Some notes on phonemes and allophones in synchronic and diachronic descriptions

A review of the traditional procedures in analysing phones leads to the conclusion that in some cases it is the systemic distinctions and not the lexical context which are decisive in establishing phonemes. The phoneme should therefore be defined as the smallest phonological unit which is contrastive at a lexical level and/or distinctive at a systemic level. It is further argued that a new phone can acquire phonemic status when it becomes distinctive in the phonological system of the language irrespective of the context in which it occurs at a lexical level. It is hard to imagine what linguistic description would be like without a phoneme concept of some sort


Phonemes and allophones in synchronic descriptions
It is a well-known fact that since the early 1960s the approach to phonology has become more and more abstract. 1However, the 'classic' approach is still frequently adopted by writers who wish to describe the sound system of a particular language or reconstruct its historical development.The present discussion is intended as a contribution to the classic, more realistic phonological tradition, which is still an inescapable premise even for those who challenge it.
In the traditional phonemic approach to phonology, the phoneme is defined as the smallest contrastive unit and the phone as the smallest perceptible segment in the phonology of a particular language.A phoneme may include several phones, which are known as allophones, contextual variants, or conditioned variants, since their occurrence in the lexical context is restricted by conditioning factors.
The traditional procedure in deciding whether the phones of a particular language should be analysed as phonemes or allophones is distribution.If two phones never occur in the same phonetic environment, they are said to be mutually exclusive and since they complement each other, they are said to be in complementary distribution.If two phones are in complementary 1 It would be pointless here to mention all the theories that have been developed in the wake of discontent over the traditional phonemic approach to phonology.Summary accounts can be found in Clark/Yallop/Fletcher (2007); Dresher (2011) and in other works on phonology, but the most comprehensive treatment of the subject is that of the recently published Oxford History of Phonology, edited by B. Elan Dresher and Harry van der Hulst (2022).
distribution, they are analysed as allophones of the same phoneme.And since they are conditioned by the context in which they occur, they are also said to be conditioned variants of the same phoneme.On the other hand, if two phones do occur in the same phonetic environment, they are said to be in contrastive distribution and are analysed as two independent phonemes. 2 Thus, for example, initial (strongly) aspirated [pʰ], as in pan, and final unaspirated (or weakly aspirated) [p], as in nap, are analysed as allophones of the same phoneme /p/, whereas [m], as in map, and [n], as in nap, are analysed as two distinct phonemes: /m/ and /n/.
Distribution is a very logical and productive criterion, since by experimenting with a whole series of commutations it is possible to establish an inventory which includes all the phonemes (or most of the phonemes) of a particular language.Thus, by experimenting with all the possible vowel commutations in English, we find that, for example, the [e] of pet and the [ae] of pat should be analysed as two distinct phonemes (/e/ and /ae/) and that the slightly nasalized [ẽ] in pen and [ae ͂ ] in pan should be analysed as allophones of /e/ and /ae/, respectively.
Obviously, the relevant commutations should be carried out in all word positions, not only in initial and final position.Thus, for example, we can establish an English phoneme /ʒ/ by comparing words like leather and leisure (/ð/ ↔ /ʒ/) or Caesar and seizure (/z/ ↔ /ʒ/).In these instances, /ʒ/ occurs word medially.Yet if we analyse the two pairs by syllable, we find that in the latter the phoneme /ʒ/ occurs in initial position, whereas in the former it occurs in both initial and final position, since the /ʒ/ of leisure (like the /ð/ of leather) is ambisyllabic, that is, it belongs to both the first and the second syllable. 3

English and German [h] and [ŋ]
A well-known 'exception' to what has been said so far, is offered by the English phones [h] and [ŋ], which never occur in equivalent or nearly equivalent phonetic environments.The former occurs only in prevocalic position (as in hat and behave), the latter is to be found only in postvocalic position, as in bank, finger (with /-ŋg-/), thing.A parallel case is that of German prevocalic [h] (as in Hut 'hat' and behalten 'keep') and postvocalic [ŋ], as in Bank 'bank', Finger (with /-ŋ-/) 'finger', Ding 'thing'.
In either case, no one would be prepared to set up a phoneme /X/ with the two allophones [h] and [ŋ] simply because the two phones are in complementary distribution.The traditional solution to such an 'exception' is to analyse the two phones as two independent phonemes /h/ and /ŋ/, on the grounds that they are not phonetically similar (see, for example, McMahon (2002: 53-55) and Cruttenden (2014: 46f.)).Now, apart from the fact that phonetic similarity or dissimilarity are somewhat vague criteria, it is obvious that in such cases even those who insist on mere oppositions without regard to phonetic features must have recourse to the phonological inventory of the language.When we speak of phonetic similarity, we are not referring to the merely impressionistic perception that [h] and [ŋ], for example, are phonetically dissimilar.What we have in mind is the phonological inventory of a language (English and German in this case), in which we find a systemic distinction between fricative and nasal consonants, and it is on this basis that we assign the glottal phone to a fricative phoneme /h/ and the velar phone to the nasal phoneme /ŋ/.This solution is practically uncontroversial, since [h] and [ŋ] have no feature in common: [h] is voiceless, oral, glottal, and fricative, whereas [ŋ] is voiced, nasal, velar, and frictionless (cf.Kohler (1995: 90f.). 4ince all the English and German consonant commutations produce distinctive sets which include a fricative series and a nasal series, we can establish that the phones [h] and [ŋ] are, so to speak, in contrastive distribution on the systemic level.On the lexical level, we can only find oppositions within the two contrastive series of fricatives and nasals, such as /h/ ↔ /f/ (as in English hat ↔ fat and German halten 'hold' ↔ falten 'fold') and /ŋ/ ↔ /n/, as in English sing ↔ sin and German singen 'sing' ↔ sinnen 'meditate'.
An inescapable consequence of this logical solution is that in these cases the systemic situation becomes prominent.Once we have set up the two phonemes /h/ and /ŋ/, their restricted occurrence (prevocalic position for [h], postvocalic position for [ŋ]) will be relevant only with regard to distribution.
Those who would set up two phonemes argue that [ç] and [x] should be analysed as phonemes because [x] occurs initially in a word like Chassidismus 'Chassidism', or because the two phones occur in opposition in forms like Kuhchen 'little cow' and Kuchen 'cake' or Tauchen 'little rope' and tauchen 'dive'.
Before examining the question at a systemic level, it may be useful to deal with these and other cases that have been brought into the discussion.
The initial [x] in Chassidismus 'Chassidism' and in other learned loanwords is a mere prescriptive phone, which is normally replaced with [ç] or [k] by non-specialist native speakers.But the very fact that [x] can be prescribed as a standard phone could be adduced as an indication that the velar fricative cannot be regarded as a mere allophone.
As for diminutives like Kuhchen 'little cow', Tauchen 'small rope', Frauchen 'little lady' and the like (adduced, for example, by Krech et al. 2009: 30), it should be noted that they are all treated as compounds even when no pause is observable before -chen. 7This appears to be confirmed by the fact that hypothetical forms like *Kühchen, *Täuchen and *Fräuchen (Philipp 1970: 37) are not in standard use, and the same is true of forms like *Kühlein and *Täulein, whereas Fräulein is obviously something different from a 'little lady' or the 'mistress of a dog'.Moreover, the fact that [ç] is not assimilated to [x] in regional (though perhaps obsolete) forms like Bachchen = Bächelchen 'small brook', Bauchchen = Bäuchelchen 'little belly', etc. 8 tends to confirm that the sequence [xç] occurs only in forms that are treated as compounds.
But what about the systemic level?In addressing this point it is necessary to restrict the analysis of [ç] and [x] to a specific variety of Standard German, since the whole of the German-speaking area offers a great variety of pronunciations, as attested by the material published by the AADG (Atlas zur Aussprache des deutschen Gebrauchsstandards). 9 The choice here falls on the specific variety of Standard German recorded in Krech et al.'s pronouncing dictionary (Deutsches Aussprachewörterbuch 2009).
The fact that [ç] and [x] normally occur in complementary distribution would seem to imply that the two phones should be analysed as allophones of a phoneme /ç/.Yet this runs contrary to the observation that a native speaker regards the two phones as two distinct sounds, and that the nouns Ich-Laut '[ç]-sound' and Ach-Laut '[x]-sound' are regular entries in German dictionaries, and not only in pronunciation dictionaries. 10True, the two terms (like Zischlaut and so many others) may well have been coined by a linguist, but the fact remains that non-specialist speakers, who are normally unaware of allophonic variations, regard the two phones as two distinct sounds and recognize at once a different type of pronunciation in which, for example, [x] is used instead [ç].These speakers (as well as other speakers) would never regard the devoiced allophone [ʁ̥ ], as in krass 'crass', and the voiced allophone [ʁ], as in Gras 'grass', as two distinct sounds.
Here one might counterargue that an objective analysis should not take account of a speaker's judgement, but if we turn to the phonological inventory of the type of speech recorded by Krech et al. (2009), we find that the layman's judgement is supported by a systemic distinction between the two phones (see Table 1).
Fricatives Palatal Velar Table 1 Again, since the two phones [ç] and [x] differ by only one feature, one might object that the same is true, say, of aspirated [t h ] and unaspirated [t], as in Tor 'gate' and rot 'red', respectively.However, aspiration can easily be shown to be irrelevant in the German phonological system, whereas the difference between [ç] and [x] is distinctive, since it is paralleled by that of [ʝ] and [ʁ] (cf.Krech et al. 2009: 30).
It is on the basis of this systemic distinction that the palatal fricative must be assigned to the phoneme /ç/ and the velar fricative to the phoneme /x/.Their restricted occurrence will be relevant only with regard to distribution.
All the controversial instances presented above (Kuhchen, Tauchen, Frauchen, etc.) have been included in this treatment because they appear in most discussions of German [ç] and [x].However, they would all become uncontroversial if only we accepted (as indeed we should) that the systemic distinction between palatal and velar implies that the palatal fricative should be assigned to the phoneme /ç/ and the velar fricative to the phoneme /x/.

Russian [i] and [ɨ]
Another distribution which deserves attention when discussing phonemes and allophones is offered by Russian [i] and [ɨ].Here we have two opposing phonological schools: the Moscow school, which advocates the one-phoneme solution /i ~ ɨ/, and the Saint-Petersburg (Leningrad) school, which prefers to set up two distinct phonemes: /i/ and /ɨ/.Reformatskij (1970) gives a detailed account of the controversy, while Chew (2003) presents the main opposing arguments (60-62) and opts for /i ~ ɨ/ as "the more minimal approach" (66) in view of his computational aims.Generally speaking, the majority of scholars is in favour of the one-phoneme solution (see, e. g.Jones 1971: 157;Padgett 2003: 39;and Timberlake 2004: 40) The assumption of a systemic distinction between front and central is thus supported by the lexical level, so that we are compelled to set up a phonological inventory in which [ɨ] is systemically distinct not only from [i], but also from [u], as indeed the pronunciation of the three corresponding letters (и, ы, and у) clearly confirms (see  As in other cases discussed above, the restricted occurrence of /i/ and /ɨ/ will be relevant only with regard to distribution.

Defining the phoneme
The three cases discussed above have been purposely chosen from well-known languages, so that anyone can see that distribution cannot be adopted as the only criterion in deciding whether two phones should be analysed as allophones or phonemes.Further examples might perhaps be adduced from other languages, but care should always be taken to ensure that they are reliable and properly documented, with special attention to the speech habits of the speakers of the language and their idiosyncrasies.
Having established that distribution cannot be adopted as the only criterion in deciding whether two phones should be analysed as allophones or phonemes, it is now possible to suggest a new definition of the phoneme which takes into account the two interdependent levels, namely the lexical and the systemic level.
To define the phoneme as the smallest contrastive unit at a lexical level is not enough, and to define it as a unit of sound that can distinguish one word from another in a particular language is entirely misleading, since the occurrence of minimal pairs is by no means a prerequisite in deciding the status of a phone.A nearly equivalent context, not necessarily an equivalent context, is sufficient to establish an opposition.Thus, for example, a pair like lachen /ˈlaxǝ/ 'to laugh' and vlaggen /ˈflaɣǝ/ 'flags' is decisive evidence in establishing the contrast /x/ ↔ /ɣ/ for those Standard Dutch speakers who distinguish between a fortis velar fricative and a lenis velar fricative.
True, quasi-minimal pairs, symmetries in the phonetic system as well as phonetic similarity have been taken into account in the past (as in, for example, Pike 1947: passim), but the generally accepted definition of the phoneme with its stress on the lexical level has often been used as will be shown also in section 3 belowto deny the status of independent unit to phones that had a right to it.
On the basis of what has been said above, it is suggested that a more accurate definition of the phoneme could be this: the phoneme is the smallest phonological unit which is contrastive at a lexical level and/or distinctive at a systemic level.When analysing a particular case the two levels should always be considered together.Thus, for example, English and German [m] and [n] are both contrastive at a lexical level and distinctive at a systemic level, whereas English and German [h] and [ŋ] are distinctive at a systemic level, though not contrastive at a lexical level.On the other hand, German [ç] and [x] are distinctive at a systemic level and only marginally contrastive at a lexical level.Russian [i] and [ɨ] provide a different case, in that they are only marginally contrastive at a lexical level and consequently interpreted as distinctive at a systemic level.
All this shows that absolute generalizations are impossible and that in deciding whether or not a phone should be interpreted as a phoneme one must necessarily consider the characteristics of the particular language under examination.Thus, for example, it would be absurd to analyse Russian [ɨ] as a mere allophone, since it is deliberately pronounced as the name of the letter ⟨ы⟩ in the same way that [i] is used for the letter ⟨и⟩, and [u] for the letter ⟨у⟩.Nor would it be reasonable to claim that German [x] is not a phoneme distinct from [ç], since native speakers are well aware of the difference between the two sounds, for which they have the names Ich-Laut '[ç]-sound' and Ach-Laut '[x]-sound'.
With regard to the conditioning factor of an allophone, one might get the impression that it must necessarily reside in the syllable, since in discussing the phoneme it is customary for phonologists to adduce monosyllabic examples or to refer to particular syllables within polysyllabic words.14This is the usual practice, at least when phonologists are analysing a particular language on the synchronic level.Things are different when linguists employ the concepts of phoneme and allophone in diachronic descriptions.

Phonemes and allophones in diachronic descriptions
When we turn to the phonological history of a language, we find that in some cases the conditioning factor of an allophone is supposed to have resided in a different syllable from the one in which the allophone is supposed to have developed.This can be illustrated with an example from the well-known case of i-umlaut (i-mutation) in English and German.The Proto-Germanic plural form */mu:siz/ appears as Old English mȳs 'mice' and as Old High German mūsi, miusi (with /y:/) > Mäuse 'mice'. 15Under the influence of an [i]-sound (/i/, /i:/, /j/) in the following syllable, the phoneme /u:/ is supposed to have developed a somewhat advanced back allophone [u̟ :], which was eventually fronted to [y:].This is the generally accepted reconstruction of a distance assimilation change.However, difficulties arise when the concepts of phoneme and allophone are employed to describe the phonologization of a new allophone.The traditional explanation with regard to i-umlaut is that the new allophone came to be analysed as an independent phoneme when the conditioning factor in the following syllable changed or disappeared (see, for example, Twaddell 1938: Penzl 1949;Moulton 1961;and Fox 2015: 55). 16A serious objection to this view is that the change or loss of the conditioning factors would result in the loss of the relevant allophones.For if a phone is actually conditioned, the change or loss of the conditioning factors results in its reversal to the main phonetic features of the phoneme to which it belongs (cf.Erdmann 1972: 22).
The obvious corollary is that if an alleged allophone does not disappear, it is because the relevant phone has already attained phonemic status.Moreover, it is important to note that the factors which trigger off a change can be adduced to explain a diachronic phenomenon, not necessarily as conditioning factors in the synchronic analysis of the new situation created by the change itself.We should therefore distinguish between two aspects of phonological change: English mȳs.The diphthong in today's Mäuse 'mice' from Old High German mǖsi reflects a complex development of the original umlauted vowel (/y:/ > /ʏy/ > /øy/ > /oeʏ/ > /ɔɪ/).The ä in Mäuse 'mice' is due to the analogy of the a in Maus 'mouse', the historical spelling of the plural being Meuse (see DWDS s. v. Maus). 16For different theoretical approaches to the notion of phoneme and phonological change see Historical Phonology (2015), which provides numerous cross-references throughout the volume.The Old High German i-umlaut is frequently cited in discussions on phonologization.See, e. g., Kiparsky (2015: 563 et passim), who uses the "Stratal Optimality Theory" framework.For a psychophysical phonetic interpretation of the Old High German umlaut see Schulze (2010).
the diachronic phenomenon, which is triggered off by specific factors, and the resulting synchronic situation, which is determined by new systemic distinctions and new oppositions at a lexical level.
With regard to the i-umlaut of the Old High German back vowels we cannot, of course, reconstruct all the intermediate stages of the change, but we may assume that at an early stage the new phones would be somewhat advanced back vowels, a series of allophones that may be rendered with [u̟ :], [u̟ ], [o̟ :] and [o̟ ].However, at a later stage the new phones would be front rounded vowels ([y:], [y], [ø:], [ø]), which would be clearly distinct from a whole series of unrounded front vowels (/i:/, /i/, /e:/, /e/) in the phonological system of the language.It was at this point that the new phones became phonemes, since in the front area there now was a systemic distinction between rounded and unrounded vowels.All this may be summarized as in Table 3.This is a specific case, but the same principle applies, not only to the other Old High German umlauted vowels and diphthongs, but also to similar cases in which a new allophone becomes distinctive in the phonological system of a language.

Early stage
We thus come to the general conclusion that a new phone can acquire phonemic status when it becomes distinctive in the phonological system of the language irrespective of the context in which it occurs at a lexical level.With regard to i-umlaut we may observe that 1) when the gradual change reached its final stage, the original triggers ceased to be conditioning factors; 2) the phonemicization of i-umlaut occurred when the triggering factors were still present, not when they changed or disappeared.
All this presupposes that i-umlaut was a gradual phonetic process which eventually resulted in the phonemicization of an allophone.However, since i-umlaut is a kind of 'vowel harmony', it could also be argued that the change in question operated without intermediate stages, and that the umlauted vowels became phonemes as soon as the influence of an [i]-sound in the following syllable affected the original vowels.Now, vowel-harmony may exploit a pre-existing vowel or create a new one.Whether the new vowel should be analysed as an allophone or as a phoneme depends on distribution at a lexical level and/or distinctiveness at a systemic level.In the case presented above, the creation of the new rounded front items by vowel harmony would have resulted in the rise of the new series /y:/, /y/, /ø:/, /ø/, which would be systemically distinct from the pre-existing series /i:/, /i/, /e:/, /e/.

Conclusions
A review of the traditional procedures in analysing phones has led to the conclusion that in some cases (examples from English, German, and Russian) it is the systemic distinctions and not the lexical context which are decisive in establishing phonemes (Sections 1-1.3).The phoneme should therefore be defined as the smallest phonological unit which is contrastive at a lexical level and/or distinctive at a systemic level (Section 2).Moreover, a new phone can acquire phonemic status when it becomes distinctive in the phonological system of the language irrespective of the context in which it occurs at a lexical level (Section 3).

Table 3
Taking as example the change /o:/ > /ø:/ in the antecedents of Present Standard German schön 'beautiful', Schöne 'beauty' and [ver]schönen 'beautify', we may present the consequences of i-umlaut on distribution as in Table4.