2.1 The synchronic perspective I: allomorphy vs. no allomorphy

Modern English evinces allomorphic variation in very few inflectional domains. Basically, there is the third person singular present tense and the past tense morpheme on verbs, and the plural and the possessive morpheme on nouns. At the phonological level, these four cases reduce to two, with the third person singular, the plural and the possessive morphemes following largely the same rules (but see next subsection). The repair principles, which are highly similar across the four domains, are mainly constituted by voicing assimilation and vowel epenthesis.Footnote ^4, Footnote ⁵

The only allomorphy at the word level is found with both the definite and the indefinite article. Notably, the two types of article display quite disparate repair strategies. While the indefinite article has recourse to consonantal epenthesis (see (1)), the definite article does not discard hiatus. Although it does eliminate the schwa, the repair is restricted to replacing the schwa with the full vowel /iː/. This retains the adjacency not only of vowels but also of identical vowels, as can be observed in the eel.

Owing to the paucity of inflections in English, there is little to compare the morphological allomorphy to. In fact, the alveolar obstruents provide the phonological basis for all of the inflections. The only other regular morpheme {-ing} has an invariant shape. It attaches without any modification even to verbs ending in 〈-ing〉, as in singing and ringing. This permissiveness is not particularly surprising given the phonotactic orthodoxy of the resultant string.Footnote ⁶ It is also no problem to add {-ing} to a vowel-final verb (e.g. see-ing and glu-ing) even though this causes hiatus.

A larger basis for contrasting allomorphy to no allomorphy can be found at the word level. As noted, among the determiners allomorphy is restricted to the articles. Neither possessive nor demonstrative determiners exhibit allomorphic variation.Footnote ⁷ The same holds true of negative and indefinite determiners. Thus, there must be something that sets articles off from all other determiners.

2.2 The synchronic perspective II: plural vs. possessive

This subsection draws attention to a minor but significant difference in the allomorphy of the plural and the possessive morphemes. The general rule is that the allomorphy of both is governed by sibilant similarity. This constraint requires an epenthetic vowel after sibilants for plural and possessive alike (e.g. the bushes – Bush's administration). This rule is largely exceptionless in plural formation even though Latin and Greek plurals mar the picture slightly. Some Latin words in -us and -um form their plural with -i and -a, respectively (e.g. alumnus→alumni). The only case to remotely resemble coalescence is the pluralisation of Greek nouns ending in -is (e.g. crisis→crises), where, under one analysis, vowel lengthening and final voicing induce word-internal changes. If the suffix rule had applied, the plural form would be composed of two highly similar phoneme sequences abutting on each other (/is–iz/). This similarity was apparently felt to be intolerable and so word-internal plural marking avoids such sequencing. It is important to note that this is not a true instance of coalescence because the singular and the plural forms are phonologically distinct. To conclude, the relevant cases of English pluralisation almost invariably rely on epenthesis. No genuine example of coalescence could be found.

The case of the genitive marker is somewhat different. Many commonalities notwithstanding, the possessive morpheme makes greater use of coalescence than does the plural. In their grammar of the English language, Quirk, Greenbaum, Leech & Svartvik (Reference Quirk, Greenbaum, Leech and Svartvik1985: 320) mention five areas where the genitive may be realized as phonetically null: (A) Greek names in /z/ of at least two syllables in length, (B) some biblical names ending in /s/, and (C) fixed expressions of the type for X's sake. Quirk et al. consider coalescence to be the rule in (A) and (C). For (B), they introduce the hedge normally. Case (D) involves English surnames ending in /z/ such as Burns. Here, the authors note vacillation between epenthesis (/bɜːnziz/) and coalescence (/bɜːnz/), with the former being more common than the latter (see also Zwicky Reference Zwicky, Cohen and Wirth1975: 133). Finally, coalescence is exceptionless when the genitive morpheme is stacked onto a regular plural noun (case E). These areas suffice to show that coalescence sits more comfortably with the possessive than the plural marker in English.

2.3 The diachronic perspective: the rise and fall of allomorphy

Old English had no phonologically conditioned grammatical-marker allomorphy (Berg Reference Berg2009). It thus offers an excellent point of departure for an examination of the rise of allomorphy. As Middle English exhibited morphophonological variation where there is none in Modern English, we also have occasion to trace the downfall of morphophonology. This subsection may therefore serve to outline the life cycle of a morphophonological system.

Let us begin with the determiner system, in particular the possessive determiners. Old English pronouns formed inflectional paradigms like adjectives. The possessive determiners were the genitive forms of the personal pronouns. Of special interest are the forms of the first and second person singular, viz. min and þin, respectively. Crucially, these forms were invariant in Old English. Their significance derives from their coda consonant /n/. The alveolar nasal was one of the major inflectional markers of Old English. It functioned to code plural on weak nouns, present tense plural and past tense subjunctive on verbs (in the Midlands also plural indicative in Middle English (ME)). Furthermore, it constituted the final part of the regular infinitive marker -an, of the past tense plural indicative ending -on and of the past participle of strong verbs -en. Last but not least, it was the final part of the non-nominative suffix -an of the weak declension of adjectives.

As is well-known, inflections underwent a process of massive erosion from Old to Middle English, during which the alveolar nasal came under considerable pressure. However, it did not disappear all of a sudden. As Schlüter (Reference Schlüter, Rohdenburg and Mondorf2003) demonstrated in a lucid analysis, the probability of the loss of /n/ was a function of the nature of the initial phoneme of the following word. A following consonant facilitated, while a following vowel inhibited, its disappearance. The reason for this differential behaviour is an output-driven one. In both cases, the retention and the omission of the critical element guarantees the simple alternation of consonant and vowel, the ideal syllable structure tout court (Schlüter Reference Schlüter, Rohdenburg and Mondorf2003).

The sensitivity to the immediate context generated a relatively systematic pattern in which the old (i.e. consonant-final) forms were required before vowel-initial (as well as 〈h〉-initial) words (see (5)) while the innovative (i.e. vowel-final) forms only occurred before consonant-initial words (see (6)). It is only natural that the old forms were also found before consonant-initial words as they were not abandoned instantaneously (see (7)).

1. (5) ME myn age ‘my age’

2. (6) ME þi wif ‘your wife’

3. (7) ME min drihten ‘my lord’

It is of critical importance that the gradual loss of /n/ gave birth to morphophonology. Apparently, the sensitivity to the subsequent phonological context emerged quite swiftly. Looking back as far as the Early Middle English period, the system had developed simple allomorphic rules which were largely categorical (see Schlüter Reference Schlüter and Minkova2009). That is, the gradual disappearance of /n/ created a morphophonological paradigm consisting of the old and the new forms, whose selection became phonemically motivated.

The perspective from Modern English reveals that this morphophonological variation did not stand the test of time. The simple Middle English rule system had already been disturbed in the 16th century (Crisma Reference Crisma and Minkova2009). As examples (8) and (9) from Shakespeare's The Comedy of Errors (Act 5, Scene 1) illustrate, my and mine could occur in the same phonological context.

1. (8) (a) my order

   (b) mine oath

2. (9) (a) my honest friend

   (b) mine honesty

Shakespeare's works show the gradual loss of the consonant-final forms to be well underway. The Early Modern English period thus marks the transition from an allomorphic to an allomorphy-less usage via probabilistic variation (Schendl Reference Schendl, Hickey and Puppel1997). By the 18th century, the alveolar nasal on the possessive determiners had disappeared (Crisma Reference Crisma and Minkova2009). Nowadays, we are left with diphthongized my and thy, with the latter being thoroughly archaic (hallowed be thy name) after its replacement by your in Early Modern English.Footnote ⁸ It may be inferred from this development that the tendency to eliminate inflections as well as pseudo-inflections was stronger than the preservation of the ideal syllable structure, which itself was ensured by allomorphy but has been given up in the modern language (witness the hiatus in my eyes).

It is worth adding that the possessive determiners were not alone in undergoing temporary vacillation between consonant-final and vowel-final alternants. This Middle English process also affected verbs such as see vs. seen ‘to see’ and go vs. gone ‘to go’ as well as the negation particle no vs. noon ‘no’ (see Schlüter Reference Schlüter, Rohdenburg and Mondorf2003). In the verbal infinitive as well as the adverbial negation marker, the word-final consonant did not survive into Modern English.

A most intriguing contrast exists between the possessive determiners and the indefinite article. The latter derives from the Old English numeral an ‘one’ which was inflected much like adjectives (and could even occur in the plural). In the accusative form, it was suffixed by 〈ne〉. By Early Middle English, all inflectional variants had been lost and the consonant-final form on/an existed alongside the vowel-final form o/a.Footnote ⁹ At Chaucer's time, the following allomorphic rules were fairly consistently applied: the consonant-final variant was selected before vowels and 〈h〉 (see (10) and (11), respectively) while the vowel-final variant was selected before consonants (see (12)). The examples are all taken from Chaucer's The Wife of Bath.

1. (10) (a) an ydiot ‘an idiot’

   (b) an owle ‘an owl’

2. (11) (a) an hooly man ‘a holy man’

   (b) an hen ‘a hen’

3. (12) (a) a forest ‘a forest’

   (b) a woman ‘a woman’

Evidently, with the exception of 〈h〉-initial words, this is the distribution that has been preserved to this day in Standard English.

The preceding discussion firmly invites a comparison between the indefinite article and the possessive determiners. The parallelism between the two word classes in terms of their allomorphy in Middle English is all too obvious. Both used the same type of allomorphic rules.Footnote ¹⁰ However, it is a highly significant fact that the allomorphic variation had a differential life expectancy in the two word classes. While both morphophonological domains came into existence at roughly the same time, the allomorphy of the possessive determiners had come to an end in the 17th century whereas the allomorphy of the indefinite article survives to this day. Of course, we cannot tell whether the demise of the possessive determiner allomorphy only antedates that of the article or whether the allomorphy of the article is set in stone. What is certainly in need of an explanation is the greater historical stability of the latter than the former.

The final diachronic analysis addresses verb and noun inflection. These sets may be considered together because the diachronic processes at work are largely the same. As noted before, the Old English inflectional suffixes had disappeared by the end of the Middle English period. In particular, the different vowel qualities of Old English inflections merged in /e/, which was subsequently reduced to schwa and ultimately lost.Footnote ¹¹ The regular plural marker on all nouns in Middle English was syllabic (with a good deal of dialectal variation with regard to the rapidity with which schwa disappeared). Thus, we typically encounter disyllabic stones and names. Crucially, the change from the syllabic to the non-syllabic suffix was not carried through across the board. It happened in the case of nouns ending in a non-sibilant but failed to take place in nouns ending in a sibilant. This marks the birth of allomorphic variation. Note that two allomorphic decisions are at stake here. For one thing, there is the choice between the syllabic and the non-syllabic allomorph. For another, the elimination of schwa created adjacent consonant clusters whose members could have the same or a different voicing value. As adjoining obstruents of divergent voice specification were illicit, a further differentiation was necessary. The Middle English plural marker /(ə)s/Footnote ¹² could remain unscathed after nouns ending in voiceless consonants but had to assimilate after nouns ending in voiced consonants. It is likely that schwa omission and voicing assimilation were concurrent historical processes.Footnote ¹³ In this way, a bipartite variation between syllabic and non-syllabic allomorphs as well as voiced and voiceless allomorphs came into being. Needless to say, this allomorphic pattern is still current today.

The emergence of the third person singular present tense allomorphy respected the same principles as those laid out for noun inflection. As far as I can see, there is no evidence to suggest that morphophonological variation was implemented earlier in the verbal than in the nominal domain (or vice versa). Since the phonological basis was the same in the two domains, it is probable that their allomorphy emerged simultaneously in (12th-century) Middle English. By and large, the rise of the allomorphy of the genitive marker followed the same path as in the above cases. However, there is one critical difference, which relates to what was said in the previous subsection. The genitive marker was produced less consistently than the plural marker. When the nominal stem ended in /s/, the genitive was often left unexpressed in Middle and Early Modern English (Curme Reference Curme1935: 134), as in hors fet ‘horse's feet’. What we observe here is a certain divergence in the historical development of the two morphemes. While the plural marker uses epenthesis,Footnote ¹⁴ the genitive marker is more likely to be subject to coalescence.

In conclusion, the diachronic perspective has shown English to be a paradigm case for studying the rise and fall of morphophonology. The variation arises as a result of the gradualness of language change, which leads to the co-existence of the old and the new forms. Crucially, the concurrent forms do not stand in free variation because language change occurs faster in contexts which are favourable to the innovation but more slowly in contexts where the innovation creates new problems (see e.g. Rohdenburg Reference Rohdenburg, Rohdenburg and Mondorf2003). This is precisely the ecological setting required for the emergence of morphophonology. The contextual factors determine whether the old or the new form is to be preferred and so give rise to phonologically motivated allomorphic variation. The demise of morphophonology is essentially the mirror-image of its birth. When the new form has completely supplanted the old one, the basis for variation has evidently been lost.

On a more fine-grained level, the gradualness of language change entails a sequence of three transitional stages which may be represented thus:

1. (13) no allomorphy before the change→

   phonemically and otherwise conditioned allomorphy→

   phonemically conditioned allomorphy→

   phonemically and otherwise conditioned allomorphy→

   no allomorphy after the change

Allomorphy may be understood as a transitory phenomenon which arises between the beginning and endpoint of sound change. The sequence in (13) represents an oscillation between more and less categorical stages. While the old and the new usage as well as the middle stage are close to categorical,Footnote ¹⁵ the phases between them are more probabilistic in nature. This is because the intermediate stages are open to a larger number of influences, none of which overrides the others. These contextual factors are omnipresent but too weak to leave their mark when usage is strict. However, once this usage loosens its deterministic grip, these factors may make themselves felt and influence the linguistic patterns. During the earlier probabilistic stage, there is a bias in favour of the old forms whereas the opposite is true of the later probabilistic stage. In this view, phonemically conditioned allomorphy represents the middle stage in phonological change. Note finally that no claim is made here to the effect that allomorphy is of necessity a transitory phenomenon. A linguistic change may stop mid-way and settle in a quite stable state. Completion of a change is not an end in itself. The indefinite article allomorphy has persisted for almost a millennium and there is little to foreshadow its imminent disappearance in the standard language (but see footnote 15).

5.1 The case of French: string frequency

It was argued in the last but one section that frequency effects arise at various levels of linguistic analysis. Therefore, it is only to be expected that allomorphy is sensitive to frequency effects located at levels above the single word or morpheme. Such is the case in French, as evidenced by the data in (3) and (4), repeated here for convenience as (15) and (16).

1. (15) F. Je me demande s'il est parti. (s'il←si il)

   ‘I wonder whether he is gone.’

2. (16) F. Je me demande si Isabel est partie. (*s'Isabel)

   ‘I wonder whether Isabel is gone.’

Why do we observe contraction in s'il ‘if he’ but not in si Isabel ‘if Isabel’ even though the phonological points of contact are identical? The reason, I argue, lies in what is known as string frequency (Krug Reference Krug1998, Reference Krug, Rohdenburg and Mondorf2003). This measure assesses frequency effects at the level of word sequences, which need not even form a syntactic unit (compare Bybee & Scheibman Reference Bybee and Scheibman1999). Clearly, the string frequency of s'il is much higher than that of si Isabel. The high frequency of s'il affords it increased lexical strength which eases the way for phonological interaction between the two words, hence the coalescence of the two /i/s. As the string si Isabel is much less frequent, it cannot undergo lexicalization and hence its constituent parts remain phonologically autonomous. In this way, high string frequency may be argued to bring about allomorphy.Footnote ²²

5.2 The case of Italian: optional allomorphy

Unlike the languages discussed so far, Italian displays optional allomorphy. An epenthetic voiced alveolar stop may be added to certain function words such as e ‘and’ and a ‘to’ when the following word begins with a vowel (Maiden & Robustelli Reference Maiden and Robustelli2000). The optional nature of this process reveals one facet of allomorphy which remains invisible in compulsory allomorphies: optional processes bring into the picture gradualness and with it, the possibility of gradual differences. It may accordingly be predicted that allomorphy is found more often in more frequent items. Focusing on the two coordinating conjunctions e ‘and’ and o ‘or’, we may predict a higher degree of /d/-epenthesis in the more frequent than in the less frequent word.

Pertinent data, taken from Juilland & Traversa's (Reference Juilland and Traversa1973) frequency dictionary, are reproduced in Table 4, which shows that the additive conjunction is almost ten times more frequent than the disjunctive one. This result allows us to predict a higher rate of /d/-epenthesis in the case of e than o.

Table 4

Token frequency of selected conjunctions in Italian

Although precise quantitative information on /d/-epenthesis is apparently unavailable, Table 5 may serve as a rough approximation to a complex phenomenon which is subject to individual speakers' preferences and prescriptivist effects among other factors.

Table 5

Extent of /d/-epenthesis in Italian

An intriguing difference emerges from a comparison of the two conjunctions. Epenthesis starts ‘earlier’ in e than in o. That is, more powerful facilitators for /d/-insertion are required for o than for e. Specifically, only if o is followed by an identical vowel does epenthesis occur whereas any following vowel is good enough to induce epenthesis in the case of e. The two conjunctions are sensitive to the same hierarchy in exhibiting a higher degree of epenthesis when they are followed by identical vowels than when they are followed by non-identical vowels. The greater reluctance of the disjunctive conjunction to undergo /d/-insertion can also be seen in the higher ratio of non-epenthesized to epenthesized forms in o than e in Table 4. This result lends neat support to the research hypothesis.

5.3 The case of Greek: the scope of allomorphy

Allomorphy may vary not only in the degree to which it is implemented but also in its scope. Allomorphic variation may occur in more or less restricted circumstances, depending on the generality of the allomorphic rules. The frequency account predicts more frequent structures to develop more extensive allomorphies than less frequent structures do. This prediction will now be tested on definite and indefinite articles in Greek.

Greek articles are declined for number, case and gender. The gender system is three-valued (masculine, feminine, and neuter), and so is the case system (nominative, genitive, and accusative). Allomorphic variation occurs only in the masculine and feminine accusative singular forms of the definite and the indefinite article. Allomorphy in both types of article operates on the presence or absence of a final alveolar nasal. According to Holton, Mackridge & Philippaki-Warburton (Reference Holton, Mackridge and Philippaki-Warburton1997), the relevant definite articles take /n/ before following words beginning with any vowel or stop (which may either be singleton or the first part of a consonant cluster). By contrast, the relevant indefinite articles take /n/ only when they are followed by a vowel or a voiceless stop. The allomorphy of the indefinite article is thus a subset of that of the definite article. There is an additional difference between the two types of article. As Holton et al. point out, the insertion of the alveolar nasal is mandatory in the definite, but optional in the indefinite, article, which is reminiscent of the behaviour of the Italian conjunctions as discussed in the preceding subsection.

The obvious prediction that can be derived from the disparate behaviour of the two types of article is that, in Greek, the definite one should be used more commonly than the indefinite one. This prediction was examined on the basis of a frequency analysis of the spoken language. A video recording of a TV show lasting approximately one hour was scrutinized for all occurrences of articles.Footnote ²³ The database numbered 7590 words. The pertinent results are presented in Table 6. The few cases where the definite article occurred as a concomitant of the demonstrative determiner were left out of consideration.

Table 6

Token frequency of definite and indefinite articles in spoken Greek

Table 6 shows the definite article to be nine times more frequent than the indefinite one, thus providing strong confirmation of the research hypothesis. Not surprisingly, the frequency difference holds for invariant articles and those with allomorphy alike. It may thus be argued that frequency determines the scope of allomorphic rules. The higher the frequency, the wider the scope of allomorphy. A further remarkable insight brought forth by the Greek data is the relationship between the generality and the optionality of allomorphy. While these two aspects might seem unrelated at first sight, the present account suggests that the generality and the optionality of allomorphy are linked by frequency. High frequency is the single cause that induces both the wider scope and the (largely) compulsory nature of allomorphy.

5.4 The case of German and English: coalescence vs. epenthesis

The next analysis adopts a contrastive perspective and compares genitive marker allomorphy in English and German. A contrastive approach represents a particular challenge because, as noted above, different languages may have different allomorphy thresholds and thus do not compare easily. It is therefore necessary to ‘calibrate’ the two languages by showing that they face similar problems and tackle these in similar ways.

An appropriate area to start from is verbal inflections. The problem that arises during the addition of an inflectional suffix is highly similar in the two languages even though English inflects its verbs far less than German. The third person singular present tense inflection is 〈s〉 in English and 〈t〉 in German (G.; see (17)). The crucial question is how English deals with verbs ending in /s/ or /z/ and how German deals with verbs ending in /t/ or /d/. Actually, the two languages have evolved identical solutions to this problem: they insert an epenthetic short vowel, as shown in (18).

1. (17) (a) E. she laugh-s

   (b) G. sie lach-t ‘she laughs’

2. (18) (a) E. he kiss-e-s

   (b) G. er wart-e-t ‘he waits’

German also makes use of epenthesis in the second person singular of /t/- or /d/-final non-umlauting verbs, even though the relevant inflectional suffix is /st/, i.e. suffixing does not create two contiguous /t/s; see (19).

1. (19) G. (a) du lach-st ‘you laugh’

   (b) du reit-e-st ‘you ride’

The second person epenthesis may be understood as an analogical extension of the third person repair strategy. By contrast, umlauting verbs do not undergo epenthesis and so may have /tst/ in the second person singular, which is broken up in (19b). Consider the conjugated first and second person forms of the infinitive raten ‘to guess’ in (20).

1. (20) G. (a) du rät-st /reːtst/ ‘you guess’

   (b) er rät /reːt/ ‘he guesses’

The phonological shape of the second person singular suffix raises the question of how German copes with /s/-final verbs. Intriguingly, although this situation leads to two /s/s in a row, the epenthetic process again fails to occur. Instead, coalescence takes over, as can be observed in (21).

1. (21) G. (a) du reißt ‘you tear’ (*du reiß-e-st) (from infinitive reiß-en ‘to tear’)

   [raist] [raisεst] [raisən]

   (b) du reist ‘you travel’ (*du reis-e-st) (from infinitive reis-en ‘to travel’)

   [raist] [raizεst] [raizən]

The only difference between (21a) and (21b) is that the stem-final obstruent is voiceless in reißen ‘to tear’ but voiced in reisen ‘to travel’. However, this difference is levelled out in the inflected forms which are homophonous in both the second and third person singular. This is due to a pervasive devoicing process in German.

Cases of coalescence in English were reviewed in Section 2.2 in connection with the Saxon Genitive.

As an interim conclusion, English and German make use of both epenthesis and coalescence in verb and noun inflection. This basic similarity warrants a comparison of the two languages in terms of their repair strategies.

A notable difference is found in the inflectional possessive marker in English and German. Its similar phonological shape renders the comparison especially direct. Compare the semantically equivalent NPs in (22) and (23).

1. (22) E. Robert's ticket

   [rɒbəts tikit]

2. (23) G. Roberts Ticket

   [roːbεrts tikεt]

It seems at first sight that the only difference between English and German lies in the invariably voiceless coda [s] in German (due to final devoicing) and the allomorphic variation between [s] and [z] in English. However, an interesting contrast emerges when it comes to examining the reaction of the two languages to phoneme repetition. The critical case is represented by names terminating in /s/, as exemplified in (24) and (25).

1. (24) E. Thomas's car

   [tɒməsiz kɑː]

2. (25) G. Thomas' Auto

   [toːmas autoː]

As can be seen from examples (24) and (25), English employs epenthesis whereas German resorts to coalescence. Here, the two languages seem to follow different principles.

To investigate this issue more systematically, I ran two small experiments, one in English and one in German, with identical procedures. The task of the participants was to generate a possessor+genitive marker+possession NP on the basis of a first name that the experimenter provided. For example, participants had to expand Don into Don's wife. The experiment was announced as a free association task which was intended to elucidate the type of associations that subjects had with certain names. This was to divert subjects' attention from the genitive marker.

The German stimuli focused on the three final phonemes /s/, /ts/ and /ʃ/, the English stimuli on the six final phonemes /s/, /z/, /ʃ/, /tʃ/, /dʒ/ and /θ/. First names ending in other phonemes served as fillers and were quasi-randomly selected. A distinction was introduced between monosyllabic and disyllabic names because this variable is known to affect the choice of repair strategy. Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 320) note that Greek names ending in /z/ go for epenthesis when monosyllabic but coalescence when disyllabic (see above). I planned to test each word length and each critical phoneme on the basis of two names. This would have yielded 12 critical items in German (3×2×2) and 24 (6×2×2) in English. However, the three final phonemes /ʃ/, /tʃ/ and /dʒ/ do not combine easily with English disyllabic names. These phonemes could consequently only be tested with monosyllabic stimuli. In German, /ʃ/ is hardly ever found at the end of first names. Thus, only a singleton example was examined in the monosyllabic set. The list of stimuli included 25 items in German (11 critical and 14 filler items) and 30 in English (18 critical and 12 filler items). Twenty native speakers of Standard German and ten native speakers of English participated in the experiment.

Subject responses were near-categorical in the two languages. Consistency levels were similarly high in English and German at approximately 95%. Word length was without effect in both languages. The repair pattern in evidence in German was straightforward. Coalescence occurred after /s/ and /ts/, as in (25) above. All other phonemes (including /ʃ/) were followed by the non-syllabic genitive allomorph. Instances of epenthesis were infrequent (N=5) and randomly scattered across subjects and items.

As expected, the English subjects consistently generated epenthesis after all sibilants. All words ending in non-sibilants took the non-syllabic allomorph, although the phonemes /θ/ and /f/ were occasionally followed by epenthetic vowels. Not a single case of coalescence occurred in the data.

The conclusion that these experimental results invite is very clear. If the need for repair arises, epenthesis is the rule in English genitive marking and coalescence is the rule in German genitive marking. However, one qualification is called for in English. As noted in Section 2.2, coalescence is not completely excluded from genitive constructions. A few biblical names and the construction for X's sake prefer coalescence but are also compatible with epenthesis.Footnote ²⁴

Why does German gravitate towards coalescence whereas English tends towards epenthesis? It was argued above that epenthesis is the less disruptive of the two repair processes and should therefore be favoured in more frequent structures. It may consequently be predicted that the Saxon Genitive should be more common in English than in German.

Comparing the use of the Saxon Genitive in English and German, one cannot but notice a major difference. Whereas the possessor has lexical status in German, it assumes phrasal status in English (Hartmann & Zimmermann Reference Hartmann, Zimmermann and Maienborn2003). The lexical nature of the possessor restricts the possessor position essentially to proper nouns, as these do not normally take an article. The Saxon Genitive thus has a fairly limited use in German. In English, by contrast, the Saxon Genitive has a far wider range. In addition to proper nouns, the possessor position may be occupied by common nouns, both animate and inanimate, and preceded by articles and several premodifiers, as exemplified in (26).

1. (26) the former Deputy Prime Minister's words

   (from Rosenbach Reference Rosenbach2005: 624)

The point of interest here is that the use of the Saxon Genitive is much less constrained and hence its frequency of occurrence is much higher in English than in German (see Hansen-Schirra, Hansen, Wolfer & Konieczny Reference Hansen-Schirra, Hansen, Wolfer and Konieczny2009 for empirical support). In the light of this contrast, it makes sense to develop different repair strategies. The recourse to epenthesis in English and to coalescence in German can now be understood as a frequency effect which is embedded in a discourse situation in which speakers and listeners have to balance out articulatory effort and the risk of disruptive communication.

This account leaves some latitude for within-language variation. Since frequency is by hypothesis the major determinant, coalescence may be expected in infrequent structures even in a language like English, which favours epenthesis. Thus, part of the explanation for the cases of coalescence noted in Section 2.2 may be the infrequency of the foreign names and the construction for X's sake. There are certainly additional factors at work. For instance, the fact that epenthesis is found in monosyllabic Greek words while longer ones prefer coalescence recalls the notion of the ideal length of an inflected word. To be specific, resorting to epenthesis in monosyllabic names but to coalescence in disyllabic ones gives a syllable number of two in both cases.

5.5 The case of Spanish: the inheritance effect

In the final analysis we return to the Spanish conjunctions as introduced at the beginning of this study. The prediction from the frequency account is very clear. The conjunctions allowing allomorphy should occur more often than the invariable conjunctions. The counts in Table 7 are taken from Juilland & Chang-Rodriguez's (Reference Juilland and Chang-Rodriguez1964) frequency dictionary.

Table 7

Token frequency of selected conjunctions in Spanish

It is obvious from Table 7 that, viewed as groups, the coordinating conjunctions are much more common than the subordinating ones. Frequency thus seems to distinguish between the invariant and the phonologically variable conjunctions. However, there is a fly in the numerical ointment. The frequency difference between the disjunctive and the conditional conjunction is only minimal. It is necessary therefore to take additional factors into account. One such factor, which has already been invoked in the interpretation of the historical data in Section 4, capitalizes on the presence of y ‘and’ in the same paradigm and may be understood as an inheritance effect. It is conceivable that o ‘or’ inherits its allomorphy from its closest neighbour which is not only of extremely high frequency but also subject to allomorphic variation.Footnote ²⁵ The effect of the frequency of individual items would thus have to be gauged in conjunction with the frequency of the other members of the paradigm.

5.6 Conclusion

All five tests reported in the foregoing discussion have by and large borne out the predictions of the frequency account. Despite its general success,Footnote ²⁶ the frequency account needs to be flanked by additional factors in a comprehensive theory of allomorphy. Clearly, within-paradigm inheritance (see previous subsection) is only one of them. Another such factor is required to account for the allomorphy or otherwise in German verb inflection. As discussed in Section 5.4, umlauting verbs defy allomorphic variation whereas non-umlauting verbs evince it (contrast (19) and (20)). The reason for this difference lies in the tendency to code a given piece of information only once, that is, to avoid redundancy. In the above, epenthesis was argued to increase the salience of the inflectional marker (at the expense of extra length) and thus to be advantageous to the listener. The umlaut in umlauting verbs performs a highly similar function in that it signals the second and third person singular present tense. No umlaut occurs in the other persons. Notably, it is precisely in the second and third person where the troublesome phoneme sequence arises. We thus register a competitive relationship between umlaut and epenthesis. As both serve the same function, the language system chooses one or the other (though not both).