The grammaticalization of noun affixes: a cross-linguistic study

1 Introduction

Most definitions seem to agree that affixes are phonologically overt forms that instantiate a semantically consistent function and attach to larger units such as stems or words (e.g., ten Hacken 2005: 11). Based on such a definition, affixes are types of morphemes and an important and intuitive mechanism for the description of many word forms in many languages. However, there are also areas in which this definition runs into difficulties. For instance, the word unbeknownst appears to contain the affixes be- and -st, given that these items remain after the rest of the word is successfully segmented. Yet, it is readily apparent that be- and -st in unbeknownst do not bear any discernible function within that word. As such, they seem to be instances of “empty morphs,” i.e., forms that are left over after an otherwise plausible morphological analysis. It is phenomena such as empty morphs that have cast doubt on the appropriateness of the concept of affixes. In fact, a large portion of contemporary morphological research argues against the assumption of affixes as the primary units of morphology, preferring instead to take the fully inflected word form as the basic level at which morphological generalizations should be made (cf. Blevins 2016: Chs. 2, 3; Stewart 2016: 5 for a review of the arguments).

Rather than discussing the merits and demerits of morpheme- and word-based approaches in abstract terms, the present work will deal with one particular type of deviation from the affixal ideal. The elements of interest are another kind of empty morph and will be referred to as “noun affixes.” The main (and often only) purpose of noun affixes as understood here is to indicate that the word form in which they occur is a noun. This phenomenon is relatively widespread cross-linguistically and was famously discussed in Greenberg (1978) under the name of “Stage 3 articles.” One suggestive case of noun affixes can be found in Iroquoian. Specifically, the noun affix is described for languages of the Northern branch, but not for the only member of the Southern branch, Cherokee (chr); cf. Montgomery-Anderson (2015: Ch. 6); Uchihara (2013: 34).^[1] Example (1) illustrates the noun affix in Onondaga. Here and throughout this work, the relevant element will be highlighted in boldface.

While the prefixal element in (1) and analogous words is described as a “noun prefix,” the description in Woodbury (2018: 283–285) states clearly that this element expresses animacy distinctions. As such, it encodes a straightforward grammatical function and is therefore a “regular” (as opposed to an empty) morph. This leaves the boldfaced suffixal element, about which Woodbury (2018: 294) says that it “contributes no meaning to the word except to mark it as a noun.” Descriptions of this item in other Northern Iroquoian languages are often no more elaborate than this. For instance, with respect to the noun suffix in Seneca (see), Chafe (2015: 86) provides an almost identical and equally brief description: “This suffix contributes nothing to the meaning of the word beyond the fact that it is a morphological noun.” (Cf. also Dyck et al. 2024: 135 on Cayuga; cay.) The impression of a purely noun-marking device also emerges from overviews of Northern Iroquoian more generally. Mithun’s (1999: 427) single piece of information about this item explicitly states that it does not express a meaning: “The [noun] suffix, most often –aʔ, simply identifies the word as a noun.” Similarly, Barrie and Uchihara (2020: 430) mention that the element is generally required for the formation of nouns, but they do not mention any kind of semantic value that it might contribute.

Given the above definition of affixes, the Northern Iroquoian noun affixes constitute a paradox in that they only signal the morphosyntactic fact that the ambient word form is a noun. This information is often redundant since it can typically be inferred from syntactic, pragmatic, or morphological clues. Therefore, Greenberg (1978: 69) refers to noun affixes of this type as “non-functional.” Their non-functional character presumably also explains why noun affixes are absent from most languages. At the same time, there are good reasons to assume that the Onondaga element is an affix and not simply part of the noun’s segmental make-up. The best evidence comes from noun incorporation processes, where only the noun root, minus any inflectional affixes, is incorporated (Woodbury 2018: 224). Since the noun affix behaves like other, more typical kinds of inflectional affixes with respect to this process, it can plausibly be classified as an affix itself. A relevant example of noun incorporation is illustrated in (2), where the noun occurs in its bare form nęh.

The existence of noun affixes raises the question as to how and why such an otherwise unexpected structure comes into being. This is the main question addressed by Greenberg (1978). He argues that the elements that I refer to as “noun affixes” represent the final stage on a diachronic continuum that leads from demonstratives (Stage 0) via definite articles (Stage 1) and markers that express both definiteness and (specific) indefiniteness (Stage 2) to pure markers of nominal status (Stage 3). To the extent that this scenario is correct, it stands to reason that there is variation in terms of how much the behavior of noun affixes still resembles that of more typical affixes, both formally and functionally. While Greenberg (1978) was careful to highlight the many fine nuances involved in those domains, he did not relate his diachronic proposal to the concept of grammaticalization.

It is that gap that the present work attempts to fill. Specifically, its main contribution will be to argue that present-day noun affixes from across the world, in conjunction with Greenberg’s (1978) historical claim, can inform certain aspects of grammaticalization theory. The necessary background on those aspects will be outlined in Section 2. Meanwhile, Section 3 will provide several case studies of noun affixes, with a focus on items that are rather clear instances of Stage 3. In so doing, I will address several facets of the phenomenon that Greenberg (1978) did not discuss. For example, it will be shown that noun affixes are widely found in the Americas, whereas Greenberg (1978) primarily focused on Africa. Furthermore, these discussions will highlight that the factor of paradigm size, which Greenberg (1978) did not mention, is crucial for understanding the diachronic pathways of noun affixes. I will also present evidence that one purpose to which highly grammaticalized noun affixes may ultimately be put is to guarantee a minimal phonological word size. In Section 4, I will aggregate these points and sketch a more comprehensive account of how noun affixes might develop. The conclusion in Section 5 will point to additional aspects of noun affixes that deserve to be investigated in future research.

2 The course of grammaticalization

Grammaticalization is typically described as the process by which elements become more grammatical over time due to a number of changes on different levels. While the notion that these combinations of changes constitute a genuinely distinct process is not at issue here (cf. Campbell 2001 for critical discussion), the fact remains that the different changes that produce grammatical items have to play out in real time. Thus, it is an empirical issue whether semantic change precedes formal change or vice versa, or whether they proceed in lockstep. One typologically informed account that argues for the latter option is the “Parallel Reduction Hypothesis (PRH)” (Bybee et al. 1994: 19–21, Ch. 4). Here, the argument is that more semantically grammaticalized elements also show more signs of formal grammaticalization. Since the indicators of grammaticalization are mostly different types of reduction that the item at issue undergoes (e.g., loss of phonological segments, loss of morphological boundaries, loss of paradigmatic competitors, loss of semantic specificity), the ultimate claim of the PRH is that the most abstract grammatical meanings should be expressed by the shortest forms. Taking the PRH to its logical conclusion, form and function should then also arrive at the zero stage simultaneously. On a strict reading of this claim, then, neither empty morphs (i.e., forms without meaning) nor their converse, zero morphs (i.e., meanings without form), should arise as a product of grammaticalization.

The PRH has attracted considerable criticism. Specifically, it has been argued in more recent work that there are many counterexamples to it, all of which point in the same direction: Grammaticalization is always initiated (and thus defined) by functional changes (whether semantic or pragmatic), while formal reduction either lags behind or fails to manifest itself entirely (e.g., Bisang et al. 2020: 3–4, 7, 40–41, 56, 79; Heine 2018: 19, 21, 25; Heine and Kuteva 2007: 35, 42; Kuteva et al. 2019: 4; Narrog and Heine 2018: 1–2, 14; cf. also Gildea 1998: 41–42, 209, 264 n. 7). In addition, Heine (2018: 21, 25) states that formal changes have never been suggested to precede functional changes, and he concludes that the former are an epiphenomenon of the latter. This conception of grammaticalization is in line with an earlier account by Hopper and Traugott (2003: 76, 82, 174), who emphasize that it is changes in function that are crucial during the early stages of grammaticalization and that it is the final stages of the process that are instead characterized by formal reduction.

In sum, most accounts posit that formal and functional changes do not proceed in parallel. Obviously, this opens up the possibility that form-function mismatches such as noun affixes come about via regular grammaticalization. In that context, it is important that recent accounts continue to acknowledge that grammaticalization involves a zero stage. For instance, Kuteva et al. (2019: 6) propose that “grammaticalization ideally ends in zero – that is, grammatical forms increasingly lose in semantic and phonetic content.” However, if functional change generally precedes and both form and function ultimately progress toward zero, the question is whether this means that the semantic dimension reaches the zero stage earlier due to its “head start.” This is not a trivial question because semantic reduction might simply take longer than its formal counterpart and might subsequently not arrive at the zero stage before the formal dimension does.

While the accounts cited above explicitly address how grammaticalization processes begin, they largely leave open how they might end. The argument of the present work is that noun affixes can shed light on this issue because they are among the rare units that have approximated the zero stage on either dimension. For instance, the Onondaga noun suffix discussed in Section 1 seems to constitute strong evidence for the idea that semantic reduction is completed before formal reduction: While the item has entirely lost any meaning that it may have had, its form has not yet disappeared.^[2] A considerably refined version of this claim will be offered in Section 4.

A final note on Greenberg’s (1978) diachronic proposal is in order. Technically, his account relies on the idea that there are different pathways for Stage 2 markers (Greenberg 1978: 63, 69). One option is a simple loss of the element. However, he also mentions that, if the demonstrative that initiated the chain of changes originally distinguished noun classes/genders, the Stage 3 items resulting from that demonstrative tend to retain that function. Hence, a noun marker that goes back to a demonstrative that was sensitive to class/gender will usually not be an empty morph but a gender marker. Put differently, noun affixes are supposed to only derive from demonstratives that did not distinguish gender. Yet, some of the data discussed below suggest that there actually is a link between gender markers and noun affixes, which would require a revision of Greenberg’s (1978) cline. This will be addressed in Section 4.

3 Noun affixes: case studies

In the following sub-sections, I will provide case studies of noun affixes from different macro-areas and language families. These data are subject to methodological and empirical constraints. First and foremost, the languages for which the phenomenon will be illustrated here do not constitute the result of a dedicated sampling process. Instead, they feature noun affixes that are relatively well described, which supports the analysis that they are noun affixes to begin with. The focus on the families described below is mostly a by-product of the work presented in Zingler (2020), which used a sample of 60 unrelated languages to investigate a different topic altogether (i.e., wordhood). If the 60 families in that sample are representative, the six families illustrated here might mean that noun affixes occur in about 10 % of all families. In conjunction with the number of languages discussed by Greenberg (1978), this suggests that noun affixes are a fairly common phenomenon that deserves further study.

The second constraint concerns the description of the noun affixes in the sources available to me. Since any one feature of a language is but a small domain within the overall system, grammars usually do not analyze any given property in great detail. This tendency is arguably even stronger for rather opaque phenomena such as noun affixes. Therefore, the information that I will be able to provide in the following sub-sections is often rather limited. In an effort to make up for this lack of detail, data from related languages will be drawn on wherever possible. As will be seen, this procedure often produces insights that a focus on a single language would not have permitted.

The following case studies are roughly ordered from more to less comprehensive. The empirical analyses will be kept to a necessary minimum; a summary and a comparison of the relevant facts will be provided in Section 3.6.

4 Noun affixes and grammaticalization (theory)

The main goal of this work is to investigate to what extent the cross-linguistic behavior of noun affixes can inform the question of how grammaticalization processes end. More specifically, the focus is on the temporal relationship between semantic changes and formal changes, which are taken here to mean semantic and formal reduction. The simple answer is that noun affixes strongly suggest that semantic reduction is completed before formal reduction. All the case studies dealt with items that more or less fall into the “non-functional” Stage 3 of Greenberg’s (1978) typology. Yet, all these elements could only be identified because they still had a phonological form. This forces the conclusion that the noun affixes have lost all of their original function(s) but not all of their formal manifestation. On a more general level, this seems to mean that the noun affixes disconfirm the assumption of the PRH, according to which form and function reduce in parallel. On the other hand, they crucially complement the alternative claim that formal and functional changes in grammaticalization processes manifest themselves in an asynchronous fashion.

Of course, noun affixes constitute but a very small niche within the large network of grammatical structures, and additional studies will have to determine whether the current conclusion is representative of grammaticalization as a whole. In particular, it would be of interest if there are also processes of grammaticalization (rather than of reanalysis or analogy) that create zero morphs since it would then have to be concluded that formal reduction can also outpace semantic reduction. For the remainder of this paper, however, the relevant result is simply that noun affixes are prime examples of empty morphs and of the diachronic forces that shape them. In the following, I will sketch what else the case studies reveal about the behavior of items that are close to the end of their grammaticalization.

Arguably the most salient result that emerges from Table 1 is that the noun affixes are absent from certain morphosyntactic contexts. With the exception of the Western Otomanguean case, which might be trivial for the reasons mentioned in Section 3.5, this is true of all the items discussed. In fact, Greenberg (1981: 106) points out that noun affixes are often dropped in compounding and incorporation constructions, which are also among the most common environments attested in the case studies here. The reasons for these two restrictions are easy to see if Greenberg’s (1978) claim that noun affixes were previously definiteness markers is taken into account. More specifically, nouns that do not refer often lack some of the common nominal inflections (cf. Hopper and Thompson 1984: 710–711). Hence, since modifiers in compounds do not refer to actual real-world entities, the nouns expressing them might not be eligible for definiteness/specificity marking, whose purpose is precisely to identify a particular set or entity. It has to be said, though, that this approach does not manage to sufficiently explain the Tacanan data, where it is the more (typically) referential heads of compounds that lack the noun prefix.

In the same vein, Mithun (1984: 849) illustrates her “Type 1” of noun incorporation constructions by stating that the incorporated berry in berry-picking does not refer to a specific berry and “[b]ecause it does not refer, it is not marked for definiteness […].” Note that whether compounding and noun incorporation are different phenomena is not crucial to the present argument. Rather, the important takeaway is that the omission of noun affixes from these two contexts is a straightforward carryover from their previous status as definiteness markers. While this argument does not extend to all of the relevant contexts in Fwe and Jalkunan, it should be mentioned that one of the environments in which the Fwe augment fails to appear is with personal names, i.e., with inherently referential nouns that are relatively unlikely to co-occur with an overt definiteness marker.

A further context in which the Fwe augment is subject to restrictions is with locative markers. This roughly unites it with several other noun affixes discussed here, which cannot co-occur with adpositions. The latter restriction can be explained with reference to Himmelmann (1998), who shows that definite and specific articles are commonly restricted and/or omitted in adpositional constructions. He also shows that this interaction is partly related to the notion of referentiality hinted at above. The syntactic and semantic details that Himmelmann (1998) invokes in order to explain this interdependency are too complex to summarize here. Yet, the fundamental point is that the omission of noun affixes in the presence of adpositions can again be explained when taking into consideration that many (or most) noun affixes derive from definiteness markers.

The second-most common property mentioned in the case studies is free variation, which Greenberg (1981: 107) also named as a feature of Stage 3 articles. While free variation does not figure prominently in grammaticalization accounts,^[8] it seems compatible with the concept of “layering.” In their discussion of this phenomenon, Hopper and Traugott (2003: 126) mention that competition between a new and an old construction may involve a stage at which the pragmatic difference between them is not clear. This stage presumably corresponds to the intermediate position in the well-known A → A/B → B model of language change. In any case, since pragmatic factors are often invoked as possible explanations of apparent free variation (cf. Section 3.1), the actual absence of such pragmatic factors during “layering” would render claims of free variation much more plausible.

Free variation of noun affixes can arguably also be understood at a more intuitive level. Since semantic change (including semantic reduction/loss) is not only gradual but also specific to individual language users, a developing noun affix may still perform its previous function for some speakers and/or in some contexts but not for others and/or in other contexts. Such alternations would be a necessary prerequisite for free variation even though it has to be reiterated that no claim regarding the existence of full-fledged free variation is intended here. More importantly, the stage of (apparent) free variation is likely to betoken the impending loss of the noun affix. Put differently, it is unlikely that a noun affix will be retained despite its lack of a clear function and after the alternative of omitting it has already been introduced into the system. So, while contexts of omission indicate formal traits that characterized the ancestors of noun affixes, free variation seems to be a property that affects noun affixes only once they have arisen as a phenomenon in their own right.

Each of the remaining properties only factors into two of the genera. Yet, they still have the potential to inform the question of how noun affixes grammaticalize. For instance, the fact that the Tacanan and Numic noun affixes are restricted to certain semantic classes suggests that they have their origin in a gender system. Given that the semantic patterns described are not entirely coherent, though, they have basically become noun affixes. This is highly relevant because Greenberg (1978: 69) draws up a dichotomy between Stage 3 markers that encode gender and those that do not. Hence, if the Tacanan and Numic elements were correctly classified here, they suggest that gender markers may ultimately shed their function and become pure noun affixes. Tacanan might also be said to have progressed further along that path than Numic because the former only has a single noun affix, which clouds any potential origin in a gender system. Overall, this analysis might require an alteration to the original cline so that, for instance, a Stage 3a includes gender markers and is diachronically linked to a Stage 3b, which subsumes noun affixes. Yet, a principled decision on this matter requires a more thorough semantic investigation than was performed here.

The morphosyntactic properties of the noun affixes also raise several intriguing questions. For example, the two African noun affixes have considerably more distributional freedom in terms of the word classes they can occur with than do the other markers analyzed. In the domain of concord across the NP, which might be relevant to Fwe, such behavior is typically not remarked upon as a reason to classify the items in question as anything other than affixes. Yet, the fact remains that they are “non-selective” (Haspelmath 2011: 45–47) with respect to syntactic category and are therefore not prototypical affixes. However, the evidence provided for the Fwe augment is too sparse to determine whether it actually shows concord. If it does, this could be seen as a characteristic of a lower degree of grammaticalization because a concord marker is still sensitive to its semantic and morphosyntactic environment and has not yet become an invariant and “non-functional” appendage of the noun. On the other hand, if the Fwe augment can variably occur on any element within the NP (but only once overall per NP), it would be even less clear how to classify it given established morphological terminology. By contrast, the Jalkunan marker is a fairly obvious instance of an NP-final clitic. While it might be subject to a higher number of idiosyncratic restrictions than clitics usually are (cf. Zwicky and Pullum 1983: 505), it seems to be a straightforward enclitic whenever and wherever it does occur. On the traditional view, a clitic is less grammaticalized than an affix (e.g., Hopper and Traugott 2003: 7), which would seem to align clitics with the status of concord markers as possibly found in Fwe.

Overall, then, the two African markers show that grammaticalizing elements may not only have retained some of their phonological form but also much of their syntagmatic potential by the time they have become semantically empty. Of course, this is compatible with the general idea of form-function mismatches that arise as a consequence of grammaticalization processes. Yet, it also means that neither the Fwe nor the Jalkunan marker is an affix as typically understood and that they subsequently undermine the cover term “noun affix” employed in this work. Note in this context, though, that Greenberg’s “Stage 3 article” would be less than ideal as an alternative because, as Greenberg (1981: 106) points out himself, items at Stage 3 are no longer articles at all.

The other morphosyntactic parameter deals with a paradigmatic issue and was not addressed by Greenberg (1978). As shown in the case studies, there is a difference between those genera that have exactly one noun affix and those that have more than one. Interestingly, the genera that have more than one (Western Otomanguean, Numic) also have more than two and may have as many as six. This difference may again be accounted for with reference to their diachrony. First and foremost, Dryer (2013) shows that less than half of all languages in his large, worldwide sample have an indefinite article. Hence, noun affixes that derive from definiteness markers are fairly unlikely to have paradigmatic competitors because their ancestors often did not have any themselves.

In light of this, the larger noun affix paradigms are more likely to derive from gender systems. In Corbett’s (2013) sample, more than 43 % of languages have at least two gender markers, and almost a quarter of all languages have three or more. Since both Numic and Western Otomanguean fall into the range of three to six noun affixes, a gender system would be a perfectly plausible origin of their markers as far as paradigm size is concerned. However, this account would again necessitate the assumption that the border between gender markers and noun affixes is not rigid, pace Greenberg (1978). In any case, if the origin in a gender marker paradigm is correct, Numic and Western Otomanguean suggest that grammaticalizing items can retain (close to) their full paradigmatic strength while steadily losing semantic content. In conjunction with the syntagmatic aspects discussed for Fwe and Jalkunan above, there is thus reason to assume that any formal property may survive during the process of semantic reduction. Whether formal conservativeness manifests itself in terms of syntagmatic freedom or paradigm size, etc., might then simply come down to whether the language had a relevant paradigm to begin with.

The remaining two parameters resemble free variation in that they also concern traits that noun affixes most likely take on after coming into existence as such. In fact, the two criteria of reuse for derivational purposes and for the creation of minimal words are related. In both cases, language users take advantage of structures that are already present but that can be refashioned virtually without restriction because there is little semantic residue that might shape their future usage. If noun affixes are commonly repurposed to derive nouns, though, this would nevertheless be unsurprising. As shown in the case studies, noun affixes are usually restricted to nouns and often occur with a vast number of them. Hence, they are a salient aspect of the nominal structure, which lends itself to an interpretation on which they identify and create nouns (at least in the absence of any other plausible function). However, it should be said that this account works better for Numic than for Western Otomanguean since the noun affixes in the latter seem to derive from derivational items rather than the other way around.

If language users are generally intolerant of forms that do not perform a function, the use of noun affixes as phonological props, as posited for Numic and Fwe, is clearly motivated. In fact, the phonological use of noun affixes might be considered a maximally economical and efficient strategy to deal with linguistic leftovers. As such, it is a kind of “phonogenesis” (cf. Hopper 1990: 153–154, 1994: 32) and an alternative to the common strategy of using epenthetic vowels to reach minimal size (Gordon 2016: 265). That this kind of reuse is not in evidence more often might simply be because a minimum word size is not a relevant factor in most languages (Gordon 2016: 264; cf. also Bickel et al. 2009).

As suggested in Section 3.2, the use of semantically empty markers in novel functions is of considerable relevance to the notion of grammaticalization. This is because such a development appears to be a counterexample to unidirectionality, i.e., to the common idea that grammaticalization only ever proceeds toward loss and never toward gain. As shown by Norde (2009: 67–72), though, desemanticization is not irreversible, and so the reuse of noun affixes for grammatical functions is not entirely unexpected. (In fact, Greenberg 1991 discusses this phenomenon under the name of “regrammaticalization.”) On the other hand, since grammaticalization theory classifies changes on the semantic dimension as primary (cf. Section 2), the mirror image of semantic reduction should also be considered more relevant than the mirror image of formal reduction.^[9] Thus, reused noun affixes are indeed major counterexamples to the assumption of unidirectional semantic reduction. One reason that such developments get overlooked might be that very few elements ever become semantically empty to begin with, and presumably even fewer are resemanticized thereafter. Hence, the reuse of noun affixes does not contradict Hopper and Traugott’s (2003: 138) larger claim that counterexamples to unidirectionality are “sporadic.”

In sum, I have suggested that the seven parameters of Table 1 divide into two classes. One class consists of properties that the noun affixes most likely retained from their ancestors. This subsumes an affiliation with certain semantic classes, occurrence on words other than nouns, omission in certain specified contexts, and a paradigmatic structure. The second class consists of properties that noun affixes most likely develop themselves, all of which indicate that they are in the process of losing their status as noun affixes. This class comprises free variation as well as use for derivational and phonological purposes. Based on this dichotomy, it is possible to get at least a rough idea of which noun affixes are the most and least grammaticalized. Western Otomanguean, Numic, and Fwe each satisfy two of the three innovative parameters and thus seem to have rather advanced noun affixes. However, the empirical situation in Western Otomanguean is somewhat unclear, and Fwe also meets two of the retention parameters. Hence, both might be in an intermediate position overall. By contrast, Tacanan, Central Mande, and Northern Iroquoian each show more retention than innovative features and hence appear to be less advanced than at least Numic. Clearly, though, this is too coarse a method to arrive at any definitive conclusions, and further research on these features and their diachronic interpretation is necessary.

5 Conclusion

This work has dealt with a type of empty morph that I called “noun affixes.” The examples came from four American and two African genera. In conjunction with the diachronic work on noun affixes by Greenberg (1978, 1981, 1991), it was argued that these items do not have functions to match their forms. This was taken to mean that semantic reduction during the process of grammaticalization can be completed before formal reduction. This, in turn, is an insight that escapes many contemporary accounts, which tend to focus on the beginning of the process, not on its end. All in all, seven properties along which the noun affixes differ more or less systematically were found. Some of these properties are likely inherited from the structures that gave rise to the noun affixes, while others are probably innovations that indicate future applications. Also, it was concluded that the reuse of noun affixes for the expression of novel functions undermines the expectations of grammaticalization theory and constitutes an interesting counterexample to unidirectionality. On the whole, the analysis offers a view of how highly grammaticalized items behave within linguistic systems that they might be expected to drop out of.

The phenomenon of noun affixes raises a number of additional questions. For instance, the data analyzed here implicitly suggest that noun affixes are not found in Indo-European languages. However, it is notable that (in)definiteness markers end up as purely phonological material in the nouns of those languages with remarkable frequency. The change from ewt to newt and from ekename to nickname via absorption of the /n/ of the preceding indefinite article an is a staple of historical linguistics textbooks (e.g., Trask 1996: 103). Similar processes have also played out across French creoles (Syea 2017: 23–25) and with the definite article of Arabic loans in Spanish (Kaiser 2014: 237). On the one hand, it is worth debating if the resulting elements in French creoles, English, and Spanish should actually be considered noun affixes. On the other hand, a positive answer to this question would suggest that noun affixes can also come about by other diachronic processes than those considered in Greenberg (1978).

On the topic of creoles, it should also be noted that a possible noun affix is discussed for Saramaccan Creole (srm; Indo-European) by McWhorter and Good (2012: 68). Their discussion revolves around many of the analytical issues also brought up above. Noun affixes would be unexpected in creoles in that these items were assumed here (and by Greenberg 1978) to develop over long periods of time, which should rule out their presence in relatively recent creoles. An alternative explanation might be that noun affixes in creoles were adopted from the lexifier(s) or, more likely, the substrate(s). That, however, would still undermine the idea that creoles only tend to express rather transparent functions morphologically. All of this simply goes to show that noun affixes (and creoles) need to be studied in more detail.

The database discussed here not only lacks Indo-European languages but is generally biased toward the Americas and Africa. However, the creation of an exhaustive list of all likely or possible noun affixes was not one of the aims of this work and will have to be left for further research. As it stands, the remaining macro-areas all seem likely to have noun affixes. For instance, Dryer’s (1989: 83–84) discussion of Austronesian “articles” touches on certain similarities with noun affixes. Meanwhile, the overview of classification devices in Sands (1995) suggests that Stage 3 noun affixes also exist in Australia. It was, of course, the Australian language Kunjen (kjn; Pama-Nyungan) that, according to Dixon (1970), created a classifier system of initial consonants to eliminate vowel-initial nouns. This scenario clearly differs from the phonological uses described in the above case studies. On the one hand, the phonological purpose in Kunjen is segmental rather than suprasegmental, and on the other hand, the noun affixes (if they can be called that) apparently came into existence in response to a phonological need and were not co-opted from pre-existing material. Depending on the analysis and behavior of those markers, however, they might constitute yet another pathway for the emergence of noun affixes.

A further point mentioned by Greenberg (1978: 49, 56, 70) is that noun affixes are commonly replaced by new noun affixes that appear on the opposite side of the stem. That is, noun prefixes are replaced by noun suffixes and vice versa. In fact, Northern Iroquoian, which Greenberg (1978) did not discuss, seems to be a prime example of this since the semantically empty suffixes co-exist with only slightly more substantive prefixes. The same pattern was also established for the Western Otomanguean data, though the nature of the suffixes is more dubious in that case (cf. Section 3.5). It is not obvious why such a symmetrical effect should emerge, and to the extent that the Bantu noun class prefixes are renewals of the augment in one way or another (cf. Greenberg 1978: 57), they would present clear counterevidence in any case. Here, too, further research will have to establish whether this tendency is at all cross-linguistically robust.

This paper began by couching the investigation of noun affixes in the general debate of word-based and morph-based approaches to morphological analysis. As suggested there, empty morphs argue against agglutinating, morph-based accounts. More broadly, it might be held that morph-based approaches are appropriate for the majority of morphological phenomena across many languages and that word-based approaches hold greater appeal when it comes to the various deviations from agglutination. While noun affixes are one such deviation, they, as well as other exceptions, are also rather infrequent. Hence, instead of making the case for word-based approaches, noun affixes actually illustrate that both alternatives are required, if for different purposes.