JOURNAL OF MEMORY AND LANGUAGE 27, 87-104 (1988) Syllabification of Intervocalic Consonants REBECCA TREIMAN AND CATALINA DANIS Wayne State University Two tasks were used to study the syllabification of intervocalic consonants like the /V’s of melon and collide. In an oral task, subjects reversed the syllables in words; in a written task, they selected between alternative syllabifications. Even in the oral task, subjects’ responses were influenced by whether their spellings of the words contained a single letter (r) or a double letter (10.Responses in the two tasks were also affected by the stress pattern of the word, the phonetic category of the intervocalic consonant, and the nature of the preceding vowel. The results are discussed in relation to theories of syllabification. o 1988 Academic Press, Inc. Many investigators (e.g., Spoehr, 1981; Segui, 1984) have proposed that the syllable plays a role in the processing of spoken and written language. However, studies of the syllable’s role in language processing have been hindered by uncertainty about how words are divided into syllables. With a word like demon, for instance, is the /m/ part of the first syllable, part of the second syllable, or part of both syllables? Different theories of syllabification offer different answers. In this paper, we ask how adults syllabify spoken and written words. We focus on intervocalic consonants like the /ml of demon and the /r/ of erase, asking what factors affect their syllabification. Within linguistics, no one theory of syllabification is widely accepted. While competing theories disagree on many points, there are three general areas of agreement. The first assumption, made by all linguistic theories of which we are aware, is that every syllable contains a vowel or syllabic consonant (as in the second syllable of battle). Although this assumption produces unambiguous syllabifications for words like Joey, it does not indicate whether the intervocalic consonants of demon and erase belong to the first syllable or the second syllable. A second assumption, held by Anderson and Jones (1974), Hooper (1972), Fulgram (1970), and Selkirk (1982), but not always by Bailey (1978), Hoard (1971), or Kahn (1976), is that if a consonant or sequence of consonants is illegal at the beginnings of words it is also illegal at the beginnings of syllables. The same holds for word-final and syllable-final consonants. For example, /I-$ is said to be illegal syllable-initially because it is illegal word-initially. Thus, the /rt/ offinger is assigned to the first syllable rather than to the second. Our experiments used only intervocalic consonants that were legal both in wordinitial and in word-final positions, so this assumption did not determine syllabifications. A final assumption concerns syllabification before stressed vowels. Many linguists (Bailey, 1978; Hoard, 1971; Hooper, 1972; Kahn, 1976; Pulgram, 1970; Selkirk, 1982) assume that as many consonants as possible are placed at the beginning of a stressed syllable. In other words, the onset (initial consonant or consonant cluster) of This research was supported by NICHD Grants HD20276 and HD00769. We thank Peggy Ericson, Cathy Beiser, and Andrea Zukowski for their contributions and Anne Cutler, Juan Segui, Mark Seidenberg, and Marcus lhft for comments on a draft of the manuscript. Reprint requests should be addressed to Dr. Rebecca Treiman, Department of Psychology, Wayne State University, 71 W. Warren Ave., Detroit, MI 48202. 87 0749-596X/88 $3.00 Copyright All rights 8 1988 by Academic Press. Inc. of reproduction in any form reserved. 88 TREIMAN the stressed syllable is maximized.’ By this assumption, a VCV’ sequence (where V = unstressed vowel, C = consonant that is possible syllable-initially, and V’ = stressed vowel) is syllabified as V-CV’. For example, /r/ belongs to the second syllable of erase. This third assumption does not specify the place of/m/ in demon since the second vowel of this word is unstressed. Although most linguistic theories agree that VCV’ sequences are syllabified as V-CV’, linguists disagree on the treatment of V’CVs. According to Pulgram (1970), the nature of the first vowel determines syllabification in this case. If the vowel is one that can end a word-often called a “long” vowel-the consonant is assigned to the second syllable. If the first vowel is “short”, or illegal in word-final position, the consonant is assigned to the first syllable. Thus, the /m/ of demon belongs to the second syllable and the /m/ of lemon belongs to the first. A similar view is apparent in the syllabifications of spoken words given by Webster’s New World Dictionary (1984).* Hoard (1971) has a different view. He states that all V’CVs are syllabified as V’C-V. Thus, both demon and lemon are broken after the /ml. The theories of Kahn (1976) and Selkirk (1982) differ from those described above in that each proposes two levels or types of syllabification. To Kahn ( 1976), syllabification in slow speech differs from that in normal and fast speech. In slow speech, the onsets of all syllables are as long as possible given the phonological constraints of ’ This holds as long as there are no conflicting morphological boundaries. Theories disagree on the treatment of words like racetrack, where phonological syllabification via the maximization of onsets before stressed vowels yields a boundary between /e/ and Is/ and morphological syllabification yields a boundary between Is1 and It/. We did not use words that presented such a conflict. t The dictionary is not completely consistent, however. For example, the it/ of mofor is placed in the first syllable. AND DANIS English. Thus, the /m/‘s of demote and lemon both belong to the second syllable. In normal rate and fast speech, the /m/ of demote remains the initial consonant of the second syllable. However, the /ml of lemon attaches itself to the first syllable as well. It becomes simultaneously syllable-final and syllable-initial, or ambisyllubic. To Selkirk (1982), the two levels of syllabification are the level of basic syllable composition-an underlying or deep level-and a surface or phonetic level. At the level of basic syllable composition, the /ml’s of demote and lemon are both syllable-initial, as for Kahn. At the surface level, the /m/ of demote remains syllable-initial while the /m/ of lemon resyllabifies to become syllable-final. In addition to the linguistic theories, which were designed to describe the syllabification of spoken words, several theories of word recognition and reading include a role for syllable units. One such theory is that of the vocalic center group or VCG, proposed by Hansen and Rodgers (1968) and later taken up by Spoehr and Smith (1973). According to Hansen and Rodgers, a vowel-consonant-vowel sequence is first parsed after the vowel. Spellingsound rules are then applied in an attempt to derive a phonological representation. If the obtained phonological representation matches a representation in long-term memory, the VCG boundary is after the first vowel. If no match is obtained, the sequence is parsed after the consonant. According to Hansen and Rodgers, the initial V-CV parse will be successful for words like demon which have “long” vowels in the first syllable. Such a parsing will not yield a match for lemon since the spellingsound rules assign the pronunciation /i/ to the letter e. Lemon is therefore parsed into the VCGs lem and on. A second theory of word recognition that includes a role for syllable-like units is that of Taft (1979). Taft suggests that printed words are accessed in terms of units called BOSSes. A BOSS includes “as many con- SYLLABIFICATION OF INTERVOCALIC CONSONANTS 89 sonants following the first vowel of the consonants. Stemberger showed that consonants in V’CV environments-consoword as orthotactic factors will allow nants classified by Kahn (1976) as ambisylwithout disrupting the morphological struclabic-interacted with both syllable-final ture of that word” (Taft, 1979, p. 24). In consonants and syllable-initial consonants. this view, Canada, canary, and canine all He interpreted this result to suggest that have the BOSS cm. The stress pattern of consonants like the ltl of water are ambithe word and the nature of the first vowel syllabic. (See Buckingham (1980) for simhave no effect. Finally, it has been proposed that syl- ilar evidence from aphasics’ errors.) The psychological theories of syllabic lable units in written words can be defined units in written words have produced more orthographically (Adams, 1981; Seidenempirical research. Taft (1979) compared berg, 1987). Specifically, “syllable boundaries will be located at points where the as- BOSS units and VCG units, finding that sociations between adjacent letters are words divided into BOSS units yielded weakest” (Adams, 1981, p. 215). In anvil, faster response times in a lexical decision for example, the bigram nv has a lower fre- task than words divided into VCGs. Howquency than the bigrams an, vi, or il. In ever, Lima and Pollatsek (1983) failed to replicate this result. Prinzmetal, Treiman, Seidenberg’s terms, there is a bigram trough at the nv boundary. The syllable and Rho (1986) and Seidenberg (1987) used boundary corresponds to this trough. feature integration errors to diagnose perThe linguistic theories of syllabification ceptual groupings of letters. This task prohave not led to a great deal of behavioral duced evidence for orthographically deresearch. One exception is the study by fined units and against BOSSes. Fallows (198 1). Fallows presented children The linguistic work and the psychologaged 4-5 and 9-10 with spoken words. ical work on syllabification have been relaThe children were to double either the first tively isolated from one another. We sought or the second syllable of the words. For ex- to draw these lines of work closer together ample, in the first task a child could say by considering both factors that linguists over as o-over or as ov-over; in the second have identified as important, such as task a child could say over-ver or over-er. stress, and factors that psychologists have A potentially serious problem with identified as important, such as bigram freFallows’ study is that some of the words quencies. We used two tasks to study sylwere spelled with a single consonant (e.g., labification. The first was a syllable reover) and some with a double consonant versal task involving spoken words. After (e.g., bunny). Fallows did not analyze the practicing on words like grandfather, two types of stimuli separately, nor did she which was reversed to form father grand, assess children’s knowledge of the subjects were asked to reverse words like spellings. Even in an oral task, people’s re- lemon. The possible answers were onlem sponses may be affected by how they spell (where the critical phoneme, /m/, is placed the words (Ehri 8z Wilce, 1986; Jakimik, in the first syllable), monk (/m/ placed in Cole, & Rudnicky, 1985; Seidenberg & the second syllable), or monlem (/m/ placed Tanenhaus, 1979). A second behavioral in both syllables). After performing this study (Stemberger, 1983a) examined task, subjects were asked to spell the speech errors. A common finding is that words. A second task, run with other subsyllable-initial consonants tend to interact jects, used written words. Subjects saw a with other syllable-initial consonants in word either with a slash before the letter or speech errors, while syllable-final consoletters corresponding to the critical nants generally interact with syllable-final phoneme (e.g., lelmon, colmma), or with a 90 TREIMANANDDANIS slash after (lemlon, commla), and circled their preferred syllabification. EXPERIMENT Method Procedure 1 Stimuli There were a total of 84 stimuli that varied in stress (first syllable stressed or second syllable stressed) and spelling (critical phoneme spelled with a single letter or a double letter). As shown in Table 1, the stimuli were designed in pairs like proper, propel and comma, command. Within a pair, the words differed in stress but were similar in spelling. The spellings were the same in the critical letter or letters, the preceding vowel, and (with two exceptions) the following vowel. The letters preceding and following these vowels were often shared as well. The words in the four categories were closely equated for frequency (Kucera and Francis, 1967), number of letters in spelling, and number of syllables. They ranged from two to five syllables in length. Table 1 shows the mean frequencies for the bigrams surrounding the critical letter(s) for the stimuli in each category. The figures from Mayzner and Tresselt (1965) are total bigram frequencies; the figures from Solso and Juel(198 1) are positional bigram frequencies for words of various lengths.3 A complete list of stimuli for this and subsequent experiments appears in the Appendix. For the oral task, the stimuli were spoken aloud by the experimenter. The order of the stimuli was randomly chosen for each subject. For the written task, each stimulus was typed in two ways on a single line-with a slash before the critical letter or letters and with a slash after. For example, the subject saw pro/per and propler and colmmand and commland and chose the better syllabification in each case. Oral task. The experimenter said, “We’re going to play a game with words. I’ll show you how it goes by giving you some examples. When I say grandfather you szty father [pause] grand. When I say catfood you say food [pause] cat.” The experimenter then asked the subject to give the responses for snowman, jawbreaker, jetliner, kidnap, and ungrateful. The correct response, of course, was to produce the second (and any following) syllables, pause, and then produce the first syllable. For these practice items, the first syllable corresponded to a morpheme. The experimenter did not use the term “syllable.” Any errors on practice items were corrected. The test items were then presented. The experimenter did not comment on the subject’s responses unless the subject failed to divide a word of three or more syllables at the boundary between the first and second syllables, saying for example tude atti for attitude. If this occurred, the experimenter referred back to the practice item ungrateful, reminding the subject that the correct answer was grateful un rather than ful ungrate. The experimenter then asked the subject for another answer. After performing the syllable reversal task, subjects were asked to spell all the words. Written task. Subjects were asked to read each word aloud; the experimenter corrected any mispronunciations. For each Subjects word, subjects were asked how they would syllabify it in print. They were to circle the Twenty students from Indiana Univerbetter syllabification of the two choices provided. After completing the task, sub3 Solso and Juel (1981) consider two- through ninejects were asked on what basis they had letter words only. Some of the stimuli for Experiment syllabified the words and what rules of syl- 1 were longer; these are excluded from the reported means. labification they had been taught. SYLLABIFICATION OF INTERVOCALIC 91 CONSONANTS TABLE 1 SAMPLE S~MULI FOR EXPERIMENT 1 AND MEAN FREQUENCIES FOR BIGRAMS SURROUNDING CRITICAL LE-ITER(S) ACCORDING TO MAYZNER AND TFUXSSELT (1%5) (MT) AND SOLSO AND JUEL (1981) (SJ) Bigram Stress Spelling Example 1 I proper 1 2 propel 2 1 comma 2 2 command Note. Critical letter(s) is italicized MT: SJ: MT: SJ: MT: SJ: MT: SJ: and lowest frequencies Prior bigram VC cc cv 585 1762 585 1733 272 1973 272 1863 308* 771 308* 655 397 555 397 542 133* 383* 133* 433 383 642* 396 504* 304 614 308 391* bigram frequency is marked with Subsequent bigram 537 1431 412 789 445 1044 497 655 *. ried out both across subjects and across stimuli; only those effects that reached at least the .05 level in both types of analyses will be reported as significant. Post hoc analyses were carried out using Scheffe tests. Table 2 shows the mean proportion of responses of each type for the four categories of stimuli. Friedman analyses of variance Results and Discussion by ranks showed that the four types of responses differed in frequency for each Responses in the oral task were scored stimulus category CJJ< .OOl). Multiple comas 1 responses if the critical phoneme was parisons were carried out on the rank sums placed in the first syllable (as in er prop for as recommended by Hollander and Wolfe proper; and corn for command), 2 re- (1973, p. 151). The procedure is quite consponses if the phoneme was placed in the servative since it tests for the six possible second syllable (per pro; mand co), or l-2 pairwise differences simultaneously. When responses if the phoneme was placed in the first syllable was stressed, 1 and 2 reboth syllables (per prop; mand corn). sponses did not differ significantly. When “Other” responses were those that did not the second syllable was stressed, 2 refit any of these categories (er pro; ommand sponses significantly outnumbered 1 reco). Each subject’s data were adjusted ac- sponses. Also, 1-2 responses only outcording to how that subject spelled the numbered Other responses when the critstimuli. For example, if a subject spelled ical consonant was spelled with two letters. command with one m rather than two, To further study the effects of spelling command was placed in the category of and stress on subjects’ responses, analyses stimuli with second syllable stress and of variance were carried out. Two depensingle letter spellings for that subject. All dent variables were used. The first was the the results to be reported in this and subse- proportion of l-2 responses, that is, how quent experiments are based on the ad- often subjects placed the critical consonant justed data. Statistical analyses were car- in both syllables. The second was the prosity, Bloomington, participated in the oral task and 24 students from Wayne State University participated in the written task. The students in this and subsequent experiments received course credit or pay in return for participation; all were native speakers of English. TREIMAN 92 AND DANIS TABLE 2 RESULTS OF ORAL AND WRMTEN TASKS OF EXPERIMENT 1 Written task Oral task Mean proportion responses 1 vs 2 l-2 vs Other Prop. I responses Diff. from .50 by two-tailed t tests n.s. 2>1 n.s. 2>1 n.s. n.s. 1-2 > Other l-2 > Other .60 .14 .68 .42 ns. C.50 n.s. n.s. Multiple comparisons based on rank sums Spelling Stress 1 2 l-2 Other 1 1 2 2 1 2 1 2 .34 .08 .15 .04 .50 .87 .24 .50 .12 .03 .60 .45 .04 .02 .Ol .Ol portion of 1 responses relative to the total To summarize the results of the oral task, of 1 and 2 responses, that is, how often both spelling and stress affected perforsubjects placed the consonant in the first mance. With respect to spelling, subjects syllable given that they chose to put it in were more likely to place the critical in both syllables when they one syllable only.4 Thus defined, the two phoneme spelled it with a double letter than when variables are logically and empirically indethey spelled it with a single letter. Subjects’ pendent. responses were also affected by a phonoThe proportion of l-2 responses varied logical variable, stress. When the second as a function of spelling (F( 1,19) = 37.80 in syllable was stressed, subjects placed the the analysis by subjects; F(1,40) = 297.01 consonant in the second syllable signifiin the analysis by items; p < .OOl for both) cantly more often than they placed it in the and stress (F(1,19) = 17.57 by subjects; F(1,40) = 14.95 by items; p < .OOl for first syllable. When the first syllable was both). Subjects were more likely to place stressed, subjects showed no reliable preference between first syllable responses and the consonant in both syllables when they spelled it with a double letter than when second syllable responses. Also, subjects they spelled it with a single letter. Also, more often placed the consonant in both they were more likely to produce 1-2 re- syllables when the first syllable was sponses when the first syllable was stressed than when the second syllable was stressed than when the second syllable was stressed. The results of the written task are shown stressed. The proportion of 1 responses relative to on the right side of Table 2. Two-tailed t 1 responses plus 2 responses varied as a tests showed that when the critical consofunction of stress (F(1,19) = 29.44 in the nant was spelled with a single letter subanalysis by subjects; F(1,40) = 85.03 in the jects had no reliable syllabification preferanalysis by items; p < .OOl for both). When ence when the first vowel was stressed but subjects placed the critical consonant in generally placed the consonant in the just one syllable, they were more likely to second syllable when the second vowel choose the first syllable for stimuli with was stressed. Subjects showed no reliable first syllable stress than for stimuli with preferences when the critical consonant was spelled with two letters. second syllable stress. Analyses of variance showed a main effect of stress (F(1,23) = 76.19 by subjects; 4 If the total of 1 and 2 responses was 0, the mean F(1,40) = 102.50 by items; p < .OOl for both). As in the oral task, the proportion of value was used. SYLLABIFICATIONOFINTERVOCALICCONSONANTS 93 first syllable responses was higher when consonant. Using the two tasks of Experithe first syllable was stressed than when ment 1, we compared performance on the second syllable was stressed. In addi- words with medial liquids, nasals, and obstruents (which include stops and frication, there was a main effect of spelling research using speech (F(1,23) = 38.26 in the analysis by sub- tives). Previous jects; F(1,40) = 20.88 in the analysis by errors (MacKay, 1978; Shattuck-Hufnagel, items; p < .OOl for both) and an interaction 1986; Stemberger, 1983b), word games between spelling and stress (F(1,23) = (Treiman, 1984), and short-term memory 35.96 by subjects, p < .OOl; F(1,40) = 5.86 errors (Treiman & Danis, in press) suggests by items, p < .05). The effect of spelling that liquids have the closest association arose because subjects were more likely to with the preceding vowel. This leads us to place the critical consonant in the first syl- expect that the il! of melon would be placed lable when it was spelled with two letters in the first syllable more often than the /v/ than when it was spelled with one letter. of seven. Experiment 2 also manipulated For two letter spellings, divisions like the factors of stress and spelling in an atcommla and commland produce orthotempt to replicate the findings of Experigraphically legal first portions, in the sense ment 1. that words can end with double consonants. Divisions like colmma and co/ Method mmand are illegal, in that words hardly ever begin with double consonants. For Procedure one letter spellings, there is usually no such The procedure for the oral task was simdifference in orthographic legality. The inilar to that of Experiment 1. However, teraction between spelling and stress arose since all the test stimuli were bisyllabic because the effect of stress was smaller, alonly two-syllable practice items were used. though still significant, when the critical The experimenter said each stimulus in a consonant was spelled with two letters than phrase or sentence before asking the subwith one. People have some difficulty ject to respond. The procedure for the choosing between syllabifications like written task was identical to that of Expericommla and colmma, neither of which corment 1. responds to their preferred syllabification comlma. The results of the written syllabification test cannot be predicted based on the bigram frequencies of the stimuli. For example, subjects treated words like proper and propel differently but these two types of words had the same patterns of bigram frequencies. The implications of the results for various theories of syllabification will be further discussed under General Discussion. EXPERIMENTS Experiment 2 examined another variable that may affect subjects’ syllabificationsthe phonetic category of the intervocalic Stimuli The stimuli for Experiment 2 varied in stress (first syllable stressed or second syllable stressed), spelling (critical consonant spelled with single letter or double letter), and type of critical consonant (liquid, nasal, or obstruent). There were 10 twosyllable words of each of the 12 types. Examples, along with the mean bigram frequencies for the bigrams surrounding the critical letter(s), are shown in Table 3. Subjects Twenty Wayne State University students participated in the oral task and 24 students participated in the written task. One sub- 94 TREIMAN AND DANIS TABLE 3 SAMPLE STIMULI FOR EXPERIMENT 2 AND MEAN FREQUENCIES FOR BIGRAMS SURROUNDING CRITICAL LETTJZR(S) ACCORDING TO MAYZNER AND TRESSELT (1965) (MT) AND SOLSO AND JUEL (1981) (SJ) Bigram frequencies Spelling Stress c type 1 1 L melon 1 1 N lemon 1 1 0 seven 1 2 L relate 1 2 N canoe 1 2 0 guitar 2 1 L borrow 2 1 N bonnet 2 1 0 bobbin 2 2 L balloon 2 2 N 2 2 0 Prior bigram Example buffet MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: MT: SJ: 561 1461 335 928 543 823 541 1374 679 1712 573 1513 169* 721 2% 895 271 742 290* 2422 329 2961 235 1032 VC cc cv Subsequent bigram 431 909 %3 797 172* 379* 656 756% 445 645 222* 556 473 944 706 700 135 228* 613 886 854 828 260 260* - 396 364* 153* 206* 214 640 495* 988 319* 362* 238 162* 442 784 292 535 119 577 520 746 358 403 231 547 299* 831 389 507 570 1130 517 843 733 %5 639 1176 369 651* 531 1638 642 1076 459 572* 670 660 407 837 451 1014 49* 342* 50* 347 308 663 44* 183* 96* 3% Note. Critical letter(s) is italicized and lowest bigram frequency is marked with *. L, liquid; N, nasal; 0, obstruent. ject who originally participated in the oral cantly outnumbered 1 responses for only task was replaced because of inability to two types of stimuli, both of which had second syllable stress and single letter perform the task. spellings. The proportion of 1-2 responses varied Results and Discussion as a function of spelling (F(1,19) = 114.46 The results of the oral task are shown on in the analysis by subjects; F(1,108) = the left side of Table 4. Friedman analyses 655.17 in the analysis by items; p -C .OOlfor of variance by ranks were carried out sepa- both). As in Experiment 1, when subjects rately for each stimulus type; in all cases spelled an intervocalic consonant with two the difference among the four types of re- letters they were more likely to place that sponses was statistically significant (p < consonant in both syllables of their re.OOl). As in Experiment 1, multiple com- sponse. Stress also affected l-2 responses parisons based on rank sums showed that (F(1,19) = 17.19 by subjects; F(l,lOS) = 1-2 responses significantly outnumbered 71.52 by items; p < .OOlfor both). Subjects Other responses when the critical conso- more often placed a consonant in both sylnant was spelled with two letters. In this lables when the first syllable was stressed experiment, however, 2 responses signifi- than when the second syllable was SYLLABIFICATION RESULTS OF OF INTERVOCALIC 95 CONSONANTS TABLE 4 ORAL AND WRITTEN TASKS OF EXPERIMENT 2 Written task Oral task Multiple comparisons based on rank sums Mean proportion responses Spelling Stress C type 1 2 1 1 1 1 1 1 2 2 2 2 2 2 1 1 1 2 2 2 1 1 1 2 2 2 L N 0 L N 0 L N 0 L N 0 .41 .50 .35 .12 .08 .06 .13 .14 .14 .06 .Ol .06 .30 .32 .49 .x3 .84 .93 .08 .06 .I1 .35 .29 .41 l-2 .23 .17 .16 .lO .08 .Ol .78 .80 35 .59 .64 .48 Other .OO .Ol .Ol .Ol .oo .oo .OO 40 .oo .oo .OO .OO 1 vs 2 1-2 vs Other ns. n.s. n.s. 2>1 2>1 n.s. n.s. ns. n.s. n.s. n.s. n.s. n.s. n.s. n.s. ll.S. l-2 l-2 1-2 l-2 l-2 l-2 n.s. n.s. > Other > Other > Other > Other > Other > Other Prop. 1 responses .62 .70 .52 .28 .15 .13 .62 .61 .54 .31 .45 .32 Diff. from .50 by two-tailed t tests n.s. >.so n.s. <so <.so <so n.s. n.s. n.6. n.s. n.s. c.50 --- Note. L, liquid; N, nasal; 0, obstruent. stressed. The effect of stress replicates that found in Experiment 1. Finally, there was a main effect of consonant type (F(2,38) = 7.16 by subjects; F(2,108) = 5.07 by items; p < .Ol for both). Post hoc tests showed that subjects were more likely to place an intervocalic liquid or nasal in both syllables than an intervocalic obstruent. An analysis of the proportion of 1 responses relative to the number of 1 responses plus 2 responses yielded a main effect of stress (F(1,19) = 58.49 in the analysis by subjects; F(l,lOS) = 143.54 in the andysis by items; p < .OOlfor both). As in Experiment 1, subjects were more likely to choose 1 responses when the first syllable was stressed than when the second syllable was stressed. (The main effect of consonant type was significant in the analysis by subjects, F(2,38) = 7.97, p < .OOl, but did not reach the .05 level in the analysis by items, F(2,108) = 2.53, p < .lO). The results of the written task appear on the right side of Table 4. Analyses of variance found a main effect of spelling (F(1,23) = 6.69, p < .05 in the analysis by subjects; F(1,108) = 15.38, p < .OOl in the analysis by items). Subjects made more first syllable responses when the critical consonant was spelled with two letters than when it was spelled with one letter. There was also a main effect of stress (F(1,23) = 39.14 by subjects; F(1,108) = 188.86 by items; p < .OOlfor both). First syllable responses were more common when the first syllable was stressed than when the second syllable was stressed. The interaction between spelling and stress was also significant (F(1,23) = 22.03 by subjects; F(1,108) = 25.02 by items; p < .OOl for both), showing that the effect of stress was greater when the critical consonant was spelled with one letter than when it was spelled with two letters. All these effects replicate those previously found in Experiment 1. In addition, the main effect of consonant type was significant (F(2,46) = 16.98 by subjects, p < .OOl; F(2,108) = 6.84 by items, p < .005). Subjects made more first syllable responses for liquids and nasals than for obstruents, the same trend observed in the oral task. The results of the written task do not show a close correspondence to the bigram frequencies. For example, the words with single letter spellings and second syllable TREIMANANDDANIS 96 stress showed a predominance of second syllable responses, but did not in general have a bigram trough at the VC boundary. In summary, Experiment 2 found the same effects of spelling and stress previously observed in Experiment 1. In addition, subjects’ responses in the oral and written tasks were affected by the type of intervocalic consonant. In the oral task, subjects were more likely to place a liquid or nasal in both syllables of their response than to place an obstruent in both syllables. Also, first syllable responses tended to be more common for liquids and nasals than for obstruents, although the trend did not reach significance in the oral task. The interpretation of these results will be taken up under General Discussion. EXPERIMENT 3 One view of syllabification that has yet to be examined is the notion that syllabification differs after “long” and “short” vowels (Hansen & Rodgers, 1968; Pulgram, 1970; Webster’s, 1984). Experiment 3, therefore, examined V’CV sequences that differed in their first vowel. The type of consonant was also manipulated, as done in Experiment 2. In Experiment 3, all the intervocalic consonants were standardly spelled with a single letter. If, in the previous experiments, 1-2 responses for words with single letter spellings only arose because of the inclusion of many words with double letter spellings, few such responses would be observed in Experiment 3. Method Procedure The procedures for the oral and written tasks were identical to those of Experiment 2. Stimuli All the stimuli were two syllable words with first syllable stress. The medial consonant was standardly spelled with a single letter. The first vowel was either /I/, /E/,I=/, Id, /al, Ii/, let, /ail, lo/, or lu/. The first five vowels are called “short”or “lax” vowels in that they cannot end a word in English.5 The second five vowels are considered “long” or “tense.” They can occur in word-final position. For each vowel, there were four words with a medial liquid, four with a medial nasal, and four with a medial obstruent. An exception was the vowel /A/ with a following liquid; there was only one stimulus in this category. The total number of stimuli was 117. Table 5 gives the frequencies for the bigrams surrounding the critical letter(s). Subjects Twenty Wayne State University students participated in the oral task and 24 participated in the written task. Results and Discussion Table 6 shows the proportion of responses of each type in the oral task. Friedman analyses of variance by ranks were carried out separately for each stimulus type; in all cases the difference among the four types of responses was statistically significant (p < .OOl). Multiple comparisons showed that 2 responses significantly outnumbered 1 responses when the vowel was long and the consonant was an obstruent. Also, l-2 responses significantly outnumbered Other responses when the vowel was short and the consonant was a liquid or nasal. J While /a/ does occur word-finally in a small number of words (e.g., spa), it is generally considered “short” when it occurs in words like olive. Subjects in both the oral and written tasks did treat /al like lri, /c/, /a~/, and IA/. SYLLABIFICATION OF INTERVOCALIC TABLE 97 CONSONANTS 5 SAMPLE STIMULI FOR EXPERIMENT 3 AND MEAN FREQUENCIES FOR BIGRAMS SURROUNDING CRITICAL LETTER ACCORDING TO MAYZNER AND TRESSELT (1965) (MT) AND SOLSO AND JUEL (1981) (SJ) Bigram frequencies V type C type Example Short L relish Short N senate Short 0 credit Long L tulip Prior bigram MT: 657 SJ: 1743 MT: 387 SJ: 1008 Long N tunic Long 0 crisis MT: SJ: MT: SJ: MT: SJ: MT: SJ: 488 1178 231* 894 355 1017 138* 830 VC cv 375% 907 595 920 236% 647* 376 417* 270* 435* 257 673 371 342% 223* 144* 237 568 615 999 548 667 219 442* Subsequent bigram 426 792 485 602 640 2391 456 899 574 1047 729 2323 Note. Critical letter(s) is italicized and lowest bigram frequency is marked with *. L, liquid: N, nasal; 0, obstruent. Analyses of variance showed that the proportion of I-2 responses varied as a function of vowel type (F(1,19) = 12.31 in the analysis by subjects; F(1 ,111) = 49.44 in the analysis by items; p < .OOl for both). There were more 1-2 responses for short vowels than for long vowefs. In addition, there was a main effect of consonant type (F(2,38) = 18.92 by subjects; F(2,lll) = 11.83 by items; p < .OOl for both). As in Experiment 2, subjects made fewer l-2 re- TABLE sponses when the medial vowel was an obstruent than when it was a liquid or a nasal. Liquids and nasals did not differ signiticantly. The proportion of 1 responses relative to the total of 1 responses and 2 responses was greater for short vowels than for long vowels (F( 1,19) = 45.55 in the analysis by subjects; F(l,lll) = 66.12 in the analysis by items; p < .OOl for both). In addition, there was a main effect of consonant type 6 RESULTS OF ORAL AND WRII-~EN TASKS OF EXPERIMENT 3 Written task Oral task Mean proportion responses Multiple comparisons based on rank sums V type C type 1 2 l-2 Other 1 vs 2 l-2 vs Other Short Short Short Long Long Long L N .40 .46 .27 .26 .26 .17 .33 .29 .58 .58 .68 .I9 .26 .24 .14 .15 .06 .03 .Ol .02 .Ol .02 as. n.s. ns. n.s. n.s. 1-2 > Other l-2 > Other n.s. n.s. ns. ns. 0 L N 0 Note. L, liquid; N, nasal; 0, obstruent. .Ol .02 2>1 Prop. 1 responses Diff. from SO by two-tailed t tests I .s5 .I54 z>.50 n.s. .44 .37 n.s. <so .14 .13 c.50 c.50 98 TREIMANANDDANIS (F(2,38) = 12.33 by subjects; F(2,lll) = 21.31 by items; p < .OOl for both) and an interaction between vowel type and consonant type (F(2,38) = 4.51, p < .02 by subjects; F(2,lll) = 5.78, p < .005 by items). On the whole, the proportion of 1 responses relative to 1 plus 2 responses was greater for stimuli with medial liquids and nasals than for stimuli with medial obstruents. However, nasals showed the highest proportion in the short vowel case and liquids showed the highest proportion in the long vowel case. In the written task, the proportion of 1 responses was higher for short vowels than for long vowels (F(1,23) = 44.79 by subjects; F(1 ,111) = 119.76 by items; p < .OOl for both). Also, there was a main effect of consonant type (F(2,46) = 22.16 by subjects; F(2,lll) = 10.79 by items; p < .OOl for both) and a vowel type by consonant type interaction (F(2,46) = 24.68 by subjects; F(2,lll) = 8.88 by items; p < ,001 for both). In the short vowel case, 1 responses were most numerous for stimuli with medial liquids while in the long vowel case they were most numerous for stimuli with medial nasals. Subjects in the written task did not generally syllabify the words at the location of the bigram trough. For example, the consonant was usually placed in the second syllable after a long vowel but the bigram trough was not between the vowel and the consonant for most of these stimuli. To summarize, subjects’ responses in the oral and written tasks were affected by the phonological category of the medial consonant, as previously found in Experiment 2. In the oral task, l-2 responses were more frequent for words with medial liquids and nasals than for words with medial obstruents. Also, given that subjects placed the consonant in one syllable or the other, they chose the first syllable more often for liquids and nasals. The written task showed parallel effects of consonant type. Subjects’ responses were also affected by the type of vowel in the first syllable of the V’CV. When the vowel was “short”, l-2 responses were fairly common. Also, the proportion of first syllable responses was higher for “short” vowels than for “long” vowels. These results will be discussedfurther below.6 GENERAL DISCUSSION Several variables affected subjects’ performances in the syllabification tasks used here. We will review the influence of each variable and then discuss the implications of the results for various theories of syllabification. One factor that affected subjects’ performances was the spelling of the stimuli. In the oral syllabification task, people were more likely to place an intervocalic consonant in both syllables if they thought that the consonant was spelled with a double letter than if they thought that it was spelled with a single letter. When forced in a written syllabification task to assign a double consonant to one syllable or the other, subjects performed near chance levels. This probably occurred because the one rule of syllabification that most subjects in all experiments remembered, according to the results of the interview given after the written syllabification test, is to divide double consonants. Subjects had some difficulty when they were prevented from applying this rule. The effects of spelling on oral syllabification concur with previous findings (Ehri & Wilce, 1986; Jakimik et al., 1985; Seidenberg & Tanenhaus, 1979) that spelling may affect people’s performance in tasks that do not specifically require its use. These results pose problems for the study by Fallows (1981), which did not consider the effects of spelling on children’s oral syllabifications. Double letter spellings, however, were 6 For some of the stimuli in Experiment 3 (those marked with * in the Appendix), the classification of the vowel preceding Irl is in doubt. The data were therefore analyzed without these stimuli. The results were similar to those reported. SYLLABIFICATION OF INTERVOCALIC not the only factor that caused subjects in the oral task to place a consonant in both syllables. People made some such responses -up to 26%- when they knew that the consonant was spelled with a single letter. These responses continued to occur in Experiment 3, which did not include any words standardly spelled with double consonants. Other factors that promoted l-2 responses are discussed below. In addition to spelling, subjects’ responses were also affected by phonological factors. One such factor was the stress pattern of the stimuli. In the oral and written tasks of Experiments 1 and 2, second syllable responses predominated for words like device and canoe which have stress on the second syllable. Such responses occurred between 72 and 93% of the time for stimuli with second syllable stress provided that the critical consonant was spelled with a single letter. Another phonological factor to affect the results was the type of medial consonant. Experiments 2 and 3 found l-2 responses to be more frequent for words with medial Iiquids and nasaIs than for words with medial obstruents. First syllable responses showed the same trend. Liquids and nasals seem to have a greater degree of cohesion with the preceding vowel and are more likely to be placed in that vowel’s syllable. Finally, Experiment 3 found that subjects’ responses to V’CVs were affected by whether the first vowel was “short” or “long.” First syllable responses were more common in the short vowel case than in the long vowel case, as were l-2 responses. How do the results fit with the linguistic theories of syllabification that were discussed in the introductory remarks? Consider, first, the effects of stress. The predominance of second syllable syllabifications in the VCV’ case is consistent with many linguistic theories. Linguists such as Bailey (1978), Hoard (1971), Hooper (1972), Kahn (1976), Pulgram (1970), and Selkirk (1982) agree that VCV’ sequences are syllabified as V-CV’ provided that the C is legal in word-initial position, as was CONSONANTS 99 the case for the stimuli used here. That subjects usually placed the consonant of a VCV’ in the second syllable speaks against the theory of Anderson and Jones (1974), which states that the intervocalic consonant is ambisyllabic in such cases. Linguistic theories show much less agreement for V’CV sequences than for VCV’ sequences, and subjects’ responses to V’CV sequences showed much less agreement as well. In both Experiments 1 and 2, first syllable responses were more common for V’CVs than for VCV’s; there were also more l-2 responses for V’CVs. That second syllable responses did not predominate for all V’CV sequences shows that the results do not reflect slow speech syllabification as described by Kahn (1976) or the level of basic syllable composition as described by Selkirk (1982), where VCVs are syllabified as V-CV regardless of stress. The results are also inconsistent with Hoard’s (1971) theory, in which all V’CV sequences are syllabified as V’C-V, and with Selkirk’s surface level, where the same syllabification occurs. Current linguistic theories of syllabification do not predict the effects of consonant type observed in these experiments. Most linguistic theories do not differentiate among intervocalic consonants as a function of their phonetic category.’ However, there is some precedent for the present results in previous findings. Fallows (1981) reported that ambisyllabic responses were more frequent with liquids and nasals than with other consonants. MacKay (1978), Shattuck-Hufnagel (1986), and Stemberger (1983b) using speech error evidence, Treiman (1984) using word games, and Treiman and Danis (in press) using memory errors found that liquids have a closer bond to the preceding vowel than do obstruents. ’ Bailey (1978) does treat liquids differently than nasals and obstruents. However, he states that both nasals and obstruents in V’CV sequences are syllabified with the preceding vowel, which is not consistent with our results. 100 TREIMAN Treiman (1984) and Treiman and Danis (in press) found differences between liquids and nasals; the speech error studies did not discuss nasals. Thus, while there is some question about the status of nasals relative to liquids and obstruents, the present results agree with previous findings in suggesting that sonorants are more closely linked to a preceding vowel than are obstruents. Sonority differences may need to be built into theories of syllabification as well as theories of the structure of individual syllables. Finally, we must consider the obtained differences between “short” versus “long” vowels in light of linguistic theories of syllabification. Pulgram’s (1970) theory and the theory embodied in Webster’s (1984) dictionary predict a difference between short and long vowels; the other linguistic theories do not. According to the Pulgram and dictionary theories, first syllable responses should predominate whenever the first vowel of a V’CV is short; second syllable responses should predominate whenever the vowel is long. The results were not completely consistent with these predictions. Although second syllable responses tended to be more frequent than first syllable responses for long vowels, the results varied with the type of consonant. When the first vowel was short, the results again differed by consonant type. Indeed, with short vowels and medial obstruents there was a nonsignificant trend toward second syllable responses- the reverse of that predicted by the theories. It is possible that the observed differences between short and long vowels are an artifact of the present tasks. In the oral task, subjects indicated their syllabifications by producing the first syllable of the stimulus in the final position of their response. When the first syllable contained a short vowel, subjects may have been more likely to close that syllable with a consonant so as to produce a phonologically legal response. When the first syllable contained a long vowel, a response without a final AND DANIS consonant would be legal. If subjects pronounced the stimuli in the written task (as most said they did), similar effects would occur there. Phonological legality can explain why subjects produced more first syllable responses for words with short vowels in the first syllable and also for why they produced more l-2 responses in this case. To determine whether syllabification truly differs as a function of vowel type, we must develop a task in which subjects do not need to pronounce syllables in isolation. One possibility, following Treiman and Danis (in press), is to study memory errors for bisyllabic stimuli. For example, when trying to remember /gae’kat/ and /ma ‘ball, are lgae’kall and /me ‘bat/ more common errors than /gae’bal/ and /m e ‘kat/? Such a result might suggest that the stimuli are syllabified as /gz’k-al/ and /me ‘b-at/ and that errors may combine the first syllable of one stimulus with the second syllable of another. Until such an experiment is done, no firm conclusions can be drawn about the effect of vowel type on sylhbification. Of the linguistic theories that were reviewed, two seem to have the best potential for accounting for the data. These are the theories of Kahn (1976) and Selkirk (1982). In Kahn’s theory, an intervocalic consonant is syllable-initial if the following vowel is stressed. If the preceding vowel is stressed, the consonant is ambisyllabic in normal and fast speech. The notion of ambisyllabicity provides a possible explanation for the l-2 responses for V’CVs in the oral task. According to Kahn, ambisyllabicity does not vary with the type of consonant. However, the results of Experiments 2 and 3 suggest that liquids and nasals are more likely to be ambisyijabic t,han obstruents. Kahn’s theory may need to be modified by providing for greater ambisyllabicity for sonorants than for obstruents. In addition, Experiment 3 suggests that consonants after short vowels are more likely to be ambisyllabic than consonants SYLLABIFICATION OF INTERVOCALIC after long vowels, aithough as discussed above this result has an alternative explanation. That l-2 responses never occurred on more than about a quarter of the trials could be seen as a further problem for Kahn’s theory. If the consonants were truly ambisy+labic, one might have expected many such responses. On the other hand, subjects may have produced relatively few l-2 responses because no such responses occurred during the practice phase. Selkirk’s (1982) theory provides an alternative way of explaining the resuits. According to this theory, intervocalic consonants are syllable-initial at the level of basic syllable composition, provided that they are phonologically legal in this position and that there are no conflicting morphological boundaries. Resyllabification rules then move tile consonants in V’CVs to syllablefind position. At first glance, the present results seem incompatible with this theory. Subjects did not consistently place the medial consonant of words like proper in the second syllable, as Selkirk’s ievel of basic sytlabfe composition maintains, nor did they consistently place the medial consonant in the first syllable, the syllabification that results from the resyllabification rules. However, the resuits may be consistent with S&kirk’s theory if we hypothesize (as S&irk does, p. 379) that people are sensitive to the discrepancy between underlying and surface syhabifications in the case of words Iike proper. Because the medial Ipl is [email protected] at one level and syllablefinal at an&&r, people are unsure how to treat it. Consequently, they sometimes place it in the first syllable, sometimes in the second syllable, and sometimes in both. This view can explain why l-2 responses did not predominate in the V’CV case. However, Selkirk’s theory, like Kahn’s, cannot explain why subjects’ choices are affected by other factors such as the type of consonant. As discussed in the introductory remarks, a previous behavioral study (Stem- CONSONANTS 101 berger, 1983a) argued for ambisyllabicity. The evidence was that consonants like the /t/ of water interact both with syllable-final consonants and syllable-initial consonants in speech errors. In contrast, consonants whose syllable position is clear, like the /t/ of protect, almost always interact with consonants in the same syllabic position. While this evidence is consistent with Kahn’s (1976) theory, it does not conclusively show that the It/ of water is simultaneously represented as both syllable-final and syllable-initial, as Kahn claims. It is also possible, following Selkirk (1982), that the consonant is syllable-initial at a deep level and syllable-final at a surface level and that some speech errors reflect one level and some the other. Syllable units play a role not only in linguistic theories but also in theories of word recognition and reading. How do the units found in the written task of the present study correspond with the units that have been hypothesized for reading? According to Hansen and Rodgers (1968) and Spoehr and Smith (1973), readers use syllable-like units called VCGs. The VCGs of demon are de and mon; the VCGs of lemon are lem and on. As these examples show, the VCG theory predicts a difference between long and short vowels. Such a difference did appear in the present results, although as discussed above it could be an artifact. The VCG theory may be able to account for the effects of stress. The groupings predicted by this theory depend on the particular spelling-sound rules that are assumed and on the nature of people’s phonological representations, topics upon which Hansen and Rodgers (1968) do not elaborate. However, if the de of a word like demote is translated into /da/, or if it is translated into /di/ and this is later reduced to /da/, the word would have the VCGs de and mote, consistent with the present results. The VCG theory does not, however, account for the effects of consonant type that were observed in these experiments. According to Taft (1979), printed words 102 TREIMAN AND DANIS APPENDIX are accessed in terms of BOSS units. The units observed in the written task of the Stimuli for Experiment I present study do not correspond to the Single C spelling, first syllable stress: demon, revBOSS units proposed by Taft. For ex- erent, cover, vanity, relative, proper, elephant, opium, ample, since pairs like vanity and vanilla elegant, protestant, dilettante, delicatessen, tomaand comma and command are orthograph- hawk, novelist, retina, devil, evil, delicate, remedy, ically similar (and do not have a morpho- recipe, opening Single C spelling, second syllable stress: demote, logical boundary before the critical conso- reversal, covert, vanilla, relaxing, propel, election, nant or consonants), they have the same opinion, electric, protection, dilemma, deliberating, BOSSes. Subjects in the present study, tomato, November, retirement, device, evict, delihowever, treated such pairs differently. In cious, remember, recital, opossum Double C spelling, first syllable stress: comma, particular, subjects preferred to divide a common, ballot, attitude, buffer, correlate, written word like vanilla into vu and nillu corrugate, comment, apparatus, appetite, attic, opportune, masrather than into the BOSS unit van plus sacre, occult, illustrate, effigy, commodore, accurate, illu, However, the units involved in lexical appetizer, asset, posse Double C spelling, second syllable stress: comaccess and the units involved in decisions about syllabification could well be dif- mand, commode, commend, balloon, attorney, buffet, correction, corruption, apparently, appearance, attire, ferent . opponent, massages, occur, illusion, efficient, comA similar distinction must be made with motion, accustom, appendages, assess, possess respect to the theories of Adams (1981) and Seidenberg (1987). According to Seiden- Stimuli for Experiment 2 berg, the units that are involved in the iniSingle C spelling, first syllable stress, liquid: melon, tial processing of written words can be de- relic, malice, relish, aloe, solid, solace, Alex, Alice, fined on an orthographic basis. Breaks be- colic Single C spelling, first syllable stress, nasal: lemon, tween units are located where the panic, manage, vanish, manic, Amish, tonic, blemish, associations between adjacent letters are anise, sonic weakest. Such units did not appear to play Single C spelling, first syllable stress, obstruent: a role in the present task. Our subjects syl- seven, epic, avid, radish, acid, topic, logic, ever, labified words like propel differently than habit, robin Single C spelling, second syllable stress, liquid: rewords like proper even though the two select, delay, carouse, sarong, erase, elect, types of words have similar patterns of bi- late, alarm, Elaine, erect gram frequencies. It appears that the sylSingle C spelling, second syllable stress, nasal: lable-like units that emerge in the early pro- canoe, demand, deny, remorse, amuse, amaze, cessing of written words are not identical amount, enough, remain, emerge Single C spelling, second syllable stress, obstruent: to phonological syllables, a point also made depend, reveal, guitar, report, evade, adorn, obey, by Prinzmetal et al. (1986). adore, abuse, divorce To interpret the role that syllable-like Double C spelling, first syllable stress, liquid: units may play in various reading tasks, it borrow, sorrow, trolley, ballot, fellow, valley, pillow, is necessary to have a better understanding gully, Ellen, pillage Double C spelling, first syllable stress, nasal: of phonological syllables. The research bonnet, sonnet, comma, Dennis, tennis, annals, inner, presented here is a first step in that direc- bunny, penny, rummage tion. A complete picture of phonological Double C spelling, first syllable stress, obstruent: syllabification will require the use of other bobbin, lobby, hobby, rabbit, fetter, baffle, cribbage, tasks and the examination of other cases, buggy, Betty, luggage Double C spelling, second syllable stress, liquid: such as intervocalic clusters. At this point, collide, corral, balloon, collect, arrange, array, corhowever, it is clear that close attention rect, allow, ellipse, arrest must be paid to linguistic factors when Double C spelling, second syllable stress, nasal: studying the role of syllables in the pro- connect, command, connive, commit, immense, innate, announce, commend, anneal, annoy cessing of spoken and written language. SYLLABIFICATION OF INTERVOCALIC Double C spelling, second syllable stress, obstruent: support, suppose, suffice, buffet, massage, attire, occur, attend, effect, afford Stimuli for Experiment 3 /I/, liquid: lily, Lihth, chili, Philip /I/, nasal: limit, clinic, image, linen /I/, obstruent: critic, liver, livid, vigor /F/, liquid: relic, relish, melon, Helen W, nasal: lemon, venom, senate, tenant /d, obstruent: credit, second, devil, medal /a~/, liquid: Alice, balance, valid, salad /a~/, nasal: vanish, banish, camel, manage ire/, obstruent: acid, avid, hazard, placid IA/, liquid: color IA/, nasal: punish, money, honey, pumice /A/, obstruent: shovel, oven, study, cover /a/, liquid: column, solid, colic, olive /a/, nasal: onyx, honest, honor, promise /a/, obstruent: Robert, profit, proper, product ii/, liquid: era*, Eli, series*, serum* /i/, nasal: lemur, Venus, demon, senile /i/, obstruent: even, evil, liter, ego lel, liquid: Paris*, parent*, Eric*, valence iei, nasal: shaman, banal, Damon, Amos /e/, obstruent: bacon, baby, basis, raven /ai/, liquid: iris, lilac, pilot, silo /ail, nasal: final, lima, Simon, sinus /ai/, obstruent: Ivan, vital, libel, crisis /o/, liquid: molar, moral*, coral*, solar lo/, nasal: omen, moment, onus, bonus /o/, obstruent: motor, oval, open, yoga lul, liquid: hula, Julie, lurid*, tulip /ii/, nasal: tumor, rumor, tunic, unit /u/, obstruent: Judas, judo, Yukon, scuba tional zation zarion. concept research in experi- Hillsdale, NJ: Erlbaum. ANDERSON, J., &JONES, C. (1974). Three theses concerning phonological representations. Journal of Linguistics, 10, l-26. BAILEY, C.-J. N. (1978). Gradience in English syllabiand a revised Bloomington: of unmarked syllabi- Indiana Univ. Linguistics Club. BUCKINGHAM, H. W. (1980). On correlating aphasic errors with slips-of-the-tongue. Applied Psycholinguistics, 1, 199-220. EHRI, L. C., 8~ WILCE, L. S. (1986). The influence of spellings on speech: Are alveolar flaps Id/ or it/? In D. B. Yaden & S. Templeton (Eds.), Metalin- * See footnote 6. analysis of present-day Journal of Verbal 22, 310-332. ADAMS, M. J. (1981). What good is orthographic redundancy? In 0. J. L. Tzeng & H. Singer (Eds.), Reading guistic awareness and beginning literacy. Portsmouth, NH: Heinemann. FALLOWS, D. (1981). Experimental evidence for English syllabification and syllable structure. Journal of Linguistics, 17, 309-317. HANSEN, D., & ROWERS, T. S. (1968). An exploration of psycholinguistic units in initial reading. In K. S. Goodman (Ed,), The psycholinguistic nature of the reading process. Detroit: Wayne State Univ. Press. HOARD, J. W. (1971). Aspiration, tenseness, and syllabification in English. Language, 47, 133- 140. HOLLANDER, M., & WOLFE, D. A. (1973). Nonparametric statistical methods. New York: Wiley. HOOPER, J. B. (1972). The syllable in phonological theory. Language, 48, 525-540. JAKIMIK, J., COLE, R. A., & RUDNICKY, A. 1. (1985). Sound and spelling in spoken word recognition. Journal of Memory and Language, 24, 165-178. KAHN, D. (1976). Syllable-based generalizations in English phonology. BIoomington: Indiana Univ. Linguistics Club. K&ERA, H., & FRANCIS, W. N. (1967). ComputaAmerican English. Providence: Brown Univ. Press. LIMA, S. D., & POLLATSEK, A. (1983). Lexical access via an orthographic code? The Basic Orthographic Syllabic Structure (BOSS) reconsidered. REFERENCES Perception of print: mental psychology. 103 CONSONANTS Learning and Verbal Behavior, MACKAY, D. G. (1978). Speech errors inside the syllable. In A. Bell & J. B. Hooper (Eds.), Syllables and segments. Amsterdam: North-Holland. MAYZNER, M. S., & TRESSELT, M. E. (1%5). Iambles of single-letter and digram frequency counts for various word-length and letter-position combinations. Psychonomic Monograph Supplements, 1 (Whole No. 2). PRINZMETAL, W., TREIMAN, R., & RHO, S. (1986). How to see a reading unit. Journal of Memory and Language. 25,461-475. PULGFlAM, E. (1970). Syllable, word, nexus, cursus. The Hague: Mouton. SEGUI, J. (1984). The syllable: A basic perceptual unit in speech processing? In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes. Hillsdale, NJ: Erlbaum. SEIDENBERG, M. S. (1987). Sublexical structures in visual word recognition: Access units or orthographic redundancy? In M. Coltheart (Ed.), Attention reading. and Journal Learning of performance XII: Psychology of London: Erlbaum. SEIDENBERG, M. S., & TANENHAUS, M. K. (1979). Orthographic effects on rhyme monitoring. Experimental and Memory, Psychology: 5, 546-554. Human SELKIRK, E. 0. (1982). The syllable. In H. Van der 104 TREIMAN Hulst & N. Smith (Eds.), The structure ofphonological representations (part 2). Dordrecht, The Netherlands: Foris. SHAX-~IJCK-HUFNAGEL, S. (1986). The representation of phonological information during speech production planning: Evidence from vowel errors in spontaneous speech. Phonology Yearbook, 3, 117-149. SOLSO, R. L., & JUEL, C. L. (1981). Position frequency and versatility of bigrams for two- through nine-letter English words. Behavior Research Methods and Instrumentation, 12, 297-343. SPOEHR, K. T. (1981). Word recognition in speech and reading: Toward a single theory of language processing. In P. D. Eimas & J. L. Miller (Eds.), Perspectives on the study of speech. Hillsdale, NJ: Erlbaum. SPOEHR, K. T., & SMITH, E. E. (1973). The role of syllables in perceptual processing. Cognitive Psychology, 5, 71-89. AND DANIS STEMBERGER, J. I? (1983a). Speech errors and theoretical phonology. Bloomington: Indiana Univ. Linguistics Club. STEMBERGER, J. P. (1983b). The nature of/r/ and ill in English: Evidence from speech errors. Journal of Phonetics, 11, 139-147. TA~v, M. (1979). Lexical access via an orthographic code: The Basic Orthographic Syllabic Structure (BOSS). Journal of Verbal Learning and Verbal Behavior, 18, 21-39. TREIMAN, R. (1984). On the status of Final consonant clusters in English syllables. Journal of Verbal Learning and Verbal Behavior, 23, 343-356. TREIMAN, R., & DANIS, C. (in press). Short-term memory errors for spoken syllables are affected by the linguistic structure of the syllables. Journal of Experimental Psychology: Learning, Memory, and Cognition. Webster’s New World Dictionary of the American Language (2nd ed.) (1984). New York: Simon and Schuster.