1 Introduction

The present article addresses the issue of crosslinguistic rarity by focusing on a rarissimum of person marking. Assuming that the explanation of crosslinguistic rarities requires a diachronic dimension (Bybee 2008), any explanatory account of the problem involves two main questions. First, what are the constraints on language change that inhibit – but do not rule out – the development of the rare feature? Second, what are the developmental pathways that nonetheless lead to the emergence of the rare feature?

Linguists have given different answers to these questions, but several main types can be identified (see Table 1). First, a given feature may be (relatively) rare because there are fewer pathways that lead to the feature than away from it. Bybee (2001: 195–197) provides evidence for the argument that there are more open syllables than closed syllables, and only open syllables are (nearly) universal, because new open syllables are constantly being created by regular processes of language change (e. g., coda weakening and loss), while there are fewer processes that lead to closed syllables. Another possibility is that some rare features may necessitate numerous diachronic “steps” that occur in a certain order in order to develop, as in Harris’s (2008) account of Georgian split case marking or Udi endoclitics. Yet another possibility is that a certain feature may require rare input structures, as in Grossman et al. (2015), which argues that adverbial subordinator prefixes are rare because they are facilitated by the relatively rare VSO order and case prefixes and are inhibited by other word orders or case suffixes. Fourth, some changes simply may be more frequent than others, as proposed by Greenberg (1978), a view adopted by Blevins (2009), who states that most languages have coronal segments because coronal maintenance and coronal creation are more frequent than “coronal annihilation”. While the last might not seem like the sort of change one normally encounters in historical linguistics textbooks, it is nonetheless a documented change. Of course, it is generally thought that some changes are regular or require only a single step or causal mechanism, while others are sporadic or require multiple steps or causal mechanisms, e. g., assimilation vs. metathesis. Finally, some properties might simply be inherently unstable, and even though they could develop frequently in language, they might tend to develop into other categories or to disappear. For example, Blevins (2008) suggests that three-way vowel length distinctions may “push the perceptual envelope” and therefore be inherently prone to loss. ^[1]

What is common to these diachronic explanations is that they all take crosslinguistic rarity as a problem to be dealt with, explicitly addressing the question how a given linguistic feature can be dispreferred for some reason and nevertheless develop and (perhaps) remain stable over time. It goes without saying that all of these explanations may make reference to a variety of additional factors, whether based on usage, acquisition, or general properties of human physiology and cognition. Importantly, these types of factors that contribute to crosslinguistic rarity are not mutually exclusive, and all things being equal, one would assume that a feature that is characterized by more of the above factors would be rarer than one that is characterized by fewer of them.

This article has three goals. The first is to document a rarissimum (Wohlgemuth & Cysouw (eds.) 2010a, 2010b) of person marking: in the Sahidic dialect of Coptic (Afroasiatic; Egypt), the 2sg.f person index is zero. While the facts of Coptic are well known to specialists, thanks to descriptive grammars (e. g., Layton 2004) and an in-depth study by Uljas (2009), the existence of the proposed rarissimum is unknown to typologists. There are no documented parallels in other languages, whether within or outside Afroasiatic.

The second goal is to provide a diachronic explanation for the rarity of this linguistic feature. Specifically, it is argued that the first reason that zero-marked 2sg.f is rare is that there are relatively few pathways that could lead to such a paradigmatic zero. This derives from two main facts: first, the source constructions from which such a feature could arise are relatively rare; and second, most pathways of change involving differential loss of overt markers, e. g., through phonological erosion, would not lead to zero-marked 2sg but rather to different types of homophonous or syncretic systems without a paradigmatic 2sg at all. Second, adding to the rarity of this feature is the fact that its source construction, i. e., gender marking in person forms, is itself crosslinguistically rare. Third, the feature documented here does not result from differential phonological erosion, based, e. g., on relative frequency, but rather from a series of conditioned sound changes that occurred in particular stages. One of which – depalatalization – is of a relatively rare type. As such, the development of zero-marked 2sg.f in Coptic involves all of the factors discussed above that contribute to the crosslinguistic rarity of features.

A third goal is to document a rare pathway of change, which is the converse of the better-known “Watkins’ Law”, which leads to zero marking of 3sg through reanalyzing a 3sg marker as part of the verb stem. In the pathway of change documented here, a zero marker is reanalyzed away, with the result that an erstwhile part of a morpheme is reanalyzed as a segmentable person marker. This process in effect leads away from the zero-marked 2sg.f, reinstating an overt segmental person marker.

Taken together, these three stages – the identification of a synchronic rarity, the identification of a language-specific diachronic explanation, and the identification of crosslinguistic inhibiting or facilitating factors – provide the necessary first steps of a diachronic account of a crosslinguistic rarity. This approach basically follows Greenberg’s (1966a) line of reasoning with respect to explanation in linguistics.

The structure of this article is as follows. Section 2 presents some brief background about zero person marking in crosslinguistic perspective. Section 3 sketches the relevant Coptic data. Section 4 provides evidence that this feature is absent from other Afroasiatic languages, and cannot be treated as an inherited feature, arguing instead that this feature has a straightforward diachronic explanation, and raising questions about its diachronic stability. Section 5 discusses the reasons for the crosslinguistic rarity, arguing that there are few diachronic pathways through which such a feature could arise, while Section 6 argues that this feature is in fact diachronically unstable. Section 7 offers brief concluding remarks.

2 Zero person marking in crosslinguistic perspective

In one of the first discussions of universals of person marking, Uspensky (1972: 68, cited in Cysouw 2003: 57), claims that “[i]f a zero expression occurs in the form of a certain person in the indicative mood, then, included in the meanings thus expressed (i. e. by a zero mark) we find the meaning of the 3rd person or that of the 1st person”. Cysouw (2003) corroborates this universal, noting several exceptions in which “Latin-type” paradigms, i. e., paradigms with no syncretism or homophony between any of the singular persons, have zero marking for the 2nd person. For example, in Bongo (Nilo-Saharan; Sudan), independent pronouns have bound (“clitic”) realizations when nothing intervenes between the pronoun and the verb. Thus, independent 1sgma is realized as m- (1a), independent 3sgbu is realized as b- (1b), but independent i disappears (1c).

However, Cysouw notes that zero 2nd person marking is extremely rare in his sample, even if it appears to be slightly more frequent than zero 1st person marking. In her sample of 347 languages, Siewierska (2009: 429) finds that zero-marked 1st or 2nd person is “too infrequent to warrant sub-classification” into types of zero marking. On the other hand, the association of zero marking with 3rd person has often been noted and re-confirmed (e. g., Ariel 2000; Bybee 1985; Greenberg 1966b; Siewierska 2009), although this observation might need revision in light of a recent argument that there is in fact “no evidence for zeros to be more common among third than among non-third persons” (Bickel et al. 2015).

These statements refer to what Siewierska (2004: 24) calls zero “in the paradigmatic sense of the term”, i. e., person paradigms in which there is a combination of zero and overt person markers. Siewierska (2013) shows that in a 380-language sample, zero 3rd person marking of the S argument is documented in 17 % of the languages (n = 66), and another 9 % (n= 21) could be analyzed as having obligatory zero marking in all 3rd persons, but as Siewierska points out, such languages could also be analyzed as having no 3rd person forms at all. Based on these data, one can conclude that while zero marking in the 3rd person is well documented, it is hardly the majority case.

Bickel et al. (2015) test the hypothesis that “zeros are assumed to develop and be preserved more commonly in third than in non-third person”. They find no support for this hypothesis as a synchronic universal: only 35 % of languages have more zero markers in the 3rd person than in other persons, the other 65 % having what the hypothesis would consider “disfavored” structures. Furthermore, Bickel et al. (2015) find no evidence for the hypothesis as a diachronic universal pertaining to paradigms. However, when it is interpreted as a hypothesis about the distribution of zero person marking cutting across paradigms (i. e., “zero markers are more likely to be found in the third than in the first and second person, across all paradigms in a family”; Bickel et al. 2015:4), it is weakly supported.

Interestingly, to the best of my knowledge, large-scale typological studies of person marking have consistently left out gender distinctions in person markers, generally seeing them as a confound (Cysouw 2003; Bickel et al. 2015).

3 Background: Coptic person markers

Coptic ^[2] has a crosslinguistically unremarkable set of independent person markers, which distinguish singular and plural number in 1st, 2nd, and 3rd persons, and masculine and feminine gender in the singular 2nd and 3rd persons, as shown in Table 2. Such systems are common in Afroasiatic languages, a fact often pointed out in typological studies of person (Cysouw 2003; Siewierska 2004).

On the other hand, Coptic has a highly complex series of bound person indexes, which distinguish the same categories but show considerable allomorphy, depending on the morphosyntactic environment and the phonological properties of the host (Polotsky 1960; Layton 2004). For example, prefixed person indexes differ in a synchronically unpredictable way depending on the verbal template. Table 3 shows two paradigms of the verb -me ‘love’, the first of which shows the present tense, with verb-initial (prefixed) A/S person indexes, and the second of which shows the past tense, with verb-initial TAM/polarity markers followed by A/S person indexes. Note that while 2sg.m, 3sg.m, 3sg.f, and 2pl person indexes are identical in both paradigms, those for 2sg.f, 1pl, and 3pl (marked in bold) differ.

For a given person, A/S and P indexes are identical in some cases and differ in others, again depending on the morphosyntactic environment and the phonological properties of the host. For example, consider the A/S and P 3sg.m indexes, which are identical, in (2a, b):

However, some persons show allomorphy of the A/S marker depending on the morphosyntactic environment. For example, while the 3pl marker in the past verb form is -u- (ex. 2b), in the present it is se-, as in (3).

The allomorphy of the P indexes, on the other hand, is determined by the phonological environment. Simplifying matters somewhat, if the lexical verb ends in a short vowel, the bound P marker is -i, while if it ends in a consonant or a long vowel, it is -t (for a more precise description, see Layton 2004). ^[3]

Focusing on 2sg.f A/S indexes in main clauses, we find that zero marking is the norm. Table 4 compares several TAM-marked verbal constructions: (i) the verb form with non-2sg.f person index, (ii) the verb form with 2sg.f, and (iii) the verb form with lexical noun phrase subject. The lexical verb used here is -me ‘love’, and the lexical NP subject used is p-rôme [def.m.sg-man] ‘the man’.

In Table 4, we see that the 2sg.f person indexes in Coptic are often paradigmatic zeros, noting also that the TAM formatives occurring with the 2sg.f are precisely those that occur with lexical noun phrase subjects. In fact, 2sg.f is usually zero-marked, although it also has non-zero allomorphs. In some cases, these non-zero formatives are regular allomorphs, as in the present tense.

In other cases, zero and non-zero formatives vary in the same construction, and can even occur as textual variants in different manuscripts, as in (6).

The distribution of zero and non-zero 2sg.f variants is complex, and a full account is beyond the scope of this article, but Uljas (2009) provides an exhaustive description. ^[5] The important thing, for present purposes, is that Coptic has a number of constructions in which 2sg.f is represented by a paradigmatic zero.

Nonetheless, it is important to keep in mind two facts, to which we return in more detail in Section 6. First, Coptic 2sg.f has non-zero allomorphs. Second, Coptic has a number of constructions in which 2sg.f could be analyzed as zero, but could also be analyzed as non-zero, usually involving -r(e)-. This ambiguity results from the fact that TAM formatives that occur with 2sg.f reference are often homophonous with TAM formatives that occur with lexical noun phrase subjects. For example, the Aorist (basically, a habitual form) formative that occurs with most bound person markers is ša- (7a), but šare- when it occurs with 2sg.f (7b) or lexical noun phrase subjects (7c).

4 So where did Coptic get its zero 2sg.f?

First of all, we can safely rule out inheritance as the source of this rare feature. It is not reconstructible for Proto-Afroasiatic, since Proto-Semitic (Goldenberg 2013), Proto-Chadic (Frajzyngier 2012, personal communication), and Proto-Berber (Kossmann 2012, personal communication) have overt 2sg markers, usually involving a dental or alveolar consonant. Cushitic and Omotic languages tend to have distinctive 2sg person markers (Mous 2012, personal communication; Amha 2012). Most importantly, zero 2sg is not attested in the early stages of Ancient Egyptian, which has an overt 2sg marker, /c/ (orthographically). As such, we can safely conclude that zero-marked 2sg in Coptic is not an inherited feature, but rather a language-internal development.

When seen from a diachronic perspective, it turns out that there are several documented processes by which a language can develop zero person marking. One such process is known as Watkins’ Law, according to which “third person markers are reanalyzed as part of the verbal stem, giving thus rise to zero marking in the third person” (Bickel et al. 2015), a process that Watkins posited to account for the reanalysis of the 3sg ending -t as part of the verbal stem in the course of development from Proto-Iranian to Persian (Watkins 1962: 94; see Siewierska 2009 and Bickel et al. 2015 for a full discussion).

Another possibility is that person marking simply never developed for a particular person, but did for others (i. e., differential grammaticalization). Yet another is differential phonological erosion, perhaps due to frequency (Bybee 2007; Haspelmath 2008).

Zero 2sg.f person markers in Coptic developed due to sound change, but not in a way that is easily explained by most phonological accounts of the development of zero markers, which have looked mainly to frequency-driven explanations. In Coptic, the 2sg.f zero marker is the result of a long-term process of conditioned segmental sound changes, in which:

1. original /*k/ was palatalized to /c/ before high front vowels,

2. in some environments, /c/ merged with /t/ by depalatalization or fronting of the place of articulation, and

3. /t/ was lost in a number of environments; it was almost invariably lost when word-final, perhaps first going through a stage of debuccalizing to /Ɂ/.

The details of these changes are still poorly understood, but the main line of development is clear (Kammerzell 1998, 2005; Peust 1999), and crucially, word-final position is the position in which these changes consistently took place. Table 5 shows these developments; the changes involving word-final position (/_________#) do not mean that this is the only environment in which the change took place, but rather a weaker claim, i. e., that in this environment the change took place.

Evidence for the first change (i), i. e., the palatalization of *k > /c/ before high front vowels, is based on reconstruction and on language-internal phonological analysis (Kammerzell 1998, 2005). Evidence for the second two changes is provided in Table 6 (Loprieno 1995).

Importantly, none of these sound changes are outlandish per se, although one, the depalatalization or fronting of /c/ to /t/, does seem to be relatively rare.

1. The palatalization or coronalization (Telfer 2006) of velars is well documented in the world’s languages (Bhat 1978; Telfer 2006; Bateman 2007).

2. Change from /c/ to /t/, while rarer, is also documented, both as the result of a synchronic alternation and as the result of postulated historical changes. For example, in Korean (isolate; North and South Korea), /c/ is neutralized to /t/ in coda position (Sohn 2001: 163–165). In Tarahumara (Uto-Aztecan; Mexico), the alveolopalatal affricate can be depalatalized and realized as an alveolar affricate before /a/ (Caballero 2008: 36). ^[6]

3. Finally, loss of t, whether via debuccalization, e. g., /t/ > /Ɂ/ as in some English dialects in syllable-final position, as well as in Burmese, Kagoshima Japanese, and Yamphu (O’Brien 2010, 2012: 10), or via deletion is extremely well documented across the world’s languages.

The only change that is apparently rare is the depalatalization or fronting of /c/ to /t/.

5 Why aren’t 2sg person zero markers more common?

6 A rarissimum: (In)stability

One of the alleged characteristics of rarissima is that they might be unstable, diachronically speaking (Wohlgemuth & Cysouw (eds.) 2010a). The Coptic zero 2sg.f marker corroborates this hypothesis. As Uljas (2009) shows, the situation in which the 2sg.f marker was zero, following a TAM formative identical to that which occurs prefixed to lexical noun phrase A/S – but crucially, not to most person indexes – was reanalyzed such that -r(e)- was reanalyzed as a 2sg.f person marker. Table 9 presents this process, taking the Aorist verb form (basically, a habitual) as exemplary. Admittedly, this is a highly idealized version of the situation. As Uljas (2009) shows, there is much synchronic variation between zero 2sg.f and overt 2sg.f markers, but a clear diachronic process can be identified, in which zero 2sg.f markers are replaced by overt ones involving -r(e)-.

As Uljas (2009: 178–179) points out, this is a crosslinguistically unusual pathway of grammaticalization, leading from a sub-morphemic part of a TAM marker to a bound person marker or pronoun. What Uljas did not point out is that this is the converse of the phenomenon described by Watkins’ Law, in which a segmentable morpheme (the person index) is reinterpreted as part of a morpheme (the base):

This instability itself is of interest. It is unlikely that zero-marked 2sg.f gave way to an overt 2sg.f marker because of the simple fact of its rarity across languages. It is hard to see how a crosslinguistically uncommon situation could affect the types of synchronic factors that are implicated in language change, unless one assumes that the linguistic knowledge of speakers of individual languages somehow captures crosslinguistic generalizations. Moreover, this kind of explanation runs into problems, since crosslinguistically uncommon structures can be stable over time (Harris 2010).

However, the process could be motivated by more general principles. One such motivation could be frequency, but it is hard to see what kind of prediction a token frequency-based account would make. Based on Bybee’s “Preserving Effect”, one could predict that structures with low token frequency would be more likely to undergo analogical change. However, there is little reason to think that the 2sg.f differs significantly from the 2sg.m in terms of token frequency, and there is similarly little reason to think that the 2sg in general had a low token frequency in actual everyday speech.

Another possibility is analogy, close to what Haspelmath (2014b) called “system pressure”, in which “[r]ules of grammar generally target large classes of items, rather than individual expressions or small classes” (Haspelmath 2014b: 197). Here “system pressure” removes the anomaly of having three “unnatural” rules for a single verbal template. The allomorphy of the aorist TAM marker (the subscript numbers represent different allomorphs of the TAM marker) is:

As the result of reanalysis, this allomorphy is reduced, such that all bound person markers are preceded by the same allomorph of the Aorist TAM marker:

Another way of looking at the relatively vague notion of “system pressure” is provided by Bybee’s (1985, 2007) conception of the role of type frequency in productivity. As Bybee points out, “when a construction is experienced with different items occupying a position, it enables the parsing of the construction” (Bybee 2007: 15), illustrating the claim with the following example:

If happiness is learned by someone who knows no related words, there is no way to infer that it has two morphemes. If happy is also learned, then the learner could hypothesize that -ness is a suffix, but only if it occurs on other adjectives would its status as a suffix become established. Thus a certain degree of type frequency is needed to uncover the structure of words and phrases. In addition, a higher type frequency also gives a construction a stronger representation, making it more available or accessible for novel uses.

In the present case, it could be proposed that the higher type frequency of the shorter TAM allomorph (ša-), which occurs with all bound person markers except 2sg.f, provided the basis for reanalysis of the longer allomorph (šare-), which occurs only with the 2sg. f. The following examples repeat the sentences in (9) but are glossed differently to show the results of this process more clearly:

Further evidence for this analysis is provided by Uljas (2009), who shows that the overt 2sg.f marker -r(e)- spread to other verb forms via analogy or “affix pleonasm” (Haspelmath 1993). Example (11a) shows the older 2sg.f marker without -r(e)-, while (11b) shows an innovative 2sg.f marker:

Note that this account, even though it eliminates zero 2sg.f A/S indexes, actually leaves zero 2sg.f in the system, as a P index (see 7b) above. As such, the reanalysis discussed here eliminates complexity in one part of the person-indexing system, i. e., in the allomorphy of TAM prefixes, but increases it in another, i. e., in the allomorphy of A/S vs. P indexes.

7 Conclusions

The evidence presented here indicates that zero-marked 2sg.f is a rarissimum, i. e., rare – perhaps unique – in the world’s languages and rare – perhaps unique – within Afroasiatic. Of the four types of diachronic scenarios that contribute to the crosslinguistic rarity of linguistic features, the development of Coptic zero-marked 2sg.f shows all four. This feature developed in an ecology of a rare constellation of source construction features, i. e., a “Latin-type” system, one in which all singular persons are overtly marked; even rarer is the gender distinction in the 2sg. Moreover, few pathways of regular language change would lead to a situation in which there is a specifically zero-marked 2sg. Upon this rare situation operated a series of sound changes, one of them of a crosslinguistically rare type, leading to the ultimate loss of the overt 2sg.f marker, creating a zero-marked 2sg.f. Altogether, the development of the zero-marked 2sg.f involved multiple stages that had to occur in a particular order on particular input structures. Strikingly, this situation was diachronically unstable, and a sub-morphemic part of preverbal TAM prefixes was reanalyzed as an overt 2sg.f marker -r(e)-. This leads to the possibility that yet another type of diachronic factor may be involved in crosslinguistic rarity, namely, the inherent instability of certain features. While this may smack of teleology, it may nonetheless be the case that certain features are indeed relatively unstable. For example, Blevins (2008) observes that the “three-way contrast between oral, weakly nasalized, and fully nasalized, documented for Palantla Chinantec is also extremely rare, occurring only in this language, where it appears to be unstable”, noting that three-way length contrasts in Estonian, Saami, and Dinka may also be relatively unstable.

The case of the zero-marked 2sg.f in Coptic is yet another corroboration of the generalization that crosslinguistically rare linguistic structures are not ruled out by general principles of universal grammar, i. e., they are learnable and transmissible throughout generations. Rather, synchronically rare linguistic structures can and do arise as the result of regular processes of language change. However, even regular processes of language change operate on synchronic structures, and rare synchronic structures can be preserved or lost, or they can give rise to rarissima, which in turn can be preserved or lost. This article is intended to provide evidence for the latter, in hopes of one day having enough crosslinguistic data on pathways of change to ask what conditions favor the preservation or elimination of crosslinguistically rare features.