Vous êtes sur la page 1sur 16

Brain and Language 81, 368383 (2002) doi:10.1006/brln.2001.2531, available online at http://www.idealibrary.

com on

Sonority Constraints on OnsetRime Cohesion: Evidence from Native and Bilingual Filipino Readers of English
Angelo Alonzo and Marcus Taft
University of New South Wales, Sydney, New South Wales, Australia Published online January 18, 2002 Research in English suggests that syllables can be analyzed in terms of two subunitsthe onset (dened as the initial consonant or consonant cluster) and the rime (the unit formed by the vowel and following consonant/s). This study investigated whether nonnative readers of English, which in the case of the present study were native Filipino speakers, also make use of onsetrime units, particularly when some features of their native language (namely inxation and reduplication) appear to foster no awareness of such units. In two lexical decision experiments, monosyllabic English words were presented, divided in between their rst and second consonants (e.g., B LIND), at their onsetrime boundary (e.g., BL IND), or at their antibody boundary (e.g., BLI ND). Results indicated that the processes of inxation and reduplication did not affect the English word processing of native Filipino speakers. Rather, results for both native Filipino and native English speakers suggest that onsets composed of an s consonant sequence (e.g., STAMP) are less cohesive than onsets comprised of a stop liquid sequence (e.g., BLIND). It was concluded that not only may sonority constraints underlie onset cohesiveness, but that such phonetic properties may also be involved in visual word recognition. 2002 Elsevier Science (USA)

INTRODUCTION

Research in English suggests that syllables can be analyzed according to a hierarchical internal structure. The most widely accepted view is that monosyllabic words can be divided into two subunitsan onset (dened as the initial consonant or consonant cluster, such as tr from trip) and a rime or body (the unit formed by the vowel and any following consonant/s, such as ip from trip). Past studies have suggested that the sensitivity to onsetrime units demonstrated by native English speakers may be due to the degree of orthographic transparency, or consistency of grapheme-to-phoneme correspondences (GPCs), in the English language (e.g., Goswami, Gombert, & de Barrera, 1998; Kessler & Treiman, 1997; Treiman, 1992; Treiman, Mullenix, Bijeljac-Babic, & Richmond-Welty, 1995). This is because in English, where many vowel graphemes have multiple pronunciations (e.g., ea can be pronounced as /i/ as in beat, /e/ as in steak, or /E/ as in head), consideration of the postvocalic consonants (i.e., t, k, and d in the example cases) may increase the degree to which the pronunciation of the vowel can be predicted, whereas consideration of the prevocalic consonants usually does not (Treiman, 1992). Thus, in the
Address correspondence and reprint requests to Angelo Alonzo, School of Psychology, University of NSW, Sydney, NSW 2052, Australia. E-mail: angelo alonzo@hotmail.com. 368
0093-934X/02 $35.00 2002 Elsevier Science (USA) All rights reserved.

SONORITY CONSTRAINTS

369

case of English, perception of rime units in particular may facilitate word recognition as such units preserve some semblance of consistency in their GPCs. Such a premise also implies that other languages may either encourage or discourage onsetrime sensitivity depending on the transparency of their orthography. Supporting this assertion, in a study which compared the nonword reading performances of English, French, and Spanish children, Goswami et al. (1998) found that all language groups were more accurate and faster in reading nonwords that shared their rime units, both orthographically and phonologically, with real words (e.g., dake from cake, take, and make) than nonwords which did not (e.g., faish and zoip). However, the difference in error rates for the two types of nonword was greater for the English and French groups compared to the Spanish group. This suggests that because the GPCs in Spanish are highly predictable relative to French and English, reliance on processing units (e.g., onsets and rimes) beyond the level of individual phonemes and graphemes is not crucial to successfully decoding unfamiliar letter strings in Spanish. In contrast, because English and French possess relatively more opaque orthographies than Spanish, larger units, such as the rime which make spelling-tosound decoding more reliable, may afford a greater processing advantage than a letter-by-letter decoding strategy. In another experiment of the same study, Goswami et al. compared the reading performances of English and French children for nonwords which shared the rime of a real word, both orthographically and phonologically (e.g., dake), and nonwords with rimes that were only phonologically similar to real words (e.g., daik). While a signicant facilitation of orthographic familiarity was found for both languages, facilitation was signicantly stronger for the English children, indicating that successful decoding from spelling-to-sound in English relies more heavily on rime-sized orthographic representations compared to French. Thus, there is evidence that onsetrime units are not equally prominent across the word recognition processes of different languages with alphabetic orthographies. Rather, the degree to which they are involved in reading performance appears to be dependent on the transparency of a languages orthography. However, while comparisons betwen native speakers of different languages may address questions as to what sublexical units are important for optimal reading performance in particular languages, investigating the reading processes of nonnative speakers may potentially address questions as to how important a role certain sublexical units play in the lexical processing of a particular language. The present study addressed this latter question in an attempt to investigate the importance of onsetrime sensitivity to competent reading in English. If it were found that nonnative readers of English are indeed sensitive to onsetrime units despite features of their native language which appear to foster no awareness of such units, the specic importance of onsetrime units to the processing of English words could be further underlined. If, however, nonnative readers of English were to show no sensitivity to onsetrime units, presumably in deference to properties of their native language which would not encourage sensitivity to such units, there would be evidence to suggest that onsetrime units are not equally crucial to all readers of English and that properties related to a readers native language would more likely underlie their propensity to developing an awareness of certain sublexical units. Native speakers of Filipino, the national language of the Philippines, were considered good candidates for comparison with native English speakers, as Filipino differs from English along a number of linguistic dimensions which appear to inuence the development of onsetrime sensitivity. First, the alphabetic, roman orthography of Filipino is highly transparent compared to English. In contrast to English, where much of the irregularity between GPCs lies in the vowel graphemes (Berndt, Reg-

370

ALONZO AND TAFT

gia, & Mitchum, 1987), the vowel inventory of Filipino contains only ve vowels and six diphthongs, all of which have only one phonemic correspondence. Furthermore, the consonant inventory of the traditional Filipino alphabet contains only 15 graphemesall with a single phonemic correspondencewith an additional eight , Q, V, X, and Z) used only for words borrowed predominantly letters (C, F, J, N from English and Spanish. Second, Filipino is a highly syllabic language relative to English. Unlike English, in the case of Filipino words, there are as many syllables as there are vowel letters, with the average length of Filipino syllables being less than that of English syllables. Thus, as studies of alphabetic languages other than English suggest (e.g., Cardoso-Martins, 1995; Goswami et al., 1998; Wimmer & Goswami, 1994), the highly syllabic structure of Filipino, as well as its shallow orthography, should not encourage native Filipino speakers to develop onsetrime representations for Filipino words, nor should it predispose them toward developing a sensitivity to onsetrime units for English words. Moreover, Filipino possesses a number of relatively uncommon, but highly productive, morphological operations which, if transferred to the English reading processes of native Filipino speakers, appear to interfere with onsetrime sensitivity. One feature of interest is the use of inxes, a type of afx inserted within a root word to mainly derive verbal forms from nominal roots. The two main inxes used in Filipino are -UM- and -IN-, both of which are inserted between the rst consonant and second letter of a root word. For example, the root word TALOP (peel) can be inxed to produce the innitive form TUMALOP or the past tense TINALOP. Further, in cases where a word begins with a consonant cluster (e.g., TRIPLE), the inx -INI- is inserted within the initial consonant cluster to produce TINIRIPLE (to triple). A second feature of interest is reduplication, a process by which the future tense of a verbal root can be produced by reduplicating the initial consonantvowel string of a root word and attaching it to the root as a prex. For example, the future tense of the root BILI (buy) can be produced by reduplicating its rst consonant vowel string BI and prexing this to the verbal root to produce BIBILI. Pertinent to the questions addressed in the present study, if the inxation and reduplication processes were to indeed exert an inuence over the English reading process of native Filipino speakers, such operations appear to foster no awareness of onset rime units. First, in the case of words beginning with a consonant cluster onset (e.g., TRIP), the inxation process may foster a tendency to break up the complex onset unit such that a word like TRIP would be divided between the initial consonant and the remainder of the word (i.e., T-RIP) rather than into the more widely accepted onsetrime division (i.e., TR-IP). Second, the reduplication process may favor the isolation of the initial CV- or CCV- sequence of a word (a unit known as the antibody) from its postvocalic consonant/s. Thus, a word such as TRIP may be divided between its antibody unit TRI and its postvocalic consonant P, a division which notably straddles the onsetrime boundary. Thus, the rst experiment of this study addressed these possibilities by investigating how native Filipino speakers analyzed monosyllabic English words beginning with complex onsets. Specically, this experiment attempted to identify whether native Filipino speakers would demonstrate an awareness of onsetrime units, presumably in deference to the relative irregularity of English GPCs. If it were the case that native Filipino speakers do exhibit an awareness of onsetrime units, despite features of their native language appearing to encourage no awareness of such units, there would be further evidence to suggest that the relative opacity of English orthography may exert a strong inuence on the establishment of onsetrime units. Conversely, if the performance of native Filipino speakers were to be associated more with processes found in their native language, it may then be the case that awareness of onset

SONORITY CONSTRAINTS

371

rime units is an adaptation that nonnative readers of English are unlikely to make, despite the evidence which suggests that an awareness of such units may negate to some extent the irregularity of English GPCs.
EXPERIMENT 1

Method
Participants. Thirty monolingual English speakers, all of whom were undergraduate students at the University of New South Wales, participated in the present experiment and were given course credit in return for their participation. In addition, 30 native Filipino speakers residing in the Philippines were selected. As English, being one of the ofcial languages of the Philippines, is also the ofcial language of instruction and commerce, all Filipino participants selected were university educated, bilingual Filipino English speakers. This criterion was applied in order to obtain a nonnative English speaking group with a high level of prociency in English. Stimuli and Design. A 2 3 factorial design was used, with the between-participants factor being native language background and the within-participants factor being type of intraword division (C-C, Onset-Rime, and Antibody). Thirty monosyllabic English words and 30 nonwords, all beginning with a complex onset, were selected as experimental items. Of each subset of word and nonword items, 10 were divided, using two slash symbols (//), in between their rst and second consonants (e.g., S//TAMP); 10 were divided at their onsetrime boundary (e.g., ST//AMP); and 10 were divided at their antibody boundary (e.g., STA//MP).1 Participants within each language group were divided into three groups such that while each group was presented with all 60 word and nonword items, the way in which the items were divided varied between the three groups.2 Thus, for example, while one group was presented with the items C//LERK, TW//IST, and STA//MP, a second group was presented with S//TAMP, CL//ERK, and TWI//ST and a third group was presented with T//WIST, ST//AMP, and CLE//RK. Participants were also presented with 10 practice items and 2 ller items prior to the commencement of the experiment. Apparatus. For native English speakers, all items were presented on a video display unit using Dmastr for Win32 (DMDX) software. For native Filipino speakers, all items were presented on an IBM compatible Toshiba T4700CS laptop computer using DMASTR-III software. Procedure. A standard lexical decision task was used. Participants were tested individually and were seated facing a computer monitor which presented the items one by one in random order. Participants were informed that single letter strings would appear on the screen one by one and that their task was to classify, as quickly and as accurately as possible, whether each presented item was an English word or a nonword by pressing one button labeled YES if they judged the item to be a word and another button labeled NO if they judged the item to be a nonword. They were also informed that each letter string would be divided by two slash symbols but that they should ignore the slash symbols and analyze the item as one whole word. All participants pressed the YES button with their dominant hand such that right-handed participants pressed the YES button with their right hand and the NO button with their left hand and vice versa for the left-handed participants. The session began with the presentation of the 10 practice items and the 2 ller items, followed by the randomized presentation of the 60 word and nonword items. Each item was presented at the center of the screen in uppercase letters for a maximum of 3 s, during which time participants were required to make their response. If a participant failed to make a response within 3 s of an item appearing, the item disappeared and their response was recorded as an error for that particular item. Response times were measured from the onset of an item presentation to the time the participant recorded a response.

Results The results of participants with overall error rates that exceeded 30% were excluded from the data analyses and replaced. Under this criterion, two participants from the native Filipino group were replaced. In this, and the subsequent experiment to be reported, the effects of isolated trials with exceptionally long or short latencies
1 In a minority of cases, when words were split, the segment following the slash marks formed an English word. However, to counterbalance this, an equal number of nonword items also contained segments which formed an English word following the slash marks. 2 Note that because the word items were of primary interest in the present experiment, only the presentation of the word items varied in this way. For the nonword items, the way in which these items were divided did not vary between the three groups.

372

ALONZO AND TAFT

TABLE 1 Decision Latencies (in Milliseconds) and Errors (in Percentages in Parentheses) of Native English and Native Filipino Speakers for Words
Conditions C-C OnsetRime Antibody English 707 (1) 702 (2) 727 (3) Filipino 803 (3) 834 (2) 839 (4)

were minimized by establishing cutoff points 2 standard deviation units above and below the overall mean latency for each participant and setting any outlying values equal to the appropriate cutoff value. Lower and upper cutoff limits were set at 200 and 3000 ms respectively. Trials in which an error was made were omitted from the analysis of reaction times (RTs). Mean reaction times and error rates (ERs) averaged across participants for each language group are presented in Table 1. Separate analyses of variance (ANOVAs), treating participants (Fp) and items (Fi) as random variables, were performed for the RT and ER data. The analysis of RTs revealed a signicant main effect of native language [Fp(1, 54) 10.64, p .05; Fi(1, 58) 50.04, p .05] and one signicant main effect of intraword division. There was a signicant difference between the C-C and Antibody conditions [Fp(1, 54) 8.49, p .05; Fi(1, 58) 5.95, p .05], but there was no signicant difference between the C-C and OnsetRime conditions [Fp(1, 54) 1.06, p .05; Fi(1, 58) 0.93, p .05] nor between the OnsetRime and Antibody conditions [Fp(1, 54) 3.12, p .05; Fi(1, 58) 1.98, p .05]. There were also no signicant interactions between native language and intraword division. The analysis of ERs yielded no signicant effect of native language nor of intraword division. There were also no signicant interactions between native language and intraword division. As the primary interest of the present study was to investigate how English words are processed by native and nonnative readers of English, results for nonwords were not analyzed. Furthermore, it is believed here that nonword data derived from lexical decision tasks may be somewhat limited in the sense that, as suggested by many accounts of nonword processing (e.g., Coltheart, Davelaar, Jonasson, & Besner, 1977; Grainger & Jacobs, 1996; Ziegler & Perry, 1998), nonwords are rejected by default. That is, a default NO response is given when there is an insufcient amount of lexical activation within some set time. Discussion The results from Experiment 1 showed that, overall, dividing the words at their onsetrime boundaries did not facilitate word recognition relative to dividing their onset clusters or dividing them at their antibody boundaries. Instead, breaking up the onset units of the words facilitated word recognition relative to dividing the words at their antibody boundaries. Thus, the results suggest that overall, the native English and native Filipino speakers did not particularly utilize the onsetrime distinction in their recognition of words. However, in light of the signicant difference between the C-C and Antibody conditions, it would be unwarranted to conclude that participants reactions indicated no sensitivity to onsetrime units. What the results do suggest is that while complex onsets may not be particularly cohesive units since separating

SONORITY CONSTRAINTS

373

their constituents appeared to facilitate decision latencies, rimes may be more rmly established units, as dividing the rimes between their peaks and codas (i.e., the antibody division) did hinder decision latencies. Moreover, that a similar pattern was found for both native English and native Filipino speakers suggests that the rime unit may be particularly important to how English words are represented and accessed since even readers from a nonnative language group indicated some sensitivity to such units. Although no signicant interactions were found between the two factors of native language and intraword divisionsuggesting that native language did not differentially affect how English words were analyzedthe pattern of results did indicate some differences in performance between the two language groups. Interestingly, the principal difference appeared to be the degree to which the onsetrime division was utilized. The average RT of the native English speakers for the OnsetRime condition was 25 ms faster than that of the Antibody condition and only 5 ms faster than that of the C-C condition. For the native Filipino speakers, though, the average RT for the OnsetRime condition was only 5 ms faster than that of the Antibody condition, but was 31 ms slower than that of the C-C condition. Indeed, when the two language groups were analyzed separately, the difference between the OnsetRime and Antibody conditions reached signicance for the native English group albeit only in the participant analysis [Fp(1, 27) 5.63, p .05; Fi(1, 29) 2.38, p .05], but there were no signicant differences between the C-C and OnsetRime conditions [Fp(1, 27) 0.17, p .05; Fi(1, 29) 0.2, p .05] nor between the C-C and Antibody conditions [Fp(1, 27) 3.5, p .05; Fi(1, 29) 1.27, p .05]. In the analysis of the native Filipino group, the difference between the C-C and Antibody conditions reached signicance [Fp(1, 27) 5.09, p .05; Fi(1, 29) 5.51, p .05], but the differences between the C-C and OnsetRime conditions [Fp(1, 27) 1.99, p .05; Fi(1, 29) 2.44, p .05] and the OnsetRime and Antibody conditions [Fp(1, 27) 0.16, p .05; Fi(1, 29) 0.25, p .05) did not reach signicance. Thus, despite the lack of an interaction between native language and intraword division, to conclude that the native English and native Filipino speakers were sensitive to the same units in their analysis of English words would belie the pattern of results. The native English speakers did indeed demonstrate some sensitivity to the onsetrime division as indicated by the difference between the OnsetRime and Antibody conditions, whereas the native Filipino speakers did not appear to utilize the onsetrime division as suggested by the marginal 5 ms difference between the Onset Rime and Antibody conditions. Moreover, as the results suggest that the decision latencies for the native Filipino group were specically facilitated when complex onsets were split, there is some indication that the inxation process may have exerted an inuence over their processing of English words. In contrast, there is no evidence pointing to an inuence of the reduplication process; dividing words at their antibody boundaries, which corresponds to the reduplicated segments in Filipino morphology, actually hindered rather than facilitated word recognition. Admittedly, though, because of the absence of an interaction between native language and intraword division, conclusions at this point regarding any observed differences between the two language groups must remain tentative. An examination of the RTs for each item, however, indicated that while decision latencies for some items showed an advantage of the C-C condition over the Onset Rime and Antibody conditions, other items showed no clear effect. This raised the speculation that perhaps native Filipino speakers may have shown a greater preference for the C-C condition over the OnsetRime and Antibody conditions for words with onset clusters that also exist in Filipino. If the inxation process does indeed inuence the way in which native Filipino speakers process onset clusters, it is con-

374

ALONZO AND TAFT

TABLE 2 Decision Latencies (in Milliseconds) of Native Filipino Speakers for Words with Onset Clusters that Exist/Do Not Exist in Filipino
Conditions Exist C-C OnsetRime Antibody Do not exist C-C OnsetRime Antibody Filipino 827 801 855 790 859 833

ceivable that they may have shown a greater effect for familiar onset clusters. To address this possibility, a post hoc analysis was conducted only for the native Filipino group, separating words which began with onset clusters that exist in Filipino from words which began with onset clusters that do not exist in Filipino. The results are presented in Table 2. Post hoc analysis. A 2 3 itemwise analysis was conducted, with the betweengroups factor being whether the words began with onset clusters that exist in Filipino. Thirteen items began with onset clusters which exist in Filipino, and 17 items began with onset clusters which do not exist in Filipino. Two signicant interactions were found, indicating that native Filipino speakers processed onset clusters differently depending on whether the onset clusters also exist in Filipino. However, results were contrary to expectations. First, recognition for words with onset clusters which exist in Filipino were facilitated when such words were divided at their onsetrime boundary compared to when their onset clusters were split, with the opposite pattern holding true for words with onset clusters which do not exist in Filipino [Fi(1, 28) 8.88, p .05]. Also, recognition was faster when words with onset clusters which exist in Filipino were divided at their onsetrime boundaries compared to when they were divided at their antibody boundaries while the opposite pattern was found for words with onset clusters which do not exist in Filipino [Fi(1, 28) 6.16, p .05]. That the results of the native Filipino group were contrary to what would be expected had the factor of onset cluster familiarity inuenced their processing of onset units brings into doubt the previous suggestion that the inxation process may have inuenced the way in which native Filipino speakers recognized English words. Contrary to what was hypothesized, native Filipino speakers demonstrated a preference for splitting the onsets of words with onset clusters which do not exist in Filipino rather than maintaining the cluster as a cohesive onset unit; for words with onset clusters which do exist in Filipino, however, a preference was shown toward processing the onset cluster as a cohesive unit rather than splitting its constituents. Had the inxation process indeed inuenced the performance of the native Filipino group, as suggested by the overall analysis, the opposite pattern to what was found in the post hoc analysis would have been expected. Importantly, the post hoc analysis raised the possibility that factors other than that of whether onset clusters exist in Filipino inuenced the likelihood of whether splitting up or preserving an onset cluster would facilitate word recognition. A further examination of the words placed in the exist and not exist conditions revealed that most words in the exist group began with consonants that can be categorized under a different phonetic class to that of the initial consonants which predominantly

SONORITY CONSTRAINTS

375

occurred in the not exist group. All 13 words in the exist group began with a stop consonant, but of the 17 words in the not exist group, only four words began with a stop consonant while 13 words began with a fricative consonant. Thus, it is possible that factors related to phonetic class may underlie the results found in the post hoc analysis. Specically, a review of the existing literature on onset clusters suggests that the differences in performance between the two onset cluster groups may be due to differences in the way onset clusters beginning with s are processed compared to other clusters (e.g., Davis, 1990; Kaye, Lowenstamm, & Vergnaud, 1990; Treiman, Gross, & Cwikiel-Glavin, 1992). As 9 of the 13 fricative-initial onsets began with s, it is possible that this subset of words may be the source of the observed effects. Admittedly though, because Experiment 1 was not specically designed to test this factor, the following conjectures drawn from a post hoc analysis must at this point be taken tentatively. Pertinent to the issue at hand though is the question of why should phonetic class or the occurrence of s in initial position affect the cohesiveness of an onset cluster? First, some linguists and psycholinguists (e.g., Davis, 1990; Kaye, Lowenstamm, & Vergnaud, 1990; Treiman, Gross, & Cwikiel-Glavin, 1992) have proposed that clusters beginning with s are not legal onsets in English. Rather, the s is instead an afx, a stray segment that is joined to the word or syllable during the nal stages of phonological derivation. Thus, in such words as stamp and snack, the initial s is not part of the words onset. Further, Davis (1990) has argued that the existence of consonant cluster onsets may be subject to language-specic constraints related to a principle which he termed minimal sonority distance. Linguists such as Davis (1990) have theorized that there are universal principles that govern some of the structural constraints on syllable composition. The phonetic notion of sonority is one such principle. This refers to the degree of loudness or vowel-likeness of a particular phoneme, with vowels being the most sonorous type of phoneme followed in descending order by liquids, the letter n, the letter m, coronal fricatives, noncoronal fricatives, voiced stops, and voiceless stops. Davis (1990) has proposed that only certain consonant clusters can form legal onsets depending on how discrepant the two consonants are in terms of their degree of sonority. The closer two consonants are in terms of sonority, the less likely they are to exist as legal onset clusters. Further, Davis (1990) has specied that the initial letter of consonant clusters can only be incorporated into the onset if there is a minimum of at least four positions between the two consonants along the sonority hierarchy. In the case of s, because s, being a coronal fricative, appears in the middle range of the sonority hierarchy, all consonants which can follow s at word onset position (i.e., c, k, l, m, n, p, t, and w) are within four positions of s along the sonority hierarchy. Therefore, as the difference, or the distance, in sonority between s and all the consonants that can possibly follow it at word onset position is always less than four positions, s cannot be incorporated into an onset cluster according to the sonority distance principle. In contrast, stopliquid onset clusters should form a cohesive onset unit. Because stop consonants are the lowest in sonority compared to all other consonants and can only be followed by liquid consonants (i.e., l and r) or the letter w, the three most sonorous of all consonants, the sonority distance between all possible stopliquid or stopw combinations is always at least four positions along the sonority hierarchy. Thus, according to the sonority distance principle, the stop consonant of stopliquid onset clusters should be incorporated into an onset unit. The results for the native Filipino group are congruent with the sonority distance principle. As expected, for the exist group, which contained words that all began with a stopliquid onset cluster, decision latencies for the OnsetRime condition were faster than for the C-C condition, suggesting that stopliquid clusters were

376

ALONZO AND TAFT

indeed processed as cohesive onset units. Conversely, for the not exist group, which contained predominantly fricative-initial or s-initial onset clusters, decision latencies were faster for the C-C condition compared to the OnsetRime condition, indicating that the s in s-initial clusters may not be incorporated in a cohesive onset unit. However, although the results for the native Filipino group lend support to the theory of onset cohesion being governed by principles related to sonority distance, there is an alternative explanation that the present results cannot resolve. The sonority contour principle implies that language users prefer a steady rise in sonority from the beginning of a syllable to the vowel or peak, followed by a gradual decline in sonority from the peak to the end of the syllable (Treiman et al., 1992). Words that begin with a stopliquid or a stopglide sequence (e.g., blind and twist) possess this characteristic, as stop consonants such as b and t are low in sonority, and liquid and glide consonants such as l and w respectively are higher in sonority. In the case of some s-initial clusters, though, such as sc, sk, sp, and st, a steady rise in sonority is not followed. An implication of the sonority contour principle is that clusters which rise in sonority from the rst to the second consonant (i.e., all stop liquid and stopglide clusters) should constitute a legal onset while clusters which violate a sonority contour (i.e., sc, sk, sp, and st) should not constitute a cohesive onset unit. Furthermore, if the sonority contour principle were to hold, s-initial clusters such as sl, sm, sn, and sw, which do show a rise in sonority, should also be processed as legal onset units. Thus, in contrast to the sonority distance principle, which would hold that all s-initial clusters should be processed similarly regardless of whether they follow a sonority contour, the sonority contour principle maintains that sinitial clusters should be treated differently depending on whether a rise in sonority from the initial consonant to the peak is followed. However, there was an insufcient number of s-initial clusters in Experiment 1 which either followed or violated a sonority contour to warrant any rm conclusions as to which sonority based principle the native Filipino group was sensitive. Thus, Experiment 2 was specically designed to resolve the differing predictions made by the sonority distance and sonority contour principles. Furthermore, it is no small issue that the explanations put forward for the results of the post hoc analysis are essentially based on phonological theory. Indeed, it may be argued that there is no strong reason to expect that a phonological factor such as sonority should inuence performances in an experimental paradigm like the lexical decision task. As items in such a task are processed visually and silently, it is conceivable that participants may successfully perform the task without necessary recourse to phonological information. That is, an afrmative response is given if a match between the visual input and an orthographic representation in the lexicon is found, whereas a negative response is given if no such match is found within some temporal threshold. There appears to be no reason to assume that a particular items phonological properties must invariably be processed before the match between a visual input and an orthographic representation in the lexicon is afrmed. Clearly, further testing is needed to resolve the above-mentioned conjectures. Therefore, before any rm conclusions can be made with regard to the possible differences identied in the present post hoc analysis, the results of Experiment 2, which was specically designed to address the sonority-based theories presently raised, are considered.
EXPERIMENT 2

There is indeed evidence to suggest that children as well as adults treat s-initial onset clusters differently than other clusters. For instance, Stemberger and Treiman

SONORITY CONSTRAINTS

377

(1986) found that although data from natural and experimentally elicited speech errors indicated that the rst consonant (C1) of an onset cluster was less likely to be lost, added, or mispronounced compared to the second consonant (C2), there is a tendency for C1 to be lost more often when it is s rather than a voiceless stop. Also, using a phoneme shifting task in which participants extracted a phoneme sequence from one item and substituted it into another item, Fowler, Treiman, and Gross (1993) found that it was easier for participants to shift the C1 of s-initial onset clusters compared to the C1 of fricativesonorant or stopliquid onset clusters. Further, the initial s appeared to be as easy to shift regardless of whether it was followed by a sonorant or a stop consonant, a nding that would not be predicted by the sonority contour principle but would support the sonority distance principle. Notwithstanding, while these studies do suggest that s-initial clusters are processed differently to other clusters, it may be argued that the results found cannot be readily extrapolated to the case of reading performance and the processes involved in lexical access. Because the tasks employed by previous studies (e.g., Fowler et al., 1993; Stemberger & Treiman, 1986) essentially required participants to produce an overt articulatory responsewith participants often required to produce a nonword as their responseit is conceivable that the differential treatment of s-initial clusters compared to other clusters may be restricted to tasks which inherently require participants to engage in phonologically based processes. Moreover, as the reasons often put forward to account for the exceptional case of s-initial clusters invariably stem from phonological theory, there is again no strong reason to expect that such theories can be equally applied to silent reading performance, unless of course evidence is found to suggest that properties related to sonority are to some extent automatically activated in the reading process. Thus, the aims of Experiment 2 were twofold. First, it attempted to explicitly test the theories raised in the post hoc analysis of Experiment 1, namely the sonority distance and sonsority contour principles. To reiterate, the sonority distance principle would predict that all s-initial clusters, regardless of whether the s is followed by a sonorant or a stop consonant, should be treated differently than all other types of clusters. Specically, decision latencies for s-initial clusters should be faster when the cluster is split (i.e., C-C condition) compared to when the cluster is preserved (i.e., OnsetRime or Antibody condition) while for stopliquid clusters, decision latencies should be faster when the cluster is preserved compared to when it is split. According to the sonority contour principle, though, s-initial clusters which follow a sonority contour should be treated similarly to stopliquid clusters. That is, decision latencies for such clusters should be faster when the cluster is preserved compared to when it is split. For s-initial clusters which violate a sonority contour, however, decision latencies should be faster when the cluster is split compared to when it is preserved. Regardless of the type of onset cluster, however, the Antibody condition should be the consistently slowest condition since, as suggested by Experiment 1, the rime unit appears to be a particularly instrumental unit in English word recognition. The second aim of Experiment 2 is a derivative of the rst aim. If the results found were to support either the sonority distance or sonority contour principle, there would be reason to suggest that either lexical representation may be organized according to properties related to sonority and/or that such properties may automatically come into play in the process of lexical access. Method
Participants. Forty-six monolingual English speakers, all of whom were undergraduate students at the University of New South Wales, and 45 native Filipino speakers residing in the Philippines participated in the present experiment. The monolingual English speakers were given course credit in return

378

ALONZO AND TAFT

TABLE 3 Decision Latencies (in Milliseconds) and Errors (in Percentages in Parentheses) of Native English and Native Filipino Speakers for Words Beginning with sStop, sSonorant, and StopLiquid Onset Clusters
Conditions sStop C-C OnsetRime Antibody sSonorant C-C OnsetRime Antibody StopLiquid C-C OnsetRime Antibody English 710 (8) 710 (5) 730 (6) 682 (6) 725 (4) 707 (6) 682 (3) 669 (2) 686 (3) Filipino 783 (4) 779 (2) 808 (4) 764 (6) 791 (4) 845 (4) 757 (2) 733 (2) 748 (1)

for their participation. As in Experiment 1, all Filipino participants were university educated, bilingual FilipinoEnglish speakers. Participants in Experiment 2 did not participate in Experiment 1. Stimuli and design. A 2 3 4 factorial design was employed with the between-participants factor being native language background (native English or native Filipino) and the within-participants factors being the type of intraword division (C-C, OnsetRime, and Antibody) and the type of onset cluster (s- clusters following a sonority contour, s- clusters violating a sonority contour, and stopliquid clusters). Eighty-one monosyllabic English words, and 81 monosyllabic nonwords, all beginning with a complex onset, were selected as experimental items. Of each subset of word and nonword items, 27 items began with s- clusters following a sonority contour, 27 began with s- clusters violating a sonority contour, and 27 began with a stopliquid sequence. The design of the experiment was such that each participant was presented with all 162 word and nonword items. Of these, 27 words were divided in between their rst and second consonants (e.g., S WAP), 27 were divided at their onsetrime boundary (e.g., SM ACK), and 27 were divided at their antibody boundary (e.g., SNA KE).3 The nonword items were divided in exactly the same way. Thus, each participant was presented with items divided according to each type of intraword division but did not see any particular item more than once. Further, participants within each language group were divided into three groups such that while all groups were presented with the same items, the way in which the items were divided varied between the three groups.4 For example, while one group was presented with the items S WAP, SM ACK, and SNA KE, a second group was presented with S MACK, SN AKE, and SWA P, and a third group was presented with S NAKE, SW AP, and SMA CK. In addition to the experimental items, participants were presented with nine practice items and two ller items prior to the commencement of the experiment. Apparatus. Same as in Experiment 1. Procedure. Same as in Experiment 1.

Results Four participants from the native English group and one participant from the native Filipino group were replaced due to overall error rates exceeding 30%. One word item (SCOLD) was also discarded from the analyses due to an excessively high error rate in one group. Mean RTs and ERs averaged across participants for each language group are presented in Table 3. Separate ANOVAs, treating participants (Fp) and items (Fi) as random variables, were performed for the RT and ER data.
3 See footnote 1. In addition, note that instead of two slash symbols dividing the items as in Experiment 1, items in Experiment 2 were divided by a space as it was thought that there was a chance that participants could confuse the two slash symbols as the letters L or H. 4 See footnote 2.

SONORITY CONSTRAINTS

379

The analysis of RTs revealed signicant main effects of native language [Fp(1, 85) 11.03, p .05; Fi(1, 154) 74.83, p .05] and intraword division. There was a signicant difference between the C-C and Antibody conditions [Fp(1, 85) 23.04, p .05; Fi(1, 154) 16.12, p .05] and between the OnsetRime and Antibody conditions [Fp(1, 85) 13.56, p .05; Fi(1, 154) 13.72, p .05], but the difference between the C-C and OnsetRime conditions was not signicant [Fp(1, 85) 1.55, p .05; Fi(1, 154) 0.63, p .05]. There were also no signicant interactions between native language and intraword division. There were, however, a number of signicant interactions between the two factors of intraword division and onset cluster type. In summary, these interactions may be grouped into two categories. The rst group of interactions resulted from differences in performance for words beginning with s onsets compared to words beginning with stopliquid onsets, while the second group resulted from differences in performance for words beginning with s onsets which follow a sonority contour compared to words beginning with s onsets that violate a sonority contour. Regarding the rst group of interactions, a difference was found between the C-C and Antibody conditions for words beginning with s onsets, but for words beginning with stop liquid onsets, there was no such difference [Fp(1, 85) 11.15, p .05; Fi(1, 154) 9.83, p .05]. Also, it was found that the C-C condition was faster than the Onset Rime condition for words beginning with s onsets while for words beginning with stopliquid onsets, the OnsetRime condition was faster than the C-C condition [Fp(1, 85) 13.34, p .05; F i(1, 154) 7.3, p .05]. Turning to the second group of interactions, there was a difference between the C-C and OnsetRime conditions for words beginning with s onsets that follow a sonority contour, while for words beginning with s onsets which violate a sonority contour, there was no such difference [Fp(1, 85) 4.92, p .05; Fi(1, 154) 6.07, p .05]. Also, the difference between the C-C and Antibody conditions was greater for words beginning with s onsets that follow a sonority contour compared to words beginning with s onsets which violate a sonority contour [Fp(1, 85) 6.06, p .05; Fi(1, 154) 4.2, p .05]. In the analysis of ERs, there was a main effect of language group only in the participant analysis [Fp(1, 85) 7.3, p .05; Fi(1, 154) 0.56, p .05], and there was only one main effect of intraword division. There was a signicant difference between the C-C and OnsetRime conditions [Fp(1, 85) 5.34, p .05; Fi(1, 154) 6.8, p .05], but the difference between the OnsetRime and Antibody conditions [Fp(1, 85) 3.16, p .05; Fi(1, 188) 2.59, p .05) and between the C-C and Antibody conditions was not signicant [Fp(1, 85) 1.55, p .05; Fi(1, 154) 0.76, p .05]. Like the analysis of RTs, there were no signicant interactions between native language and intraword division with all F values being less than 1. Only one signicant interaction involving intraword division and onset cluster type was found, although it reached signicance only in the participant analysis. There was a difference between the C-C and OnsetRime conditions for words beginning with s onsets while for words beginning with stopliquid onsets, there was no such difference [Fp(1, 85) 5.24, p .05; Fi(1, 154) 2.82, p .05]. No other interactions that were found to be signicant in the RT analysis reached signicance in the ER analysis. Discussion The results from Experiment 2 show that although the average RT of the native English group was faster than that of the native Filipino group, the two language groups did not differ in their preferences for how words were divided. Words were

380

ALONZO AND TAFT

recognized faster either when they were divided at their onsetrime boundaries or when their onset clusters were split compared to when they were divided at their antibody boundaries. Further, dividing the words at their onsetrime boundaries did not provide a greater advantage to word recognition compared to splitting their onset clusters. Thus, the results suggest that while the onsetrime distinction was utilized to some extent by both language groups, as indicated by the signicantly faster RTs for the OnsetRime condition compared to the Antibody condition, it was not the only optimal way in which word recognition was facilitated. Compatible with the results from Experiment 1, that there was no signicant difference between the C-C and OnsetRime conditions, suggests that while the onset unit may not be a particularly cohesive unit, the rime unit may be more instrumental in the processing of English words, since splitting the rime (i.e., the Antibody condition) produced the slowest decision latencies, whereas splitting the onset (i.e., the C-C condition) did not interfere with word recognition relative to preserving the onset (i.e., the OnsetRime condition). Taken together, the results from Experiments 1 and 2 offer evidence in support of the involvement of rime units in the way English words are accessed and represented. As proposed by previous studies (e.g., Taft, 1994; Treiman, 1992; Treiman & Zukowski, 1991), rime units may be represented in the lexicon at a level that is intermediate between the layers of representation for syllabic and graphemic/phonemic units. That the native Filipino group performed similarly to the native English group namely that the Antibody condition was the slowest condition for both language groupsalso suggests that a sensitivity to rime units is an adaptation that the native Filipino group had made in their processing of English words. Moreover, as the Antibody condition hindered the performance of a nonnative group as much as it did that of the native English group, there is strong evidence in support of the conception of rime units as being particularly important when processing a language such as English, which has a relatively opaque orthography. That is, because rime units have been shown to preserve some degree of consistency in the spelling-to-sound decoding of English words, sensitivity to such units appears to be particularly important to the efcient processing of English words that even readers of a non-English background, such as native Filipino readers, indicate an awareness of rime units comparable to that of the native English group. Furthermore, as divisions at the antibody boundary, which corresponds to a reduplicated sequence utilized by native Filipino speakers, produced the slowest decision latencies for both the native Filipino and native English groups, there is again no evidence of the reduplication process having specically inuenced the performance of the native Filipino group. If the reduplication process were to have inuenced the English word recognition of the native Filipino group, the Antibody condition should not have hindered decision latencies relative to the C-C and OnsetRime conditions. Rather, decision latencies for the Antibody condition should have been faster than those for the OnsetRime, if not the C-C, condition. Regarding the inxation process, as dividing the words at their onsetrime boundaries compared to splitting their onset clusters, did not differentially affect the two language groups, there is also no evidence to suggest that the inxation process inuenced the English word recognition of the native Filipino group. Had it done so, at the very least, a greater difference in favor of the C-C condition over the OnsetRime condition would have been expected for the native Filipino group compared to the native English group. However, that both language groups equally preferred the C-C and OnsetRime conditions refutes any assertion that onset clusters may be less cohesive units for native Filipino speakers due to the feature of inxation in their native language. The pattern of results for the two language groups also suggests that the factor of

SONORITY CONSTRAINTS

381

sonority may be comparably involved in the lexical processing of both language groups. Lending some support for the sonority distance principle, which would imply that onset clusters beginning with s should be processed differently compared to onset clusters beginning with a stopliquid sequence, signicant interactions were indeed found which suggest that s onset clusters were treated differently to stop liquid onset clusters. Together, these interactions suggest that onset clusters beginning with s were less cohesive than onset clusters composed of a stopliquid sequence. Words beginning with s onsets were recognized faster when their onset units were split compared to when they were divided at their antibody boundaries while for words beginning with stopliquid onsets, splitting their onset units did not facilitate word recognition relative to dividing the words at their antibody boundaries. In addition, words beginning with s onsets were recognized faster when their onset units were split compared to when they were divided at their onsetrime boundaries while for words beginning with stopliquid onsets, recognition was facilitated when they were divided at their onsetrime boundaries compared to when their onset units were split. Such results support the sonority distance principle, which would maintain that because the distance along the sonority continuum between s and all the consonants which can possibly follow it at word onset position is always less than four, onset clusters beginning with s should not function as cohesive units. That onset clusters comprised of a stopliquid sequence were recognized faster when they were divided at their onsetrime boundaries compared to either when their onset units were split or when they were divided at their antibody boundaries suggests that stopliquid onset clusters functioned as cohesive onset units and lends further support to the sonority distance principle. In showing that stopliquid onset clusters were more cohesive units compared to s onset clusters, the results conrm predictions stemming from the sonority distance principle that because stop and liquid consonants are separated by a minimum of at least four positions along the sonority continuum, stopliquid onset clusters should function as cohesive onset units. In contrast to the evidence in support of the sonority distance principle, the results provided no support for the sonority contour principle. Contrary to what would be expected according to the sonority contour principle, performances for words with onset clusters which followed a steady rise in sonority from the initial consonant through to the vowel did not show a particular sensitivity to onsetrime units. That there were no signicant interactions when comparing performances for onset clusters which violate the sonority contour (i.e., sstop onsets) to performances for onset clusters which follow the sonority contour (i.e., ssonorant and stopliquid onsets) suggests that the factor of whether onset clusters follow a steady rise in sonority from the initial consonant through to the vowel did not differentially affect word recognition performance. Specically, there is no evidence to suggest that onset clusters which follow the sonority contour function as more cohesive onset units compared to onset clusters which violate the sonority contour. Indeed, the nding that the onsetrime condition, relative to the C-C condition, facilitated word recognition for only stopliquid onsets is more compatible with predictions derived from the sonority distance principle. However, the two interactions found indicating differences in performance between ssonorant onsets and sstop onsets cannot be readily accounted for by the sonority distance principle. Together, these interactions suggest that s sonorant onsets were even less cohesive than sstop onsets. First, for words beginning with ssonorant onsets, recognition was faster when the onset clusters were split than when the words were divided at their onsetrime boundaries, whereas there was no such difference for words beginning with sstop onsets. Second, splitting the onset cluster relative to dividing the words at their antibody boundaries facilitated

382

ALONZO AND TAFT

recognition to a greater extent for words beginning with ssonorant onsets compared to words beginning with sstop onsets. These results are incompatible with the sonority distance principle, which would maintain that there should be no differences in performance for ssonorant and sstop onsets. That differences did exist suggests that factors in addition to the difference in sonority between the two constituents of an onset cluster may inuence word recognition. In the case of ssonorant onsets, it is possible that the more sonorous second consonant may be especially cohesive with the following vowel, a result that has some support from Fowler (1987) using a phoneme shifting task. Requiring participants to exchange specic phonemes in two visually presented words, and then pronounce the two resulting words, Fowler (1987) found that more sonorous consonants took longer to shift than less sonorous consonants regardless of whether they were prevocalic or postvocalic consonants. Thus, the nding that splitting the onset relative to dividing words at their onsetrime boundaries facilitated recognition only in the case of the ssonorant onsets, and not the sstop onsets, could be explained by the second consonant of the ssonorant onsets (i.e., l, m, n, and w) being higher in sonority than the rst consonant and therefore having a greater tendency to be attracted to the following vowel compared to the stop consonants (i.e., k, p, and t) of the sstop onsets. Notwithstanding the difference in performance for the ssonorant onsets, the results of Experiment 2 do indeed suggest that despite properties in their native language which appear to be incongruent with the onsetrime division, native Filipino speakers processed English words similarly to native English speakers. For both language groups, the rime appeared to be a particularly salient unit in their processing of English words, whereas the onset in general did not appear to be as instrumental. English word recognition for both language groups was inhibited when the rime unit was split, whereas splitting the onset unit did not interfere with the recognition process. In addition, the present results also suggest that properties related to sonority, specically the difference or the distance in sonority between two consonants occurring at word onset position, may inuence the likelihood of such sequences functioning as cohesive onset units. Furthermore, in using what is essentially a phonologically based theory to explain the results of a visual word recognition task, there is the implication that phonological properties may be involved in the word recognition process. However, it is beyond the scope of the present study to identify the locus of these phonologically based effects. Such effects may conceivably arise from processes prior to or after lexical access or may simply reect the way in which words are organized in the mental lexicon.5 Future studies which specically isolate prelexi5 As noted above, it is no small issue that the explanations put forward for the present results are essentially based on phonological theory. It could perhaps be argued that orthographic factors such as bigram frequencies may have been the bases of the present results. However, bigram frequencies were indeed calculated for the items in the present study and it was found that bigram patterns were not congruent with the present results. For words beginning with stopliquid clusters, the lowest bigram frequencies existed at their initial C-C boundaries, whereas for words beginning with s clusters, the lowest bigram frequencies existed at their onsetrime boundaries. Such a pattern suggests that stop liquid clusters should actually have been less cohesive than s clusters, a conjecture that is not supported by the present results. In addition, it may be reasoned that an analysis of the nonword items used in the present study could have provided conclusive proof that the effects found in the analysis of word items were indeed phonologically based. However, the view held here is that an absence of congruent effects for nonword items would not necessarily mean that the effects found for the word items were not phonologically based, as there is the possibility that the effects found for word items may have arisen from postlexical processes. Thus if such a case were to hold true, there would be no strong reason to expect nonword processing to indicate sonority-based effects.

SONORITY CONSTRAINTS

383

cal or postlexical components in the word recognition process may provide evidence to suggest where the effects identied in the present study, that is the sonority-based effects, exert their inuence in reading.
REFERENCES
Berndt, R. S., Reggia, J. A., & Mitchum, C. C. (1987). Empirically derived probabilities for graphemeto-phoneme correspondences in English. Behavior Research Methods, Instruments & Computers, 19, 19. Cardoso-Martins, C. (1995). Sensitivity to rhymes, syllables, and phonemes in literacy acquisition in Portuguese. Reading Research Quarterly, 30, 808828. Coltheart, M., Davelaar, E., Jonasson, J. T., & Besner, D. (1977). Access to the internal lexicon. In S. Dornic (Ed.), Attention and performance VI. London: Academic Press. Davis, S. (1990). Italian onset structure and the distribution of il and lo. Linguistics, 28, 4355. Fowler, C. A. (1987). Consonant-vowel cohesiveness in speech production as revealed by initial and nal consonant exchanges. Speech Communication, 6, 231244. Fowler, C. A., Treiman, R., & Gross, J. (1993). The structure of English syllables and polysyllables. Journal of Memory and Language, 32, 115140. Goswami, U., Gombert, J. E., & de Barrera, L. F. (1998). Childrens orthographic representations and linguistic transparency: Nonsense word reading in English, French, and Spanish. Applied Psycholinguistics, 19, 1952. Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual word recognition: A multiple read-out model. Psychological Review, 103, 518565. Kaye, J., Lowenstamm, J., & Vergnaud, J.-R. (1990). Constituent structure and government in phonology. Phonology, 7, 193231. Kessler, B., & Treiman, R. (1997). Syllable structure and the distribution of phonemes in English syllables. Journal of Memory and Language, 37, 295311. Stemberger, J. P., & Treiman, R. (1986). The internal structure of word-initial consonant clusters. Journal of Memory and Language, 25, 163180. Taft, M. (1994). Interactive-activation as a framework for understanding morphological processing. Language and Cognitive Processes, 9, 271294. Treiman, R. (1992). The role of intrasyllabic units in learning to read and spell. In P. B. Gough, L. Ehri, & R. Treiman (Eds.), Reading acquisition (pp. 65106). Hillsdale, NJ: Erlbaum. Treiman, R., Gross, J., & Cwikiel-Glavin, A. (1992). The syllabication of /s/ clusters in English. Journal of Phonetics, 20, 383402. Treiman, R., Mullennix, J., Bijeljac-Babic, R., & Richmond-Welty, E. D. (1995). The special role of rimes in the description, use, and acquisition of English orthography. Journal of Experimental Psychology: General, 124, 107136. Treiman, R., & Zukowski, A. (1991). Levels of phonological awareness. In S. A. Brady & D. P. Shankweiler (Eds.), Phonological processes in literacy (pp. 6783). Hillsdale, NJ: Erlbaum. Wimmer, H., & Goswami, U. (1994). The inuence of orthographic consistency on reading development: Word recognition in English and German children. Cognition, 51, 91103. Ziegler, J. C., & Perry, C. (1998). No more problems in Colthearts neighborhood: Resolving neighbourhood conicts in the lexical decision task. Cognition, 68, 5362.

Vous aimerez peut-être aussi