1 Introduction

In Modern Hungarian, vowels take part in two types of harmony patterns: backness harmony and labial (or rounding) harmony (for a recent survey and analysis see Siptár & Törkenczy 2007: 157–170). These are exemplified in (1a) and (1b), respectively, from Standard Hungarian (the authors’ native variety).

Both types are progressive root-controlled harmony, but they differ from each other in important ways. Backness harmony is the older of the two, traced back as far as Proto-Uralic (Suomi 1983; Aikio 2022). It is also much more pervasive: it is triggered by all vowels, undergone by most vowels (though the front unrounded vowels [i iː eː ɛ] can be neutral), and it actively manifests itself in widespread alternations in many of the suffixes. By contrast, labial harmony has a much narrower scope; although it is a stable pattern, it only affects the short mid front vowels [ɛ] and [ø], and it seems to have appeared only in the Old Hungarian period.¹

The fact that we set out to explain in this paper is that the emergence of labial harmony, i.e., the progressive spreading of the feature [labial] in Late Old Hungarian results from an apparent change in the earlier direction of spreading for the same feature. In Early Old Hungarian [labial] demonstrably spread regressively, from final syllables to vowels in the preceding syllables; in Late Old Hungarian, however, the spreading began to stabilize into the progressive pattern that has been preserved ever since. The question we address is this: what caused the change in the direction of the spreading?

The broader context of the issue is the much-discussed question of what motivates vowel harmony phenomena. The formal side of vowel harmony is commonly described in phonology as feature spreading, i.e, a subcase of assimilation (e.g. van der Hulst & van de Weijer 1995; Krämer 2003; Rose & Walker 2011). In terms of its motivation, two possible avenues of explanation have been explored in the literature. A coarticulatory basis is advocated by Ohala (1993; 1994a; 1994b, cf. also Mailhot 2010; 2013), with perceptual factors playing a role at the later stage of phonologization. An acoustic and perceptual basis is assumed by others (Nearey 1980; Suomi 1983; Kaun 1994; 1995; 2004), which explains among other things why labial harmony never occurs on its own (mainly because it is an enhancement of the F2 differences achieved primarily via a front–back contrast).² Przezdziecki (2000) demonstrates that coarticulatory phenomena and vowel harmony in Yoruba operate in the same direction with only a difference of graduality between them and concludes from this that harmony is indeed based in coarticulation. Beddor & Yavuz (1995) show that in Turkish vowel coarticulation and vowel harmony can have opposite direction; from this they conclude that while vowel harmony is based in coarticulation, its phonologized form, i.e. vowel harmony in the strict sense of the word, may be divorced from its original motivation and obey different regularities. Hansson (2001) as well as Gafos (1996; 1998; 2021) focus on a crucial distinction between spreading and copying, both of which can lead to harmony-like phenomena, but of which only the former involves coarticulation, i.e., contiguous articulation of a gesture extending across intervening segments. They argue that vowel harmony typically involves spreading, whereas consonant harmony – with the possible exception of coronal harmony – involves copying.

What will be presented in our paper resonates closely with the findings of Beddor & Yavuz (1995): coarticulatory spreading and phonologized vowel harmony go in opposite directions. There are two important differences, however. One is that in Turkish word stress is final, which influences the direction of coarticulation (but not that of vowel harmony); in Hungarian word stress has always been initial. The other difference is that in Beddor & Yavuz (1995) the two phenomena were studied in one and the same period, whereas in Hungarian a diachronic sequence is assumed.

Thus, our main concern is directionality; the hypothesis we advance is the following. Early Old Hungarian regressive spreading of labiality did not affect lexical cues (the recognizability of lexical items), since [labial] was a redundant or at most marginally contrastive feature; the spreading was probably motivated by simple coarticulatory mechanisms. However, as [labial] became a robust contrastive feature in the course of the Old Hungarian period due to a variety of sound changes, it aligned with backness (an inherited robust contrast) in its spreading direction to enhance lexical recognizability. The latter point crucially hinges on the general morphological character of the language throughout its documented history, viz. widespread agglutination without prefixation, lexical morphemes being on the left of word forms, followed by suffixes. The argument also implies that while labial harmony owes its origin to a pattern rooted in coarticulatory mechanisms, in its mature form it also acquired an important perceptual motivation. The central claim that this paper makes is that Old Hungarian labiality stopped spreading regressively onto roots and became a rightward-spreading harmonizing feature when it became lexically specified and thus crucial to identifiability.³

We illustrate the role of harmony in lexical recognizability at this point with two examples from Modern Hungarian. The type of harmony crucial to the first is backness harmony. In the most recent period there is a noticeable (and perhaps growing) overlap between the realizations of two lexically and etymologically distinct vowels, [ɛ] and [aː] for many speakers of Standard Hungarian (see e.g. Gósy & Siptár 2015). Both vowels can be realized as [æ] with variable length depending on the consonantal environment, syllable structure and other factors. This leads to frequent confusion between minimal pairs such as hely [hɛj] ‘place’ and háj [haːj] ‘fat’ or mert [mɛrt] ‘dare pst’ and márt [maːrt] ‘dip’. Since, however, independently of their actual realization, [ɛ] harmonizes as front, whereas [aː] harmonizes as back, suffixed forms are no longer ambiguous. This is shown in (2), with the overlapping vowels indicated as half-long [æˑ].

The second example concerns labial harmony in different Modern Hungarian dialects. The multiplicative-temporal suffix is invariant [ser] in certain dialects (Hegedűs 2001: 358), but in the majority of varieties (including Standard Modern Hungarian) it harmonizes both in backness and in labiality. Because of this, the words meaning ‘once’ vs. ‘five times’ (written egyszer and ötször in the standard orthography) have the forms given in (3) in the two types of dialects, respectively.

In this particular pair of words, the difference is encoded only in the labiality of the vowels. If because of noise or lack of attention or some other masking factor the quality of the first vowel is missed by the hearer, in dialect A it is not recoverable, and information is lost. In dialect B, however, the stem vowel is recoverable because the harmonizing suffix indicates whether it was labial or not, hence the lexemes remain identifiable.

The specific understanding of the perceptual motivation of vowel harmony that is crucial to our hypothesis has this disambiguating capacity in its focus. The insight is essentially identical to that articulated in Baković (2003). In that paper root-outward harmony is seen as instrumental in achieving stem-identity, a property that ensures lexical identifiability at the expense of introducing alternations within affixes. This relation is encapsulated by Baković’s (2003: 10–11) stem precedence generalization.⁴ It is clear that contrastive features play a greater role in the identification of lexical items than redundant ones; hence the promotion of a redundant feature to a contrastive status can change the way it interacts with the morphology. As we shall argue, this is what we see in the case of labiality in Old Hungarian vowels.

The paper is structured in the following way. In section 2 we discuss the nature of the data and present the Early Old Hungarian vowel system, which serves as our starting point. Then in section 3 we discuss those changes that led to a significant increase in the lexical frequency of labial vowels (especially front labial vowels, the chief protagonists of this paper). In section 4 we give a detailed presentation of regressive labialization, also highlighting those factors that were to prove essential in the reversal of the direction of spreading. In section 5 we present the emergence of progressive labialization, i.e., labial harmony that is almost identical structurally to its Modern Hungarian continuation. In section 6 we relate our analysis to two issues that are prominent in current theoretical debates. Section 7 concludes. In the Appendix and the accompanying files, the data are presented and explained in detail.

2 The Old Hungarian data and the vowel system

The Early Old Hungarian period (henceforth EOHu) covers the 10th–14th centuries AD. From this time very little continuous text remains; what we have is mostly onomastic material and glosses embedded in Latin, and occasionally Greek, texts. The Late Old Hungarian period (henceforth LOHu) covers the late 14th, the 15th and the early 16th centuries, its end conventionally identified as coinciding either with the destruction of Hungarian statehood by Ottoman invasion (1526) or with the printing of the first Hungarian book (1541). From the LOHu period we have extensive surviving material in the form of more than 40 codices (some incomplete) and a substantial corpus of minor texts (letters etc.). Most of the surviving OHu texts are freely accessible in searchable form in the Old Hungarian Corpus (http://omagyarkorpusz.nytud.hu/en-intro.html, cf. Simon 2014; Simon & Sass 2012), an online database which we used extensively in our research.

It is to be acknowledged that some of the phonological developments to be discussed in the paper, especially those in section 3, took place in EOHu, and therefore the pool of data on which their reconstruction is based is far smaller than would be ideal. The EOHu onomastic material, which could potentially provide a larger data pool, has not yet, as of this date, been organized into a usable data base. As is often the case with reconstruction, hypothesized changes are distilled from later patterns, so these hypotheses are not data-driven in the strongest sense of the word. Nevertheless, they still provide, at present, the most plausible explanations for the patterns seen at later stages. It is also true that numerically speaking the LOHu data pool is not large either, at least it is not comparable to what is available for some better documented languages. The hypotheses and the arguments we formulate are based at certain points on a higher number, at others on a smaller number of data. We are still confident that the larger picture that emerges will present a plausible scenario.

The EOHu vowel system is usually reconstructed as shown in Table 1 (cf. Kiss & Pusztai 2003: 336, 344). The front/back contrast as well as the three-way height contrast is beyond doubt. The status of the back (or at least non-front) high unrounded vowel (here [ɯ]) is a moot question: there is a tradition of assuming such a sound, but it has never been generally accepted and has come under serious criticism recently (most prominently in Kis 2005). The contrast between [i] and [y] most probably had very low functional load and involved a great deal of variability (cf. Proto-Finno-Ugric *kitV or *kütV ‘middle, gap’ > Hu [køz] as per Benkő 1993–1994 s.v. köz); it is possible that some dialects of Pre-OHu or Early OHu had no [y] sound at all (cf. Bárczi 1958: 34–35).

It seems that no long labial vowels can be securely reconstructed at least for the earliest stage of EOHu (Bárczi 1958: 82; E. Abaffy 2003: 344). During this period, however, many long vowels appeared via various lengthening changes and contractions of short vowels with tautosyllabic approximants ([Vj, Vɣ, Vw] > [Vː], e.g. *[fɛɣ] > *[fɛɥ] > [føː] ‘head’, *[low] > [loː] ‘horse’, *[mundaj] > [mondaː] ‘he said’).⁵ Many of these recent long vowels were labial, and these, along with the vowels discussed in the next section, came to play an important role in the emergence of labial harmony.

Table 2 shows the vowel system at the end of the OHu period, the stage that was reached by the time the changes discussed here were completed or at least had progressed significantly (Bárczi 1958: 107, E. Abaffy 2003: 346). Variation among dialects was widespread, but many details are unclear. Some dialects may have lacked front rounded vowels even in this period; some may have retained non-labial short [a]; some may have displayed a severely restricted length contrast, especially among the labial and/or the high vowels. By and large, when abstracting away from dialectal variation, the Modern Hungarian vowel system is nearly identical to the LOHu system; Modern Standard Hungarian has lost only [e] and [ɛː] and has not added any new vowels.

Table 3 shows the written forms for the OHu vowels. Since vowel length was not marked in the period, in the left column of the table we only indicate vowel qualities. Diacritics marking nasal consonants in OHu texts are not included.

3 The proliferation of labial vowels in Old Hungarian

Next to the two inherited labial vowels [u o], further labial vowels appeared in the OHu period via a series of sound changes; borrowings — mainly from Turkic — also increased their token frequency (see below). The precise chronology of the relevant sound changes is not always securely established and may have varied from dialect to dialect; but their outcomes are amply attested in the LOHu period at the latest. The changes are the following (Bárczi 1958: 47–61; E. Abaffy 2003: 323–327).

(i) Next to labial consonants, the short front vowels [i e] were frequently labialized early in this period; examples are given in (4), including inherited as well as borrowed words.

In EOHu [kørtveʎ] from Turkic *kertweliɣ (Róna-Tas & Berta 2011: 602), the [w] affected the preceding vowel across a consonant cluster. In LOHu [ʦøvek] from MHG zweck an epenthetic vowel was inserted to break up the complex onset, a non-existent structure in OHu. Such epenthesis was systematic in the language, and the vowel was either a copy of the following vowel, or [i] or [e] (for a detailed discussion see Nyirkos 1993); in the case of [tsøvek] the labial quality of the epenthetic vowel can be attributed to the following [v].

(ii) In the environment _[l]C[coronal], [e] and [ɛ] were labialized to [ø], and [a] was labialized to [o], as the examples in (3) show (cf. Dudás 2002; this may have taken place in the latter part of the EOHu period, but for many affected words attestation is later).

(iii) In an unconditioned change, [a] was labialized to [ɒ] by the end of the EOHu period, with long [aː] remaining intact: EOHu [kard] ‘sword’ > LOHu [kɒrd].¹⁶

(iv) In word-final position, a sequence of a vowel and an approximant systematically developed into a long vowel. We have seen a few examples above in section 2; two more examples are given here for front labial vowels, since these are the ones of immediate interest to us. The general pattern here was *[iɣ *iw *eɣ *ew *ɛɣ *ɛw] > [iɥ eɥ] > [yː øː]; when a vowel-initial suffix followed, the contraction to long vowel did not take place, which led to extensive alternations. For examples see (6).

The four changes briefly described here significantly increased the number of rounded vowels in the language in various phonological environments. To these one must add the many loanwords that were adopted in the Pre-OHu or the EOHu period; these further increased the number of front labial vowels either directly or by providing input to the above sound changes. Here belong borrowings mainly from the Turkic languages with which speakers of Hungarian came into contact in the steppe region (Pre-OHu period) and the Carpathian basin (EOHu period). In (6) above the word [ɛrøː] ‘strength’ is a Turkic loanword; some more examples are given in (7).

In the next section we turn to yet another labialization process, that involving word-final vowels, whose labiality exerted an assimilatory influence on preceding vowels and thus introduced regressive spreading. This needs to be discussed in a separate section because it played a crucial role in the development of labial harmony and because it showed a complex interaction with morphology.

4 Final and regressive labialization

The available data show quite clearly that short word-final vowels became high and then labial within the period spanning Pre-OHu and EOHu (Bárczi 1958: 17–24; E. Abaffy 2003: 321–323). These vowels harmonized in backness: they were realized as [u] after back vowels, and as [y] after front vowels. The closing was apparently exceptionless; the labialization shows a few exceptions (i.e. [i]-final words). The effects of the two changes are seen both in inherited and in borrowed vocabulary. However, since the final short high vowels themselves were lost by the latter half of the thirteenth century, their attestation in terms of numbers is rather poor. Examples are given in (8).

In suffixed forms, we see extensive variation in the labiality of stem-final vowels throughout the OHu period. Thus, for instance, the plural of the EOHu word [bereɡy] > LOHu [bereɡ] or [berek] ‘grove’ can be either [ber(e)kek] or [ber(e)køk], where the final [k] is the plural suffix.²⁵ The most commonly accepted explanation for this variation is analogical transfer of the labial vowel from unsuffixed forms, where the vowel was in final position and hence labialized. This explanation, elaborated in great detail in D. Bartha (1964), rests on the three observations introduced in this section: (i) word-final original short vowels are invariably high and overwhelmingly round in EOHu, (ii) the same word-final short vowels are lost by the mid-thirteenth century, (iii) the same short vowels show variation in rounding and/or height before suffixes, where they were not lost. But the explanation also rests on a crucial assumption, viz., that in the case of variation it is the unrounded and/or lower vowel that is the original; to take the previous example, [ber(e)kek], rather than [ber(e)køk], is assumed to be the older form of ‘groves’.

This assumption itself is based on an observation: from the earliest period of attestation onwards, short vowels before suffixes (the original stem-final vowels) are lexically specific and cannot be derived from the phonological environment in a significant — and conceiveably very old — part of the word stock. This is true even in Modern Hungarian, but is exemplified here by five EOHu adjectives formed with the suffix s: [feːɲyʃ] ‘shining’ vs. [eːzeʃ] ‘sweet’ vs. [sɛrɛlmɛʃ] ‘beloved’ vs. [utɒʃ] ‘having a road’ vs. [vilaːɡuʃ] ‘light-coloured’.²⁶ It is also to be noted that even in Modern Hungarian, many short monomorphemic stems of Finno-Ugric provenance are lexically specified as taking suffixes with a low vowel (e.g. MoHu [haːzɒk] plural and [haːzɒt] accusative of [haːz] ‘house’, [lovɒk] plural and [lovɒt] accusative of [loː] ‘horse’). The majority pattern, to which most loanwords adhere, is to take the same suffixes with a mid vowel (if any, e.g. [ɡaːzok] plural and [ɡaːst] accusative of [ɡaːz] ‘gas’). Thus, the low stem-final vowel certainly represents a very old pattern. Generally speaking, with respect to Old Hungarian, it is concluded on the basis of such observations that when not high, these vowels represent the original variant of the stem-final vowel.

For the argument we shall adhere to this assumption, though things are complicated by several factors, not least among them the adoption of loanwords, which often received stem-final vowels on the analogy of the inherited vocabulary.²⁷ But the picture is muddied further by the general lowering of short vowels whereby many of the high vowels became mid vowels in LOHu (witness [ber(e)køk] rather than *[ber(e)kyk] as the plural of [berek] in LOHu).

As was first systematically argued in Losonczi (1915) (for a recent discussion see Dudás 2004), the labialization of final short vowels and the analogical introduction of these labialized vowels into non-final (i.e. pre-suffix) position had a pervasive effect on the phonological form of words in many dialects of OHu. This effect can be described as the regressive labialization of front vowels, a tendency that becomes visible already in the EOHu period. It became a widely and robustly attested phenomenon, resulting in pervasive but phonologically patterned variation. The geographical aspect of this variation, while interesting in the context of historical dialectology, will not be discussed in this paper (but see Juhász 2018). As was said above, when reconstructing the phonological origins of labialization, Losonczi attributed the initial effect to the final short labial vowel, which itself was lost before LOHu times. In fact, long final labial vowels, usually resulting from the contraction of vowel + approximant sequences, as shown in (6) above, had the same effect (Berrár 1961). The attested variation from the OHu period is exemplified in (9).

The variation is extensive, but patterned: descriptively, a labial vowel resulting from regressive labialization implies a labial vowel to its right but not to its left.³⁵ The word for ‘riverside, grove’ can be any of the forms listed in (9), but it cannot be ⁺[børeɡ] or ⁺[børek]; ‘noble’ cannot be ⁺[nømeʃ], ‘delightful’ cannot be ⁺[ɟøɲeryː], and so on. The forms marked here with ⁺ are either completely absent from the OHu corpus or are vanishingly rare.³⁶ Even in the 20th century Hungarian dialects showed extensive variation in labiality among the front vowels which tended to pattern along the same lines (Imre 1971: 203–212). This latter observation, as well as the variation observed in OHu, was crucial for Losonczi’s argument that the feature [labial] had spread regressively from the final syllable of the word.

In view of the fact that Hungarian most likely always had initial word stress, it is frequently claimed in the literature that both final and regressive labialization were essentially a form of vowel reduction (Bárczi 1958: 20; E. Abaffy 2003: 330). Indeed, in word-final position it seems to have been the precursor to the complete loss of short vowels (see the data in (8) above). Nevertheless, whether we regard it as reduction in any sense or not is immaterial for our purposes. It may have been reduction word-finally, but the regressive spreading affected stressed vowels as well (cf. the first syllable in [nømøʃ] ‘noble’, [ɟøɲøryː] ‘delightful’ etc. in (9) above), which would be highly unlikely in the case of a reductive process.

As was said above, regressive labialization was a robust but variable tendency. Furthermore, it interacted with two other tendencies which proved crucial in the emergence of the progressive spreading of labiality. One was the paradigmatic levelling of stem-final vowels, the other was the set of changes that labialized front vowels in the stressed — or at least non-final — syllables of words. The latter set of changes was presented in section 3 above; the paradigmatic levelling, or analogical transfer, was introduced earlier in this section.

Word-final short vowels were lost in the course of the thirteenth century at the latest — indeed after certain consonants considerably earlier. As was seen above, final short vowels had been labialized in most words before being apocopated altogether. When such a stem was suffixed, however, the phonological conditions (viz. word-final position) were not present, hence alternations between a labial and a non-labial vowel resulted (EOHu [bereɡy] ‘grove’ vs. *[bereɡek] > LOHu [ber(e)kek] pl). Subsequently these alternations were variably levelled within paradigms, which resulted in labial front vowels appearing before suffixes as well. Examples from LOHu are given in (10).

This paradigmatic levelling and redistribution of stem-final vowels proceeded with many exceptions and varies among dialects to this day. Add to this that several of the labializing changes presented in section 3 had exceptions or were undone via morphological levelling, and this explains why the resulting picture appears chaotic at first. There were three variable processes interacting with each other: regressive labialization, morphological levelling in stem-final position and the conditioned labializing changes affecting non-final vowels. There is, however, a very important observation to be made with respect to the precise relation between the first two of these processes: the regressive spreading of labiality presupposed morphological levelling to a labial vowel in stem-final position, but not vice versa.⁴⁰ This generalization emerges quite clearly from the variation apparent in LOHu, as is exemplified by the data in (11). These show the attested variant forms of the words [ɟɛrmekek] ‘child-pl’ and [piʃpekek] ‘bishop-pl’⁴¹ (singular [ɟɛrmek] and [piʃpek], respectively).

The variants that do not seem to exist are those with regressive labialization but no levelling before the suffix (⁺[ɟɛrmøkek], ⁺[piʃpøkek]).⁴⁵ This is consistent with what one sees in unsuffixed forms (or forms with obscured suffixation). In both types it appears that labiality in a short front vowel had to be licensed either by the consonantal environment, as in [fyʃt] ‘smoke’ or [tølt] ‘fill’ (see section 3), or by a labial vowel in a following syllable, as in [ɟøɲøryː] ‘delightful’ or [kørøst] ‘cross’ (see (9) above). At this stage in the history of the language, in a putative form ⁺[ɟɛrmøkek] neither licensing factor would have been available.

Figure 1 provides a summary of this section by presenting the sequence of changes through the example of the singular vs. plural form of the ‘child’ word. Some simplifications have been made but these do not impinge on the points being made. The fact that some of the forms are unattested (e.g. the ‘child’ word with final [y]) is irrelevant because they represent types (final [y] is attested in other lexical items). The forms themselves are given in the middle two columns, while comments are given in the first and last columns regarding the highlighted parts of the forms.

Figure 1: Conspectus of final and regressive labialization.

5 The emergence of progressive labial harmony

With the token frequency of front (as well as back) labial vowels growing steadily throughout the OHu period, and Turkic loanwords such as [økør] ‘ox’, [øzøn] ‘flood’ also increasing the number of front labial vowels, the conditions were finally given for progressive labial harmony to appear, which would be structurally similar to the inherited backness harmony. Is it possible to get a grasp of the details of the process whereby regressive labial spreading turned into progressive labial spreading? And is it possible to find the motivation for this apparent reversal of direction? We believe that the answer to both questions is positive. Many details that cumulatively contribute to an answer to the former of the two questions can be culled from the time-honoured historical grammars of Hungarian, most comprehensively Benkő (1991; 1992). These, however, need to be placed into a proper context so that these findings will contribute to our understanding of the general properties of harmony systems. As we see it, the process of the reversal of spreading direction took place along the following lines.

It seems certain that many of the suffixes that were affected by stable progressive labial harmony beginning in LOHu had earlier had a labial vowel, that is, they can be reconstructed with [u] or [y] in EOHu. There are two reasons for this. One is that the vowels in several of these suffixes result from the labialization and levelling of etymological stem-final vowels, which were reassigned to affixes subsequent to the loss of word-final short high vowels. The example in (12) shows the most plausible path to the development of the vowelled superessive suffix [on/en/øn] via reanalysis of stem-final high vowels followed by the lowering of [y] to [ø]. Many other suffixes acquired vowelled variants in this way.

The other reason is that some of these suffixes result from the grammaticalization of nouns that had an etymological labial vowel. A good example is the allative suffix [hoz/hez/høz], which derives from a Proto-Ugric noun reconstructed as *kućV ‘side’ (Benkő 1993–1994 s.v. hozzá). Systematic sound changes resulted in *[xuz] > EOHu [huz] with variable lowering to [hoz]; in this period this word is already attested in a grammaticalized form (ozchuz ‘to it’ in the late 12^th c.). At some stage, probably during the EOHu period, it underwent backness harmony and developed the front variant [høz] after front-vowelled stems. Backness harmony for this suffix is securely established for the LOHu period; this simply means that it had aligned by this time with other suffixes not resulting from grammaticalization, and we see the same happening to other suffixes grammaticalized from nouns (e.g. inessive [bɒn/bɛn]).

In several codices within the LOHu corpus the back variant of the allative suffix was [hoz], the front variant [høz]; in some, however, a non-labial variant [hez] appeared (for a full conspectus of the data see the Appendix and the relevant data file). The relationship between the two front allomorphs in some of the corpus is one of free variation, with the labial form dominating. In some codices, however, the progressive harmonizing pattern appears in which after front vowelled stems the labiality of the last vowel before the suffix determines the labiality of the vowel in the suffix itself. But how did the delabialized variants appear in the first place? Their emergence can be explained on the analogy of unlevelled suffixed forms such as [ɟɛrmekek] ‘child-pl’ and [piʃpekek] ‘bishop-pl’ (see (11) above). Since both the morphological levelling of stem-final vowels and the regressive spreading of labiality were variable, but the latter presupposed the former, there was a varied but restricted set of analogical patterns that a suffix of the same type as the allative could follow. With a suffix for which the original/default front variant was labial, these patterns were as given in (13); the example is [ɟɛrmek] ‘child’ in the allative and in the plural.

The crucial point here is that the appearance of the innovative non-labial variant was based on an existing pattern in which there was neither stem-final levelling, nor (by implication) regressive spreading of labiality. Since this resulted in forms in which a non-labial suffix vowel followed a non-labial stem vowel, such forms could easily be reanalysed as resulting from progressive spreading of the [–labial] feature value. It is also important that since there were no (or nearly no) forms of the type ⁺[ɟɛrmøkek] — because of the implicational relation between stem-final levelling and regressive spreading —, there was no pattern on which forms such as ⁺[ɟɛrmøkhez] could be analogically created. This must have contributed to the perception that the relation between a labial vowel in the stem and a labial vowel in the suffix (in this order) was rule-governed.

The changes affecting suffixes such as allative [hoz/hez/høz] exemplify reanalysis resulting in the progressive spreading of [–labial]. The progressive spreading of [+labial] also appears in certain dialects in the LOHu period. In a group of closely related texts called the Hussite Bible, where suffixes similar to the allative show exceptionless labial harmony, we also see the first clear instances of labiality appearing on suffix vowels that are certain to be etymologically non-labial. Examples include the conditional suffix [ne] and the 3^rd singular suffix [e]. In both the original [eː] shortened when in word-final position (this was a systematic sound change in all dialects); in the language variety represented by the Hussite Bible this shortened [e] was progressively labialized after front labial vowels (see 14). Here we see the progressive spreading of [+labial] onto short front vowels next to the progressive spreading of [–labial].

Another suffix that began showing progressive labial harmony with [+labial] spreading, though not in the same dialect at this time yet, was multiplicative-temporal [ser] (see Benkő 1992: 370–372, for a full conspectus of the data see the Appendix and the relevant data file). This was grammaticalized from a noun meaning ‘order, sequence’ originally (Benkő 1993–1994 s.v. szer2). In a large part of the OHu corpus it shows no harmony at all; in some LOHu codices it shows consistent backness harmony, appearing as [sor] after back vowels and as [ser] or [sør] in free variation after front vowels, independently of the labiality of the preceding vowel. Interestingly, in the Hussite Bible, which shows innovative progressive labial harmony for some affixes, [ser] is invariant⁴⁹ — this may reflect its status as not fully grammaticalized in that language variety. There are very few codices (e.g. the Virginia Codex, written in the early 16^th c.), in which [ser] already shows consistent harmony both in backness and in labiality, see (15).

Thus, progressive labial harmony emerged quite consistently for a variety of suffixes both in verbal and in nominal inflection, and it replaced the variability based on regressive labial spreading. In this way the feature of labiality became capable of fulfilling a disambiguating function with respect to lexical stems by spreading over a portion of word forms, but via a process controlled by the stem rather than by the final syllable of words. It became capable of contributing to lexical identifiability, thus assuming a perceptually relevant function next to the coarticulatory pattern in which it originated.

The following period, Middle Hungarian, shows the stabilization of labial harmony for many of the suffixes, among them the recently grammaticalised multiplicative-temporal [ser]. For reasons beyond the scope of this paper, the only such suffixes were those with mid vowels (with [e] later lowering to [ɛ] in the dialects on which Modern Standard Hungarian is based). High vowels never participated in progressive labial harmony, and long vowels did so only marginally for a brief period.⁵¹

6 Perspectives for further research

As we see it, there are two specific aspects of the analysis presented above that may be of further theoretical or typological interest. We offer no more here than tentative remarks regarding these. The first (directionality) has been inherent throughout the discussion; the second (implications for feature theory) less so.

One aspect concerns the direction of the harmonic process. Hyman (2002: 25–26) compares the findings regarding vowel harmony to Hansson’s (2001) findings regarding consonant harmony and comes to the conclusion that both types predominantly display two major patterns of directionality: one is sensitive to morphological structure (root- or stem-controlled), the other is insensitive to morphological structure and typically proceeds right-to-left. The emergence of OHu labial harmony represents a shift from the latter to the former type. As was pointed out earlier, a similar difference was described in Beddor & Yavuz (1995) in Modern Turkish between vowel coarticulation and vowel harmony (see also Conklin 2019). It is interesting that the two languages have different stress patterns, but word stress is redundant in both. It has also been claimed (e.g. Nicolaidis 1999, Recasens 2015) that while stress clearly influences the direction of coarticulatory spreading, this impact can be moderated by intervening consonants. Whether any such effect exists in the OHu data is for future research to determine.

Analyses of Old High German umlaut, a change structurally similar to the first (regressive) phase of labial spreading in EOHu, correctly highlight the role of coarticulatory factors in the process (e.g. Iverson & Salmons 2003; Walker 2011: 111–115). There are, however, some indications that perceptual factors may have played a minor role too; in particular, Walker (2011: 113–114) points this out on the basis of the interaction between umlaut and stress. Perceptual factors may indeed have played a functional role in umlaut since this change frequently accompanied suffixation and thereby contributed to the secondary marking of grammatical differences (e.g. plural vs. singular for nouns, gradation for adjectives, present vs. past for weak verbs). One could perhaps argue for a similar perceptual motivation for OHu regressive labialization if such patterns had ever emerged, e.g. singular [ɟɛrmøk(y)] vs. plural [ɟɛrmekek] or vice versa, singular [ɟɛrmek] vs. plural [ɟɛrmøkøk] with a consistent distinction in labiality corresponding to number. It is just possible (see Appendix) that the language variety of the Érdy Codex represents the latter case, but we cannot be entirely certain. What is then typologically interesting in the OHu case is a shift from a regressive spreading pattern with a basis predominantly in coarticulation to a progressive spreading pattern with a basis probably still in coarticulation but with an important perceptual function superimposed on it.

The other aspect concerns the implications of the phenomena and their analysis for a hierarchical theory of features (as advocated and applied in Dresher 2009; see further Dresher & Zhang 2005; Ko 2011; Oxford 2015; Hall & Hall 2016). One basic tenet of this theory is that there is an intimate relationship between contrastivity and phonological activity: non-contrastive features are typically phonologically inactive. With special reference to harmony phenomena, Dresher & Zhang (2005: 55) say that “only segments with a contrastive specification for a feature [F] can trigger harmony based on [F].” While this clearly implies a categorical notion of contrastive status, Hall & Hall (2016) interpret Dresher’s hierarchical model of features as involving a continuum of contrastiveness, marginally contrastive features being those that are low in the hierarchy and thus do not take scope over other features. The analysis we have presented resonates with this interpretation: indeed, when labiality began to spread regressively in the EOHu period, it was either a redundant or a marginally contrastive feature, just appearing at the lowest level in the feature hierarchy. As Hall & Hall (2016) say, “contrasts with low scope might… be expected to be diachronically unstable—either emerging or disappearing … —and synchronically more susceptible to neutralization… Similarly, but potentially independently, contrasts with low entropy may offer less evidence to the learner, and consequently also be diachronically precarious.” While labiality became a robust contrastive feature in LOHu, it displays a certain amount of dialectal variability even today.

It is interesting to note that Meadow Mari, a distantly related language, also developed a very limited form of labial harmony independently of Hungarian (Fejes 2022: 26–27). In that language, however, labiality takes precedence over backness, another harmonizing feature, i.e., in the [o] ~ [e] ~ [ø] alternating series the back vowel appears only if labial harmony allows it, e.g. the inessive of ‘Moscow’ is [moskva-ʃte] rather than *[moskva-ʃto] because [a] is unrounded, despite the conflict in the backness of vowels (example from Fejes 2022: 26). What this probably means in terms of the hierarchical theory of features is that labiality takes scope over backness in Meadow Mari, whereas the reverse is true in (Old) Hungarian. Of course, a full analysis along these lines is impossible to undertake in this paper.

7 Conclusion

We have presented what we believe to be a plausible account of the emergence of labial harmony in Old Hungarian. The roots of the diachronic sequence of events go back to a stage that is poorly documented, and thus reconstructions need to be projected backwards, in some cases on the basis of very little data. For some phenomena, later and better documented stages already show lexicalized reflexes rather than change in progress, but even these can be helpful in formulating hypotheses about how a given state of the language came into being. The limited amount of data at our disposal necessarily means that no strict quantitative methodology can be applied. Significantly, however, the data that exist do not contradict the claims that have been articulated here, and we are confident that the main thrust of the argument will stand the test of time.

The emergence of progressive labial harmony in late Old Hungarian can be traced to two factors affecting labiality in Pre- and Early Old Hungarian. One is the regressive spreading of labiality from word-final position onto short front vowels, most probably a coarticulatory process; the other is the gradual stabilization of labiality as a contrastive feature in the vowel system. At the beginning of the OHu period, the labiality of vowels was a redundant feature, or at most a feature with marginal distinctive function. However, in the wake of (i) phonological changes that increased the number of front labial vowels, (ii) the adoption of loanwords including such vowels, and (iii) the enhancement by labialization of the contrast between [aː] and [ɒ] (< [a]), by LOHu labiality was a robustly entrenched contrastive feature. This meant that its role in the identification of lexical morphemes, i.e. the left of word forms, became as crucial as that of [±front] (which had always been a progressively spreading feature). Gradually, stems came to be lexicalized with vowels specified for labiality and, parallel to this, several affixes with mid vowels began to display consistent stem-controlled labial harmony. The latter development can clearly be understood as a step towards increased lexical identifiability (cf. also Suomi 1983 and Baković 2003), and this change in the contrastive force of labiality explains the reversal in the direction of its spreading. The perceptual motivation of progressive harmony – after the shift in direction – as purported here is to be understood as provided by this increase in its role in lexical identifiability.

Appendix

The data for regressive labialization in the two nouns ‘bishop’ and ‘child’ as well as for labial harmony in the allative and the multiplicative suffixes are given in a separate file with a separate readme file that give technical explanation (Data_1, Data_2, Data_3, Data_4, Data_5, Data_Readme, respectively). Here we explain how the data were gathered and what they show.

Abbreviations

abl: ablative

acc: accusative

adj: adjective

adv: adverb

allat: allative

cond: conditional

EOHu: Early Old Hungarian

illat: illatve

iness: inessive

LOHu: Late Old Hungarian

MHG: Middle High German

MiHu: Middle Hungarian

nom: nominative

OHu: Old Hungarian

pl: plural

sg: singular

superess: superessive

Supplementary files

Data_1 lists all occurrences of the word ‘bishop’ in the codices, respectively, as well as its derivatives.

Data_2 lists all occurrences of the word ‘child’ in the codices, respectively, as well as its derivatives.

Data_3 lists all occurrences of word forms affixed with the allative suffix in the codices.

Data_4 lists all occurrences of word forms affixed with the multiplicative suffix in the codices.

Data_5 includes all the remaining data referred to throughout the paper; these are presented here for illustration rather than any kind of quantitative inquiry. Sources are indicated from which the data were taken.

Data_readme gives a more detailed explanation to the data file.

The data referenced above can be found here: https://doi.org/10.16995/glossa.10536.s1

Funding information

The research was supported by the Eötvös Loránd Research Network (HUN-REN) grant KSZF-14/2023 Fehér foltok; part of the research was supported by by the Faculty of Humanities and Social Sciences of Pázmány Péter Catholic University (project number PPKE-BTK-KUT-23-2).

Acknowledgements

We are greateful to Ágnes Kalivoda for her help in data collection.

Competing interests

The authors have no competing interests to declare.