Structured Heterogeneity and Change in Laryngeal Phonetics

JEngL 33.4 Purnell et al.(December / Upper Midwestern 2005) Final Obstruents 10.1177/0075424205285637 Structured Heterogeneity and Change in Larynge...
Author: Evelyn Gordon
1 downloads 2 Views 548KB Size
JEngL 33.4 Purnell et al.(December / Upper Midwestern 2005) Final Obstruents

10.1177/0075424205285637

Structured Heterogeneity and Change in Laryngeal Phonetics Upper Midwestern Final Obstruents THOMAS PURNELL

University of Wisconsin–Madison JOSEPH SALMONS

University of Wisconsin–Madison DILARA TEPELI

University of Wisconsin–Madison JENNIFER MERCER

University of Wisconsin–Madison American sociolinguists have largely ignored obstruents as invariant, including how speakers distinguish /s, t/ from /z, d/. Upper Midwestern final obstruents provide clear evidence that the realization of such contrasts can and does vary. In a once German-speaking Wisconsin town, we have found that speakers systematically produce final laryngeal distinctions differently than reported for American English, with an apparent partial neutralization of the distinction. Here, we seek the historical antecedents of this pattern, comparing acoustic characteristics of recordings from speakers throughout the region born from 1866–1986. Analysis by date of birth shows distinct obstruent phonetics over this whole period, revealing striking changes in which acoustic cues have been exploited to maintain the distinction: The oldest speakers used primarily glottal pulsing, younger ones exhibit a “trading relation” between pulsing and preceding-vowel duration, and the youngest have reduced the acoustic cues of the distinction dramatically. Keywords:

obstruents; sociophonetics; variation and change; voicing; trading relations; cue exploitation

This article provides evidence that the phonetics of laryngeal or “voicing” distinctions in regional American English, observable in minimal pairs sing~zing, bussing~buzzing, hiss~his, vary and change in systematic and previously unappreciated ways. Examination of multiple acoustic characteristics for the phonological voicing distinction in syllable codas indicates that the cues are used in tandem, and Journal of English Linguistics, Vol. 33 / No. 4, December 2005 307-338 DOI: 10.1177/0075424205285637 © 2005 Sage Publications

307

308 JEngL 33.4 (December 2005)

evolve along not one isolated parameter, but along several parameters at once. In the coming pages, we first provide a description of temporal acoustic voicing cues of American English coda obstruents (duration of the obstruent, the preceding vowel, and pulsing during the obstruent), and place them in the context of trading relations. We then report results of an acoustic study, showing a progression of key acoustic characteristics in speakers born over 120 years in eastern Wisconsin English. The resulting picture, initially rather chaotic, becomes much clearer when we chart this history in terms of a trading relation between vowel duration and pulsing. As such, we contextualize these changes within the concept of the general concept of voicing running through each step in the speech chain.

The Phonetic Description of American English Coda Obstruents The production and perception of American English laryngeal distinctions are famously complex, with over a dozen different perceptual cues (PvC, for perceptual cues for VOICING, Figure 1) and acoustic characteristics (AvC, for articulatory and acoustic VOICING characteristics, Figure 1) thought to be involved in medial position (Lisker 1986; Bregman 1990; Pilus 2002; among others). Ladefoged and Maddieson’s (1996, 95–99) precise overview shows how difficult this terminological terrain is, and we agree with Catford (1977, cited in Ladefoged and Maddieson 1996) that no terms can be used without explicit clarification. Traditionally, pairs of sounds such as English /z/ versus /s/ have been referred to as “voiced” and “voiceless,” but we follow a growing consensus (Iverson and Salmons 1995, 2003)1 that in English the distinction is fundamentally one where the traditional voiceless sounds are phonologically specified by [spread glottis] or Glottal Width (in the analysis of Avery and Idsardi 2001 and Iverson and Salmons 2003) and traditional “voiced” sounds are phonologically unmarked or inert and unmarked for laryngeal features or Glottal Width. On this view, in English and other aspirating languages such as German, phonetic voicing per se is only a low-level phonetic enhancement in voicing-friendly environments, where phonetic voicing is phonologically “passive.” Here, we use “fortis” and “lenis” as a convention for labeling contrastive classes such as /s/ ≠ /z/, without endorsing proposals for these as phonetic features (see Kohler 1984). Figure 1 depicts our broader view of “voicing” through the sound related portion of the speech chain. When we use “voicing” and do not disambiguate as to whether it means phonological ([spread glottis]), articulatory (phonation, glottal pulsing, AUTHORS’ NOTE: Versions of this article benefited from presentations at New Ways of Analyzing Variation 33, October 2004, and the annual meeting of the American Dialect Society, January 2005. We specifically appreciate Stuart Davis, Steven Huxmann, Bill Idsardi, Greg Iverson, Mark Louden, Monica Macaulay, Bert Vaux, and Malcah Yaeger-Dror for providing us with useful comments on this article and discussions on this topic. All errors are our own.

Purnell et al. / Upper Midwestern Final Obstruents 309

Figure 1: A Conceptual Depiction of VOICING throughout the Speech Chain. NOTE: Note that the distinctive features in (a) represent two distinct traditions for describing the obstruent contrasts in (b) and the items in (a) are not aligned with those in (b) just as elements in PeV are not aligned with those in AvC. Moreover, FORTIS and LENIS are used as nothing more than convenient labels for the contrasts in (b). Also, note that PvC is assumed to be either identical to AvC or a subset of AvC, and there is no condition whereby statistically significant individual characteristics in AvC are necessarily significant in PvC.

aspiration), acoustic (F0), or perceptual (vowel duration, glottal pulsing as a percent of consonant duration) voicing, the term is a generic one characterizing any or all of the processes, and we use the cover term VOICING throughout.2 The two items under Phonemic Voicing (PeV) represent two phonological traditions of analysis. An auditory object—that is, the product of an auditory stimulus being interpreted mentally as a single entity—contains perceptual cues from the set of acoustic characteristics that listeners integrate in order to understand units of speech. These cues often overlap such that no one characteristic is solely responsible for conveying a perceptual precept, boundary, or exemplar from the speaker to the listener (Hodgson and Miller 1996; Lisker 1975; Miller 2001; Parker et al. 1986; Repp 1982, Repp and Liberman 1987; Repp et al. 1978; Summerfield and Haggard 1977). For example, in oral plosives such as /p b t d k g/, spectral and temporal cues such as the transitions into and out of the stop, the stop closure, the voice onset time, and the spectral shape of the burst all indicate to the listener something about the place of articulation of the stop (Port 1976). Repp observes that while cues for one characteristic may vary in strength cross-linguistically, they often vary in strength even within one instance of speech “so as to maintain the original phonetic percept” (1982, 87). Continuing the place of articulation example, in fast speech the temporal cues (stop gap and voice onset time durations) would be affected. In hypothetical terms, if speakers weakened all of the cues for a specific characteristic such as place of articulation to the point of neutralization (e.g., the complete German laryngeal neutralization in codas, cf. Bad “bath” and bat “offered, 1st/3rd SG.PRET.”), then we might expect listeners to misperceive the sound produced altogether (as is

310 JEngL 33.4 (December 2005)

Figure 2: Schema Depicting Two Arbitrary Phonetic Dimensions and Their Potential Relationships. NOTE: This depiction assumes that {a, b} is the set of means of two distinct types of sounds along dimension x whereas {A,B} is the set of means of the same group of distinct sounds along dimension y. Panel (a) depicts how dimension x has priority over dimension y, while panel (b) represents the opposite situation. While panel (c) represents how both dimensions might agree in tandem, panel (d) represents trading relations between dimensions x and y. In (d), the (b,B) value can fall anywhere to the right of the diagonal boundary and (a,A) can fall anywhere to the left of the boundary.

the case in German). As we know from contrasts in English, however, cues often “trade off” with each other in terms of strength, preserving speech comprehension.3 In other words, as one cue is strengthened, other concomitant or complementary cues can be diminished.4 Such a relation is depicted in Figure 2. As becomes apparent from the vast literature on the topic (Gimson/Cruttenden 2001; Jessen 2001; Pilus 2002; Nittrouer 2004),5 no single item from the AvC or PvC clusters comfortably captures the phonological distinction. If we assume that a “trading relation” among cues or characteristics is within the range of our linguistic competence (such as shown in Figure 2d, relations among cues or characteristics in dialect-specific speech should be exploited or privileged along typical social dimensions (age, class, gender, occupation, race, etc). But just how AvC or PvC items co-vary either to reflect acoustically or perceptually significant VOICING within the context of dialect variation remains unclear. Thus, if glottal pulsing (in the fam-

Purnell et al. / Upper Midwestern Final Obstruents 311

ily of AvC characteristics) was plotted on the y axis and vowel duration on the x, panel (a) in Figure 2 would depict a distinction based on vowel duration, (b) would show one organized around pulsing. If both dimensions are active, then it is reasonable to assume that extreme values of both dimensions (the quadrants labeled (a,A) and (b,B) in Figure 2c) enhance each other and voiceless tokens will occur in the quadrant with lesser values on both dimensions while voiced tokens will occur in the diagonal quadrant. What characterization, however, is made of the remaining quadrants? As we show below, these quadrants are split because the aspect of dimension trading is active. Thus, the panel in Figure 2d represents a trading relation between vowel duration and glottal pulsing. In the hypothetical situation described by Figure 2d, voiceless tokens can display a great amount of glottal pulsing as long as they have very short vowels preceding the obstruent. With regard to the relation between AvC and PvC, some characteristics within the AvC family have been argued to transfer more robustly into the PvC family than others: glottal pulsing, vowel duration, and “low frequency” cues, in particular. Actual voicing—that is, glottal pulsing throughout the consonant itself—is often largely or even entirely absent in syllable codas (Gimson/Cruttenden 2001, 161–7). Still, even a reduced presence of pulsing in lenis or “voiced” obstruents may serve to support the contrast. A large body of work proposes that the most prominent cue to English final laryngeal distinctions is greatly increased vowel duration before lenis obstruents, typically being over one and a half times as long as before fortis or “voiceless” ones (Chen 1970; Crowther and Mann 1992; Denes 1955; Eliason 1942; Flege and Port 1981; Heffner 1937, 1942; House and Fairbanks 1953; Lehmann and Heffner 1943; Lisker 1957; Lloyd 1936; Raphael 1972; Rositzke 1943). Another prominent view is that a “low-frequency property” is associated with lenis obstruents, where F0 and F1 drop before final lenis obstruents to cue their presence (Castleman and Diehl 1996; Kingston and Diehl 1994, 1995; Stevens and Blumstein 1981). The body of evidence just discussed indicates that AvC and PvC manifestations of laryngeal distinctions are extremely complex. They proffer, in short, an excellent opportunity for investigating the very kinds of “structured heterogeneity” on which modern sociolinguistics has been built since Weinreich, Labov, and Herzog 1968. Yet, sociolinguists, dialectologists, and historians of American English have taken little notice of obstruents—with respect to the distribution of their moderately lowlevel sociophonetic characteristics—beyond occasional calls for more work in the area. Rositzke (1943, 41) concludes that his: . . . quite tentative remarks are based on the very slender evidence of the lexical speech-habits of five subjects and plainly do not represent any sort of statistically reliable description of their speech-habits or of [American] speech-

312 JEngL 33.4 (December 2005)

habits in general. They rather indicate the nature of the correlations that a more comprehensive study of these final consonants ought to investigate. . . . A third of a century later, Pederson (1977, 265) essentially repeats this call, writing that: Despite the acceleration of research in general phonetics, description of the consonants and their interrelationships remains unsatisfactory. Compared with vowels, consonants have been neglected by phonologists, phoneticians, and dialectologists in North America. Docherty and Foulkes (1999, 53–54) echo the point that the phonetics of consonants is understudied by observing how less than 10 percent of the voluminous Labov 1994 is devoted to discussing consonants, while there are no acoustic data presented on consonants. In fact, some work has been conducted on coda obstruent variation, mostly phonological and often limited to particular words, for example, voicing alternations such as in the medial fricative of greasy and other patterns observed by Kurath and McDavid (1961). More relevant is Allen’s brief observation that Upper Midwestern speakers of Scandinavian ancestry devoice final [z], for example hi[s] and her[s] (1973–79), unsurprising given the lack of [z] in Norwegian and Swedish. Knack (1991) shows that in Grand Rapids, Michigan, Jewish speakers devoice more than their non-Jewish counterparts. Davis and Anderson (2004) have begun to explore the distinct laryngeal phonology of so-called Dutchified English in southeastern Pennsylvania. More recent research in this area has been sociolinguistic, on “geek’ girls” stop release (Bucholtz 1996, cited in Podesva, Roberts, and Campbell-Kibler 2001), stop release in gay male speech (Podesva, Roberts, and Campbell-Kibler 2001), and of course variable deletion or glottalization of final coronal stops in African American English (e.g., Wolfram 1969; Kohl and Anderson 2000), with similar patterns in other varieties. Most important, in Docherty and Foulkes 1999 we find results parallel to the present article. Their sociophonetic analysis of Derby and Newcastle pre-pausal [t] shows that while the variants of [t] retain characteristic properties in the two locales, . . . subgroups within the accent community are realising these accent specific allophones in slightly different ways. What we are seeing is subgroups within a community adopting what we may term different phonetic realisation strategies (PRSs). The net outcome of the variation is simply one of widening the range of possible allophonic realisation. It is not transparent how

Purnell et al. / Upper Midwestern Final Obstruents 313

the linguistic system can be held responsible for inducing such variation. (1999, 69). The goal of the current study is to investigate three test hypotheses under the assumption that the presence of multiple acoustic characteristics for a single phonological distinction encourages speakers to exploit AvC for marking social group membership. The first hypothesis is that a distinctive pattern found in earlier work on speakers from Watertown, Wisconsin (Salmons et al. forthcoming; Tepeli et al. forthcoming; Purnell, Salmons, and Tepeli 2005; Purnell et al. 2004), is not only synchronically coherent, but can be traced over time. To test this, we analyze recordings made as far back as the early 1950s, yielding a 120-year span of birth dates. Which characteristics are used and to what extent they are used may prove susceptible to change over time. The distinction may be carried primarily by one characteristic or a subset from AvC at one point in time, but a very different way in subsequent generations. Our second test hypothesis is that the Watertown pattern extends across a broader regional and ethnic population. If various configurations of characteristics can signal a distinction, such acoustic changes need not, of course, alter the fundamental phonological relationships within a given dialect. To test this, we examine recordings of speakers from the points shown on the map in Figure 3. All of the subjects’ dialect locales lie within a region bounded by the urban centers of Milwaukee to the east, Green Bay to the north, and Madison to the west. European settlement of this area occurred approximately between 1830– 1850 and forms “a zone of German domination” in the words of Ostergren (1997). The history of this region, moreover, was further influenced by a significant influx of Polish, Dutch, Irish, and members of other linguistic and ethnic groups. We expect the pattern to be similar across ethnic and geographic groups, but different from those of a set of American English–speaking controls from outside Wisconsin. Third, based on the use of multiple acoustic characteristics to carry the phonological fortis~lenis distinction, we hypothesize that there will be a kind of cue exploitation here. To test this, as already noted, we focus on the interrelations and possible trading relations among what are reported to be the central acoustic characteristics in the extensive literature on English final laryngeal distinctions. Finding systematic variation in the relationships within the AvC family opens the possibility that speech communities can modify secondary cues, not to signal phoneme categorization but rather to signal social group membership. While perceptual studies will be required to confirm it, our expectation is that some of these acoustic characteristics marking final laryngeal distinctions may serve social functions in the Upper Midwest.

314 JEngL 33.4 (December 2005)

Figure 3: Geographic Distribution of Wisconsin Speakers in Present Study. NOTE: Some locations have more than one speaker. The gray area identifies the “Inland North,” where the Northern Cities Shift is found (Labov et al. 2005).

Method Subjects This article uses data from four test groups (often referred to throughout as groups 1 through 4; see Table 1) and two control groups. The first test group consists of five speakers born from 1866 to 1892, recorded as part of Frederic Cassidy’s Wisconsin English Language Survey (WELS, 1951–1953; see Cassidy 1948), and the 1968 Dictionary of American Regional English (DARE, Cassidy and Hall 1985). The second group, also WELS and DARE consultants, were born from 1899 to 1918. The third group contains mostly Watertown speakers from our initial study, with birth dates from the mid-1920s to the late 1930s. The final group includes younger subjects from the University of Wisconsin’s X-ray Microbeam database (Westbury 1994) and one subject recorded in the course of our earlier work

Purnell et al. / Upper Midwestern Final Obstruents 315 TABLE 1 Subject Distribution by Gender, Birth Date, and Test Group Number (Female/Male) Group 1: 1866–1892 WELS DARE Group 2: 1899–1918 WELS DARE Group 3: 1920–1939 DARE Watertown Group 4: 1966–1986 UW XRMB Watertown

Birth Dates

Recording Dates

2 (1/1) 3 (1/2)

1866 to 1872 1888 to 1892

1949 /1950s 1968

4 (3/1) 5 (3/2)

1899 to 1906 1898 to 1918

1949 /1950s 1968

1 (1/0) 8 (6/2)

1926 1920 to 1939

1968 2004

4 (3/1) 1 (1/0)

1966 to 1971 1986

1990 2004

a

a

a

a. Actual recording date uncertain.

(Purnell et al. 2004). This last subject was raised in Watertown, but had parents who were not from Watertown. Two control groups consist of speakers from outside Wisconsin on one hand and one native speaker of German on the other. Work reported in Salmons et al. (forthcoming) and Tepeli et al. (forthcoming) suggests that the lack of final laryngeal distinctions in German led to a distinct pattern in a previously German-speaking area in Wisconsin. We now look at whether similar patterns may be found across, and thus have generalized to, a broader population, most likely fed by other immigrant languages with final devoicing, such as Dutch and Polish.6 The Watertown evidence also suggests that the pattern would have permuted into something different from both English and those immigrant languages with final fortition or devoicing. Following this line of thought, several possible diachronic trajectories exist for eastern Wisconsin English speech patterns in relation to General American English (GAE) and languages such as German. First, all four test groups could show complete neutralization, unlike GAE but like those reported for the devoicing languages brought by German-, Dutch- and Polish-speaking immigrants. Second, all four groups could exhibit GAE patterns and show no resemblance to the patterns of devoicing languages. Third and fourth, we could find a shift from being more like German and other languages either between groups 2 and 3 or between groups 3 and 4. Fifth, we might see a shift at each time period from a point where the speakers may move toward the standard variety and then, in a sense, past it in younger generations. Finally, the earlier forms could display some degree of interlanguage (i.e., like neither GAE nor German).

316 JEngL 33.4 (December 2005) TABLE 2 Table Tokens by Recording Text 2004 Word list Arthur XRMB

bead, beat, bed, bet, bids, bits, cease, chop, dock, dog, edge, etch, job, sees, soup, tube always, at, back, blade, but, choice, dead, face, feet, foot, gave, half, have, his, noise, said, shook, stood, this, unsafe, yet ask, food, greasy, had, has, Hispanic, shoot, that, these

TABLE 3 Distribution of Tokens Across Recordings by Group Stops Fortis 1866–1892 1899–1918 1920–1939 1966–1986 Subtotal English L1 controls German L1 control Subtotal Total

Fricatives

Lenis

Fortis

Lenis

Total

3 13 53 7 76

24 30 54 13 121

18 28 12 13 71

7 14 11 19 51

52 85 130 52 319

33 7 40 116

38 6 44 165

6 1 7 78

6 1 7 58

83 15 98 417

Tasks Subjects in the WELS and DARE recordings were taped telling the Arthur story (Cassidy 1948; Cassidy and Hall 1985). Subjects in the Watertown recordings were recorded with word lists (embedded in a carrier sentence, see Purnell et al. 2005). Those subjects who are a part of the X-Ray Microbeam protocol were recorded with word-list, sentence, and short passage readings (Westbury 1994).

Tokens Because of the variable means of data collection, the words shown in Table 2 were selected for analysis. Table 3 presents the distribution of obstruents throughout our data by manner of articulation and speaker group.

Analytical Procedure Tokens were down sampled to 10 kHz. Using Praat, landmarks were identified (e.g., vowel onset, steady state, and offset; stop gap/fricative duration; pulsing duration). Landmark durations (e.g., pulsing), as well as the F0 and F1 at the end of the

Purnell et al. / Upper Midwestern Final Obstruents 317

steady state, were measured. Two ratios were computed: the ratio of pulsing in the consonant to the duration of the whole consonant, and the ratio of preceding vowel duration to consonant duration. These recordings posed a number of problems for acoustic analysis. The older recordings exhibited a good deal of ambient noise. To offset the interaction of noise during consonant production, the spectrogram and waveform were examined to help determine the landmarks. In particular, we first tried to locate a change in F1 of 50 Hz, and then looked for a sharp decay in energy at the end of the vowel, which often coincided with change in F1. Additionally, we encountered statistical noise. For instance, several lexical changes were made to the Arthur story over the years. Also, it is possible that some, though surely not all, of the variation across groups can be attributed to speakers’ attention in different styles of elicitation (reading words versus reading passage tasks).7 Moreover, while we did all we could to keep clear minimal pairs of a given place of articulation for stop or fricative after a given vowel (such as bead versus beat), we were forced to include a somewhat broader range of vowel and consonant combinations because of limitations in the available data.

Results We focus our attention here on results from four areas of change, all connected in some way to duration (i.e., temporal cues to VOICING): coda obstruent duration, vowel duration before coda obstruents, the ratio of vowel-to-consonant duration during the rhyme, and glottal pulsing during obstruents. T-tests and analyses of variance (ANOVA) were calculated on these measures. Where appropriate, the conservative Scheffé post-hoc test helped determine which means were significantly different. In order to contextualize the results, it is important to keep in mind the tendencies of these measures in American English. For the duration of the coda obstruents themselves, the fortis obstruents are generally longer than the lenis obstruents. Vowels before lenis obstruents are generally much longer than before fortis obstruents. Lenis obstruents also have a larger vowel-to-obstruent ratio, which suggests that as the vowel gets longer, the consonant shortens (and vice versa). Finally, for the proportion of glottal pulsing to consonant length, the vocal fold pulsing often covers more than half of the lenis obstruents, while the vocal fold pulsing for fortis consonants is approximately less than half that of the fortis obstruent.

Consonant Duration Overall, consonant durations support the generalization that fortis tokens are longer than lenis ones in coda position. ANOVA was computed for fortis conso-

318 JEngL 33.4 (December 2005) TABLE 4 Mean Consonant Duration by Group Fortis (msec) (5,6)

Lenis (msec)

Lenis - Fortis (msec)

1866–1892 1899–1918 1920–1939 1966–1986

172‡ 128 130 114

100 76 105 83

–72* –48* –25 –31*

English L1 Controls German L1 Control

100 76

102 77

2 1

*p < 0.05: significant difference between means for voicing found in t-test. NOTE: ‡(x,y) indicates significant differences found by an ANOVA, p < 0.05, and Scheffé post-hoc test with other group (5 = English controls; 6 = German control).

Figure 4: Coda Obstruent Duration for Four Groups of Wisconsin Speakers as Compared to NonWisconsin (AE) and German (G) Controls.

nants, showing significance across groups (F(5,188) = 4.91, p < 0.001). The means of the oldest group were found to be significantly different from both control groups (Table 4). The fortis bars in Figure 4 show what ANOVA confirms, namely that the means of the oldest group of speakers is significantly different from the American English control group, while the general pattern of Wisconsin English speakers is toward the GAE pattern. Review of the lenis bars in Figure 4 show that the distance between lenis and fortis is decreasing, although t-tests indicate that this difference

Purnell et al. / Upper Midwestern Final Obstruents 319 TABLE 5 Mean Vowel Duration by Group Fortis (msec) (4)

Lenis (msec) (3,5)

1866–1892 1899–1918 1920–1939 1966–1986

145ç 123 110 (4) 90‡

103‡ (3,5) 95‡ 156 125

English L1 Controls German L1 Control

101 89

157 115

Lenis – Fortis (msec) –42* –28* 46* 35* 56* 26*

*p < 0.05: significant difference between means for voicing found in t-test. NOTE: ‡(x,y) indicates significant differences found by an ANOVA, p < 0.05, and Scheffé post-hoc test with other group (3 = 1920–1939; 4 = 1966–1986; 5 = English controls).

is significant only for three of the four Wisconsin groups, but not the control groups. ANOVA for lenis tokens was also significant (F(5,217) = 2.70, p < 0.05), although the post-hoc test did not identify any significantly different group means.

Vowel Duration As noted above, previous research has shown a consistent pattern of vowel lengthening before lenis obstruents as opposed to fortis, the former being over one and a half times as long as the latter. T-tests show fortis values to be significantly different from lenis values for each group examined in the present article (Table 5). A downward trend in fortis values is apparent in Figure 5, and so an ANOVA was calculated for fortis obstruents (F(5,188)=3.71, p < 0.005). The means were found to be significantly different for the oldest and youngest Wisconsin groups. ANOVA was also calculated for lenis tokens (F(5,217)=11.74, p