Evidence from Rebétika and Folk Songs

In Arvaniti and Joseph (2000) we studied the variability in the pronunciation of the Greek phones spelled mp, nt, gg/gk, which in speech are said to consist of a nasal consonant, e.g., m, and a “voiced” stop consonant, e.g., b. Our data showed that the presence of the nasal depended largely on age, with younger speakers producing many more nasalless instances of these phones than older speakers. Here we examine the pronunciation of mp, nt, gg/gk in original recordings of early twentieth-century Greek rebétika and folk songs to see if these show similar variation, as linguistic theory would predict, or not (as traditional studies of Greek dialectology suggest). Our new data show variation in the pronunciation of these phones in a period for which no variation had been reported before. This early twentieth-century variation confirms our earlier conclusion that variation at the end of the twentieth century betokens a change to a new nasalless pronunciation, away for a previously stable variation pattern. This study reports on variability in the way in which speakers of Modern Greek pronounce the phones spelled mp, nt, gg/gk, which in speech are typically said to consist of a nasal consonant (m, n, or Ω1 ) and a “voiced”2 stop consonant (b, d, g). Our starting point is the situation found in the contemporary standard language, drawing on quantitative sociolinguistic data presented in Arvaniti and Joseph (2000). Here, however, we extend the empirical basis for understanding this situation by examining data from sound recordings of early twentieth-century Greek, in particular rebétika and folk songs from the period from 1910 to 1937, a corpus not previously exploited for the study of the historical development of Greek pronunciation. Before we proceed with our study, a brief presentation of the nature of variability in language is in order. 74 Amalia Arvaniti and Brian D. Joseph The Nature of Variability in Language One of the constants, so to speak, when it comes to language, is change. Even a glimpse at an earlier stage of a language, especially in comparison with a later stage, will bring to light numerous points of difference, the key to determining that a change has occurred. With such comparison, it is usual to talk about a change in “real time,” since different temporal “slices” of a language’s chronological development are involved. The differences between compared stages can be quite striking, especially when the temporal distance is great, as in (1), where both examples translate into English as “I don’t want to give”: (1) a. oÈk §y°lv dom°nai (adapted from Iliad 7.364) [eighth century B.C.] (literally, not want/1sg give/infin) b. den y°lv na d≈sv (contemporary Modern Greek) (literally, not want/1sg subjunc give/1sg) Moreover, virtually all aspects of a language can be subject to change. In (1) there is evidence for change involving: (i) vocabulary: e.g., oÈk vs. den for ‘not’; (ii) form: e.g., dom°nai (to give), an “infinitive,” is not part of the repertoire of verbal forms now found in Greek; (iii) syntax: e.g., the verb §y°lv (want) occurs with an infinitive as a subordinate form in Ancient Greek, and thus with no personal agreement endings on the subordinate verb, but its Modern Greek equivalent, y°lv, occurs with a subjunctive and thus with agreement; (iv) pronunciation: e.g., the verb want has three syllables in its Ancient Greek form but only two in Modern Greek.3 It should be clear that recognizing that a change has occurred can be easier with regard to some aspects of a language’s structure than with others. For instance, changes in the syntax of subordination with want are as obvious in (1) as they are striking; the availability of written records from earlier stages makes the comparison possible from which an observation of change can be made. However, if one is interested in matters of change in sounds (phonological or phonetic change), dealing with written records can present many problems, since writing systems often mask facts about details of pronunciation. Thus, as Labov (1994:11) puts it, the historical linguist often has to “mak[e] the best use of bad data.” While Janda and Joseph (2003:14) suggest “imperfect” as a better characterization of such data, in that the information available from earlier stages of a language “will of necessity be fragmentary or otherwise incomplete, possibly misleading,” Labov’s point is well-taken. It is thus incumbent on the linguist interested in studying a particular change in a given language to consider as much relevant data as possible, and thus when studying sound change to exploit sources of 75 Early Modern Greek /b d g/ information on actual pronunciation of sounds in earlier, otherwise not directly accessible, stages of a language. Because of such problems when working with historical data, many linguists have turned to the examination of ways in which the relevant data can be directly observed or inferred from a single stage, without comparisons across real time. In particular, they have focused attention on variation in usage found in the contemporary period, treating such variation as a window into change in language. Of special significance in this regard are the notions of “change in progress” and “change in apparent time.” These notions are rooted in the observation that variability is inherent in language. No two speakers, even of the same language, are exactly identical in all aspects of their language use, and differences among them may well depend not only on geographical (dialectal) variation, but also on social factors such as the class or gender of the speaker. Even one and the same speaker can show internal fluctuations in his/her own usage, depending on factors such as the style of speech and the social context in which the speech is uttered. This effectively means that a given linguistic feature—such as the pronunciation of a vowel, or the syntax of the English verb to need—can show variant forms; when these functionally equivalent “variants” acquire social significance for the speakers of a language, they form a “linguistic variable.” Linguists have reason to believe, moreover, that the linguistic variables can be evidence of a language change in progress, i.e., of a change that is becoming established in the speech community and is in the process of becoming a new usage norm in that community. This is especially so when there is “age-grading” evident in the distribution of variants across a speech community, i.e., when greater or lesser use of one variant as opposed to another correlates directly with age. For instance, younger speakers of American English are at the forefront of the use of be like as a way of introducing direct quotation (e.g., I’m like ‘Oh my God . . .’ ). As discussed by Romaine and Lange (1991) this usage is clearly innovative yet rapidly becoming the new norm, replacing the older use of go (e.g., And then I go ‘Oh my God . . .’ ) in the same function (on go see Butters, 1980). In such a case, one can talk of “change in apparent time,” since no real temporal “slices” are at issue; rather, the time dimension is reflected in the different ages of the speakers examined, with older speakers presumably reflecting usage fixed when they were young, and younger speakers reflecting innovations in usage.4 At the same time, though, not all variation indicates change in progress; there can be stable variation, even if age-grading is evident. For instance, the use of a term of address such as Mommy shows age-grading in American English, with younger speakers using it, while older 76 Amalia Arvaniti and Brian D. Joseph speakers tend to use Mom or Mother. This variability, however, reflects a maturational aspect of each individual speaker’s development rather than a shift in usage in the speech community as a whole. However, even when techniques for identifying changes in a language (such as looking for age-grading in variation and interpreting it) are employed, it is usually necessary to have some fixed point of reference against which to consider the data in order to see that a change has taken place—or is taking place—and what the direction of that change is. Finding a suitable point of reference for comparison often involves the interpretation of written records or an inference about what a likely prior stage was like (e.g., based on comparisons of related or similar situations). Interpreting data from earlier stages can be straightforward where written texts provide clear indications, as with elements of sentence syntax evident in (1). However, getting at fine details of pronunciation from written records alone is often quite difficult. Most importantly, spelling does not always reflect current pronunciation. The use of the multiple breathing and accent marks in the spelling of Greek into the second half of the twentieth century is a case in point, as these reflect simply a continuation of Ancient Greek spelling practices that have had no phonetic reality after the period of the Greek Koine. In such cases, the use of oral data, such as the recordings of folk and rebétika songs used here, if available, can be indispensable. Before we present our findings from the examination of such songs, however, we need to present some details on the situation with the nasal plus stop combinations that were the focus of our investigations. The variability in the pronunciation of mp, nt, gg/gk In the above mentioned earlier study by Arvaniti and Joseph (2000), the results of which are summarized below, we used variationist techniques on data collected in the 1990s from a sampling of Greek speakers of different ages, thus looking at “apparent time,” in order to assess the situation with an aspect of the pronunciation of contemporary Standard Modern Greek. In particular, we investigated the ways in which our sample of thirty speakers produced the phones represented in Greek orthography as mp, nt, gg/gk. These spellings reflect not one but several possible pronunciations. Thus, the pronunciation of mp, nt, gg/gk can include a combination of a nasal consonant and a following “voiced stop consonant,”5 making them similar to the English sequences mb, nd, ng in words such as limbo, end, and finger; a Greek example of this pronunciation would be sugkalÊptv (to cover up). However, mp, nt, gg/gk can also be 77 Early Modern Greek /b d g/ pronounced simply as voiced stops—as in the words mpÊra (beer) or ntomãta (tomato)—or as a combination of a nasal consonant and a voiceless stop consonant (i.e., like the sequences mp, nt, nk in the English words limp, dent, and ink respectively), e.g., in memptÒw (blameworthy). For convenience, we are going to refer to the three different phonetic categories mentioned above, nasal consonants, voiced stop consonants and voiceless stop consonants as N, D, and T respectively. The mp, nt, gg/gk combinations have largely arisen in Modern Greek from Ancient Greek clusters of consonants involving nasals (N in our notation) and two types of stops, voiced and voiceless ones (D and T respectively). Examples of these two types of clusters include the words ãndra (man) and p°nte (five) in Ancient Greek. In the first word, nd was pronounced nd (as in the English word send; ND in our notation); in the second word, nt was pronounced nt (as in the English word cent; NT in our notation). The pronunciations of these ND and NT clusters are known to have begun to merge to ND by the sixth or seventh centuries A.D. at the latest and most likely even earlier than that (Tonnet 1993: 40–46; Horrocks 1997:112). From this ND outcome in Middle Greek, several discrete outcomes are found in regional dialects of Modern Greek6 (Mirambel 1933; Mirambel 1959; Newton 1972). The key ones for our purposes are the following: (i) preservation of ND word-internally and simplification to D word-initially, e.g., p°nte with ND, but ntÊnv (to dress) pronounced with D, as in the Athenian standard and in most dialects of the Peloponnese and Northern Greece (Newton 1972:94); (ii) simplification to D in all positions, e.g., p°nte and ntÊnv both pronounced with D, as in Cretan, Thracian, Eastern Macedonia dialects, Thasos, Samothraki, Lesbos, Skiros, and Samos (Newton 1972:95). It is also important to note that occasional instances of variability internal to particular dialects (that is, from speaker to speaker) have been documented for regional Greek of the late nineteenth and early twentieth centuries. For example, Thumb (1891:107–108) describing the dialect of Aegina notes that “n is almost completely missing so that when I was transcribing the words I often doubted whether I should write the n or not; in general nt or mp . . . are very close to . . . d, b, but differ from [them] only in that a light nasal resonance precedes them during production” [our translation]. This quote suggests that although Aegina speakers for the most part had the nasal pronunciation they must at least occasionally have omitted the nasal. Similarly, Dawkins (1916:69, 81) in his description of Cappadocian Greek in the early twentieth century notes that one can hear mb, b, and even p as outcomes of historical mp, showing yet another source of variation within one and the same dialect with regard to the realization of these phones. Newton 78 Amalia Arvaniti and Brian D. Joseph (1972:95), describing a later stage of the Greek language, also mentions that there are dialects which show fluctuation in terms of how much nasality is used in the pronunciation of these phones and how often nasality occurs. As far as the standard language is concerned, it has often been assumed that the nasal is always pronounced (e.g., Mirambel 1933; Newton 1972), except word-initially (e.g., Mirambel 1933:157; Householder 1964:20; Newton 1972:96). However, as with the regional varieties of Greek, variability in the Standard has been reported in the literature on Greek linguistics, beginning in the 1960s. Householder (1964), for instance, mentions that in Greek, at least as spoken by educated Athenian speakers, there are three types of words: (i) words that fluctuate between D and ND; e.g., liontãri (lion), pougg¤ (purse), tsamp¤ (bunch); (ii) words that are pronounced (almost) exclusively with ND; e.g., kãmpow (plain), goggÊli (turnip), ant¤di (endive); (iii) words that are pronounced (almost) exclusively with D; e.g., mpampãw (dad), ntantã (nanny), str¤ggla (shrew).7 Householder’s fluctuating words would thus show variation in their pronunciation, and the system overall would show variation in that there was no single overarching generalization about how the voiced stops were pronounced. Similarly, Newton (1972:95) notes that in Athens itself, “where the ‘standard’ pronunciation would be expected” (97), the nasal is rarely perceptible at least as far as fairly rapid speech is concerned. Newton’s comment suggests that in his view there was stylistic variation between D and ND in the standard; that is speakers show a tendency to simplify ND to D word-internally in casual speech. Similar views are also expressed by Kazazis (1968; 1976). On the other hand, quantitative studies (Charalambopoulos, Arapopoulou, Kokolakis, and Kiradzis 1992; Pagoni 1989; Mikros 1997) suggest that realization depends largely on two factors: first, age, with older speakers using more instances of ND than younger speakers, and second, education, with more educated speakers using ND more than the less educated ones. In addition, we should stress that these pronunciations occur not only within words, but also when the nasal and stop occur across a word boundary, as in the case of combinations such as ton pat°ra (the father, acc.) or thn kalosÊnh (the goodness, acc.). Similarly to wordinternal instances, these nasal plus stop sequences also show variation. Although in very careful speech they may be pronounced as sequences of nasal and voiceless stop, NT, typically they are pronounced as nasal plus voiced stop, ND, or as a plain stop, D, depending on the dialect. These historical and contemporary observations raised the question of whether the variants involved represent a case of stable variation or instead a case where variation reflects a change in progress (hence a 79 Early Modern Greek /b d g/ variable, notated ND). Our earlier study (Arvaniti and Joseph 2000) was designed to answer this question. The findings of Arvaniti and Joseph (2000) The findings that emerged from Arvaniti and Joseph (2000) were quite revealing, and suggested that there is a real change going on in Greek with regard to (ND). We examined all the factors that previous studies had suggested as, but had not conclusively shown to be, relevant for the variation in the realization of (ND), namely style, age, gender, education, social class, and linguistic context. Specifically, our study included data in formal and informal style (reading of an especially composed text and approximately 30 minutes of casual conversation) elicited from a sample of 30 native speakers of Athenian Greek (15 men and 15 women) stratified according to class, education level, and age. The data yielded 5396 instances of (ND), 3660 from the two repetitions of the text and 1736 from the conversational data. These tokens were classified into three variants, ND (for tokens exhibiting both nasality and voicing, e.g., amp°li (vineyard) pronounced with mb), D (for tokens without nasality but with voicing, e.g., nt°fi (tambourine), pronounced with an initial d), and NT (for tokens with nasality but no voicing, e.g., thn Tetãrth (on Wednesday) when the sequence –n T– is pronounced nt). The percentage of each of the variants was computed separately for each speaker and context, i.e., separately for (ND) in word-initial position, word-internal position, and across a word boundary, since differences among these positions had already been noted in the literature. The computed percentages formed the basis for our statistical analysis of the data. Our results showed that the pronunciation of (ND) depended primarily on linguistic context and age, and to a much lesser extent on style and gender. Our results confirmed traditional accounts that wordinitial voiced stops are virtually always pronounced without nasality. In addition, in word-internal (ND), variation in the use of the ND and D variants shows a strong correlation with age, with speakers below the age of 45 displaying a dramatic reduction in ND pronunciations when compared with older speakers. On the other hand, gender, education, and class did not affect the speakers’ choice of variant (a finding that typically suggests that a change has already been completed, as shown by Labov 1963 and 1994, for instance). Significantly, style did not affect (ND) realization, except in the case of older speakers, who showed an increase of ND usage in reading: thus these speakers used ND 56% of the time when reading words such as d°ntro (tree) and ant¤yeta (in 80 Amalia Arvaniti and Brian D. Joseph contrast), but this percentage decreased to 42% during spontaneous conversation. We also found that the pronunciation of (ND) across a word boundary was affected by age, but that within each age group the variable was influenced in different ways by gender and style of speech. In the youngest age group these factors did not affect (ND) realization, and in the majority of cases the variant used was D. In the oldest age group, style affected the choice of variant, resulting in higher ND and lower D percentages in reading than in conversation for both men and women. In the middle age group, on the other hand, women showed an increase of NT in reading compared to conversation; this increase was at the expense of the D variant, while women’s percentage of ND pronunciations remained the same in the two styles. Unlike the youngest and oldest age groups, women in the middle group behaved differently from men, whose choice of variant was not influenced by style; this difference suggests that for the middle age group there is sociolinguistic significance in the choice of (ND) variant across a word boundary, and that women are more sensitive to it. The overwhelming effect of the age factor compared to all other factors suggests that the pattern of stable variation depicted in most traditional grammars and descriptive works (e.g., Mackridge 1990; Newton 1972), in which ND is the formal and D the informal variant, is changing, that is to say, showing change in progress. It appears from our data that for the majority of the younger speakers, ND is no longer a prestigious marker of careful speech, and D forms are no longer stigmatized. On the contrary, our results suggest that prenasalized voiced stops may have actually begun to disappear from Greek, or more accurately, from the speech of the younger speakers of Standard (Athenian) Greek, resulting in an age-grading phenomenon. What makes the observed pattern of some interest is (a) the abruptness of the change, which seems to have taken place within one generation, and (b) the direction of the change, namely the fact that the currently dominant variant, D, is traditionally thought of as less prestigious. In Arvaniti and Joseph (2000) we proposed that these phenomena are due to the overwhelming political changes which took place in Greece in the mid-seventies and led, on the one hand, to social changes, and on the other, to the official abolition (in 1976) of Katharevousa, the “high” variety in the long standing Greek diglossia, in favor of Dhemotiki, the “low” variety in the diglossic situation. Specifically, it appears that before the mid 1970s, in addition to ND, the prestigious variant linked to Katharevousa, an Athenian “low” standard with D as its reflex for older ND was emerging among those upwardly mobile strata of society— always considered innovators (Labov 1980)—that after the Second 81 Early Modern Greek /b d g/ World War formed what Lytras (1993) terms the “new middle class.” Eventually the D of the low prevailed for socio-political reasons, namely the end of the military government and the subsequent abolition of Katharevousa as Greece’s official language (the timing of these events correlates with the sharp age division in our results). The reason D prevailed was that Katharevousa related norms were generally rejected because of the connection of Katharevousa with the military government (see Frangoudaki 1992). Using folk and rebétika songs Using variationist methods on contemporary data was appropriate for the study of this phenomenon, because it is especially difficult to draw clear inferences about ND combinations for earlier stages of the language, owing to the vagaries of Greek orthography; as suggested above, conventional spellings for these combinations do not always reflect pronunciation accurately. Moreover, despite the occasional reports of regional intra-dialectal variation in pre-contemporary Greek noted above, there was no reason to believe that there was such variation in the standard language. It is not entirely clear what “standard language” would mean in the context of nineteenth and early twentieth-century scholarship on Greek; for our purposes we assume that descriptions of Greek that do not make specific reference to regional dialects refer in fact to the author’s conception of standard Greek (e.g., Mirambel 1933). With that definition in mind, we note that no description of Greek from this era says anything about pronunciations of ND clusters other than, essentially, that ND occurred in word-internal position and D in wordinitial position.8 Still, our study, being an investigation of contemporary usage, could not possibly provide any hard evidence bearing on the state of affairs with nasal plus stop combinations in pre-contemporary Greek. In order to have a reference point against which to judge the patterns of variation found for late twentieth-century Greek, we made reasonable assumptions concerning the earlier situation, drawing on historical information and facts about the realization of these combinations in various modern dialects. Still, direct evidence of pronunciation from earlier stages was lacking and our conclusions were only as strong as our inferences about these earlier states. We thus turned to a heretofore unexploited source of directly observable—as opposed to inferable—data bearing on how (ND) was actually pronounced in early twentieth-century Greek. The data in question come from early recordings of folk and rebétika songs, and thus, since we are not dealing with written documents that have to be 82 Amalia Arvaniti and Brian D. Joseph philologically interpreted, we are in a better position to say with some certainty just what the range of pronunciations were that were available several generations prior to the speech of our sample in the earlier study. Our source was tapes and CDs of original recordings of rebétika and folk songs from the 1910s to the 1930s. Although we do not know what the earliest (nonwritten) record of Greek is, given that the technology of recording speech for listening has emerged only within the last 120 years or so, we expect that recordings from the 1910s, which are the earliest we have, must be among the earliest available sound recordings of the language. We note that there may be other possible sources of recorded oral Greek from that period, such as radio recordings. However, there are certain advantages to using the song corpus. First, song recordings are easily accessible, since one can buy them in stores instead of having to search through archives for which permission might be needed. Second, they are more likely to reflect real and natural usage than radio recordings, which typically are more formal, or films, which do not provide real data about pronunciation and usage, but rather, reflect attitudes speakers hold about variation and stereotypical usage.9 Third, the recordings we examined provide two types of stylistic information: (a) there are two styles of singing involved—folk songs and rebétika— and (b) there are interjections of the singer to the orchestra, so that the same person is recorded both singing and speaking. Fourth, the recordings provide us with a variety of speakers and dialects. Moreover, since the singers are well known, their birthplace is almost always known and provided in biographical notes accompanying each recording. This is very different from the situation that one faces with radio recordings, where the announcers are generally anonymous so it is impossible to ascertain their origins (and hence their accent). There are of course some necessary caveats in the use of early rebétika/folk song recordings as linguistic evidence. First, the recordings are obviously not of high quality. This would be a potential problem for certain types of detailed phonetic analysis, but would be unlikely to affect the variable under consideration (and in any case, bad sound quality is an inevitability when working with recordings of that period). Second, with performed language as in singing, stylized pronunciations that do not directly reflect actual spoken usage are possible (cf. pronunciation of final schwa in French songs, in words such as chose (thing), which are pronounced without the final schwa in regular speech). Here though there is no reason to think that singers would target (ND) for such stylization. We have some evidence that this is so, 83 Early Modern Greek /b d g/ since the same singers were often recorded singing both rebétika and folk songs, and the comparison of their pronunciation of (ND) in the two singing styles shows no evidence of conventional singing-style related differences (note, in contrast, that the folk style involves mannerisms that seriously affect vowel quality; these mannerisms are absent from the rebétika songs).