martin & Saffran, 1992; R. Martin, Breedin, & Damian, 1999; Majerus et al., 2001).
The relation between lexico-semantic retention capacities, working memory and later semantic sentence integration processes
We also obtained some evidence for an association between lexico-semantic retention abilities, as indexed by word immediate serial recall, and later stages of semantic integration, confirming at the same time the specificity of the association between phonological STM estimates and phonological analysis stages during sentence processing. The word immediate serial recall measure correlated negatively with the processing cost variable for position 5 in OR sentences where the meaning of the relative clause has to be integrated with the possible meaning(s) of the earlier occurring phonetically ambiguous word. Participants with higher estimates of semantic retention capacities showed faster semantic integration processes for this position, suggesting that they had more precise semantic STM traces of the sentence content allowing them to reconstruct the meaning of the phonetically ambiguous word more quickly. One might argue here that this association could be mediated by a third factor, such as more general semantic processing abilities. Although we cannot reject this possibility with absolute certainty, we should note that the correlation between the semantic STM task and processing times in position 5 remained significant after control for general lexico-semantic abilities as estimated by a four-choice picture-word vocabulary matching task.
This selective relationship between phonological STM and phonological aspects of sentence processing, on the one hand, and semantic STM and semantic aspects of sentence processing, on the other hand, is in line with a number of studies by R. Martin and colleagues, showing that semantic maintenance processes during sentence processing are indeed dependent upon semantic retention capacities, but not phonological retention capacities (R. Martin et al., 1994, 1999; Freedman & R. Martin, 2001: R. Martin & Freedman, 2001; Martin & He, 2004; see also Butterworth et al., 1986; Waters et al., 1991). The original finding of the present study however is to show that phonological STM capacities also play a role, even if this role is restricted to specific aspects of phonological analysis during sentence processing.
Finally, none of the different STM tasks used in the present study correlated with end-of-sentence decision times. As we have discussed, most studies have observed a relation between offline sentence decision processes and more general working memory measures (e.g., Caplan & Waters, 1999; Just & Carpenter, 1992; McElree et al., 2003; Waters & Caplan, 2004). First, we will consider the word and nonword immediate serial recall measures. These tasks are, strictly speaking, ‘passive’ STM tasks rather than working memory tasks. Yet we might expect the word immediate serial recall measure to correlate with end-of-sentence processing decision times, given that previous studies establishing a relation between semantic retention capacities and sentence processing also used offline sentence processing measures (R. Martin et al., 1994, 1999). On the other hand, we should note that this relation is established on the basis of neuropsychological case studies showing greater difficulty in sentence anomaly judgments in patients with more severe semantic STM impairments, but not necessarily for any sentence type. This association between semantic STM and sentence comprehension is most expressed for sentences where semantic integration is delayed (e.g., Haarmann, Cameron, & Ruchkin, 2003; Martin & Miller, 1992). The significant correlation with position 5 in our study, presumably reflecting these delayed semantic integration processes, is clearly consistent with this interpretation. End-of-sentence decisions on the other hand might involve semantic and conceptual storage and processing capacities for integrated sentence information rather than passive maintenance processes for individual semantic information, these semantic working memory processes not being measured by our semantic STM task.
Finally, the single nonword delayed repetition task had clearly some working memory requirements given that a single item had to be stored while a concurrent task had to be performed, combining storage and processing demands as most working memory tasks do. Hence, at first glance, we might expect this task to correlate to a greater extent with end-of-sentence decision times. But again, the task we used here differs from verbal working memory tasks that have been shown to be associated with end-of-sentence decision times. These tasks were most often reading span measures or variants of this task (e.g., Dede et al., 2004; Van der Linden et al, 2001; Waters & Caplan, 2004). This task requires the participant to make a sentence anomaly judgment on successive sentences, while retaining the last word of each presented sentence. Hence, this working memory task is already a sentence comprehension task, in some way, requiring combining semantic storage and processing as is the case in sentence comprehension tasks. On the other hand, our ‘working memory’ task purely involved phonological information (nonwords) which had to be retrieved after a backwards counting distractor task. Along these task-related differences that might explain the absence of a relation between our semantic STM, phonological ‘working’ memory and off-line sentence comprehension measures, we should however also note that end-of-sentence processing times were extremely variable, and hence the reliability of these measures might not have been optimal. The same absence of relation was observed when we considered untransformed end-of-sentence processing times rather than the response time differences of the processing time variables (range of R’s : -.01 - -.23; mean R: -.10). This large variability of response times at end-of-sentence positions had possibly been accentuated by the fact that the participants were not explicitly instructed to respond as fast as possible, some participants taking more time than others to make a sentence decision, irrespective of STM/working memory capacity.
The present study suggests that phonological STM capacities support online phonological analysis during sentence processing, although, somewhat paradoxically, in the present case, phonological STM support actually led to slowed phonological analysis processes, caused by the fact that the to-be-analyzed phonological information was ambiguous; our task was designed in such a way that ongoing phonological analysis processes, prolonged via phonological STM maintenance of the initial input, could not resolve the phonetic ambiguity. This procedure was necessary to show that phonological STM is involved in auditory sentence processing when the usually very fast phonological analysis and identification processes take more time. By extension, our results suggest that any sentence comprehension situation involving suboptimal listening conditions or ambiguous phonetic material may be more dependent upon phonological STM capacities. This could for example be the case in elderly people whose lowered auditory processing abilities lead to suboptimal phonological input and may require the intervention of phonological STM capacities during auditory sentence processing to a larger extent, as compared to normal hearing young adults. This could also be the case in persons immersed in a foreign language environment and being less experienced listeners of this foreign language. Future studies may want to explore the more practical implications of the present findings.
Steve Majerus is a Research Associate funded by the Fund of Scientific Research – FNRS, Belgium.
Borsky, S., Tuller, B., & Shapiro, L.P. (1998). "How to milk a coat:" The effects of semantic and acoustic information on phoneme categorization. Journal of the Acoustical Society of America, 103, 2670-2767.
Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory. A case study. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 38, 705-737.
Caplan, D. & Waters, G. S. (1999). Verbal working memory and sentence comprehension. Behavioral and Brain Sciences, 22, 77-126.
Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235-271.
Connine, C.M., & Clifton, C. (1987). Interactive use of lexical information in speech perception. Journal of Experimental Psychology: Human Perception and Performance, 13, 291-299.
Content, A., Mousty, P., & Radeau, M. (1990). BRULEX. Une base de données lexicales informatisée pour le français écrit et parlé [BRULEX: A computerized lexical data base for the French language]. Année Psychologique, 90, 551-566.
Cronbach, L.J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297-333.
DeDe, G., Caplan, D., Kemtes, K., & Waters, G. S. (2004). The relationship between age, verbal working memory, and language comprehension. Psychology and Aging, 19, 601-616.
Delattre, P. C., Liberman, A. M., & Cooper, F. S. (1955). Acoustic loci and transitional cues for consonants. Journal of the Acoustical Society of America, 27, 769-773.
Dunn, L. M. & Dunn, L. M. (1981). Peabody Picture Vocabulary Test - Revised: Manual for Forms L and M. Circle Pines, MN: American Guidance Service.
Dunn, L. M., Thériault-Whalen, C. M., & Dunn, L. M. (1993). Echelle de vocabulaire en images Peabody. Adaptation française du Peabody Picture Vocabulary Test. Toronto, Canada: Psycan.
Ferreira, F. & Anes, M. (1994). Why study spoken language? In M.A. Gernsbacher (Ed.), Handbook of Psycholinguistics (pp. 33-56). San Diego: Academic Press.
Ferreira, F., Henderson, J. M., Weeks Jr, P. A., & McFarlane, D. K. (1996). Effects of lexical frequency and syntactic complexity in spoken language comprehension: Evidence from the auditory moving window technique. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 324-335.
Freedman, M. L. & Martin, R. C. (2001). Dissociable components of short-term memory and their relation to long-term learning. Cognitive Neuropsychology, 18, 193-226.
Gathercole, S. E., Frankish, C. R., Pickering, S. J., & Peaker, S. (1999). Phonotactic influences on short-term memory. Journal of Experimental Psychology: Human Learning and Memory, 25, 84-95.
Gupta, P. (2003). Examining the relationship between word learning, nonword repetition and immediate serial recall in adults. Quarterly Journal of Experimental Psychology, , 1213-1236.
Jacquemot, C., Dupoux, E., Decouche, O., & Bachoud-Levi, A. C. (2006). Misperception in sentences but not in words: Speech perception and the phonological buffer. Cognitive Neuropsychology, 23, 949-971.
Just, M. A. & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122-149.
Kotz, S. A., Von Cramon, D., & Friederici, A. D. (2005). On the role of phonological short-term memory in sentence processing: ERP single case evidence on modality-specific effects. Cognitive Neuropsychology, 22, 931-958.
Majerus, S., Lekeu, F., Van der Linden, M., & Salmon, E. (2001). Deep dysphasia: Further evidence on the relationship between phonological short-term memory and language processing impairments. Cognitive Neuropsychology, 18, 385-410.
Majerus, S., Van der Linden, M., Poncelet, M., & Metz-Lutz, M. N. (2004). Can phonological and semantic short-term memory be dissociated? Further evidence Landau-Kleffner Syndrome. Cognitive Neuropsychology, 21, 491-512.
Majerus, S., Van der Linden, M., Mulder, L., Meulemans, T., & Peters, F. (2004). Verbal short-term memory reflects the sublexical organization of the phonological language network: Evidence from an incidental phonotactic learning paradigm. Journal of Memory and Language, 51, 297-306.
Marslen-Wilson, W. & Welsh, A. (1978). Processing interactions and lexical access during word recognition in continuous speech. Cognitive Psychology, 10, 29-63.
Martin, N. & Saffran, E. M. (1992). A computational account of deep dysphasia: Evidence from a single case study. Brain and Language, 43, 240-274.
Martin, R.C., Breedin, S.D., & Damian, M.F. (1999). The relation of phoneme discrimination, lexical access, and short-term memory: A case study and interactive activation account. Brain and Language, 70, 437-482.
Martin, R. C. & Freedman, M. L. (2001). Short-term retention of lexical-semantic representations: Implications for speech production. Memory, 9, 261-280.
Martin, R. C. & He, T. (2004). Semantic short-term memory and its role in sentence processing: A replication. Brain and Language, 89, 76-82.
Martin, R., & Miller, M. (2002). Sentence comprehension deficits: independence and interaction of syntax, semantics, and working memory. In A. Hillis (Ed.), The handbook of adult language disorders (pp. 295-310). Hove, UK: Psychology Press.
Martin, R. C. (2006). The neuropsychology of sentence processing: Where do we stand? Cognitive Neuropsychology, 23, 74-95.
Martin, R. C. & Romani, C. (1994). Verbal working memory and sentence comprehension: A multiple-components view. Neuropsychology, 8, 506-523.
Martin, R. C., Shelton, J. R., & Yaffee, L. S. (1994). Language processing and working memory: Neuropsychological evidence for separate phonological and semantic capacities. Journal of Memory and Language, 33, 83-111.
Martin, R. C., Lesch, M. F., & Bartha, M. C. (1999). Independence of input and output phonology in word processing and short-term memory. Journal of Memory and Language, 41, 3-29.
McElree, B., Foraker, S., & Dyer, L. (2003). Memory structures that subserve sentence comprehension. Journal of Memory and Language, 48, 67-91.
Nairne, J. S. & Kelley, M. R. (2004). Separating item and order information through process dissociation. Journal of Memory and Language, 50, 113-133.
Poirier, M. & Saint-Aubin, J. (1996). Immediate serial recall, word frequency, item identity and item position. Canadian Journal of Experimental Psychology, 50, 408-412.
Peterson, L. R. & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Verbal Learning and Verbal Behavior, 10, 346-354.
Pfutze, E. M., Sommer, W., & Schweinberger, S. R. (2002). Age-related slowing in face and name recognition: evidence from event-related brain potentials. Psychology and Aging, 17, 140-160.
Raphael, L. J. (2005). Acoustic cues to the perception of segmental phonemes. In D. Pisoni & E. Remez (Eds.), The handbook of speech perception (pp. 182-206). Oxford: Blackwell Publishers.
Roberts, R.A., & Lister, J.J. (2004). Effects of age and hearing loss on gap detection and the precedence effect: broadband stimuli. Journal of Speech, Language and Hearing Research, 47, 965-978.
Serniclaes, W., & Wajskop, M. (1979). Prevoicing as a perceptual cue in French. In H. & P. Hollien (Eds.), Amsterdam Studies in the Theory and History of Linguistic Science IV, Vol.9, (pp 765-773). John Benjamins: Amsterdam.
Sommers, M. S. (2005). Age-related changes in spoken word recognition. In D. Pisoni & E. Remez (Eds.), The handbook of speech perception (pp. 469-493). Oxford: Blackwell Publishers.
Tabachnick, B. G., & Fidell, L. S. (1996). Using multivariate statistics (3rd Ed.). New York: Harper Collins.
Thorn, A. S. & Frankish, C. R. (2005). Long-Term Knowledge Effects on Serial Recall of Nonwords Are Not Exclusively Lexical. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 729-735.
Tubach, J. L. & Boe, L. J. (1990). Un corpus de transcription phonétique. France: Telecom.
Vallar, G. & Baddeley, A. D. (1984). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141.
Van der Linden, M., Hupet, M., Feyereisen, P., Schelstraete, M., Bestgen, Y., Bruyer, M. R. et al. (2001). Cognitive mediators of age-related differences in language comprehension and verbal memory performance. Aging, Neuropsychology and Cognition, 6, 32-55.
Waters, G. S., Caplan, D., & Hildebrandt, N. (1991). On the structure of verbal short-term memory and its functional role in sentence comprehension: Evidence from neuropsychology. Cognitive Neuropsychology, 8, 82-126.
Waters, G. S. & Caplan, D. (2004). Verbal working memory and on-line syntactic processing: Evidence from self-paced listening. Quarterly Journal of Experimental Psychology, 57(A), 129-163.
Table1. Characteristics of natural and phonetically ambiguous target stimuli
Table 2. Descriptive statistics of the different STM and sentence processing cost measures.