Subject Clitics: New Evidence from Old Nubian

This article treats a set of subject cross-referencing morphemes in the medieval Nilo-Saharan language Old Nubian, traditionally called “personal endings.” Based on an analysis of their syntactic distribution and morphology, I argue that this set can be best described as a set of subject clitics, originally deriving from phonologically reduced pronominals. This set of subject clitics interacts with both topic and focus makers in the clause. Finally, by inspecting the historical development of Old Nubian subject clitics into full-fledged agreement suffixes modern Nile Nubian languages Nobiin and Mattokki (Kenzi) I argue that a purely syntactical approach to this development is impracticable, but that any morpho-phonological approach should be able to account for the diachronic data.


The distribution of Old Nubian "personal endings"
Old Nubian features a series of morphemes that appear on verbal forms after tense morphology, but before the predicate marker -ⲁ -a (Table 1). 2 These morphemes are referred to in the literature as "personal endings" (Browne 2002: 49;Bechhaus-Gerst 2011: 72;Smagina 2017: 38).
I have argued, however, that the finiteness of a verb should not be correlated with the presence of one of these cross-referencing morphemes, but rather with the presence of predicate marker -a (van Gerven Oei 2015a: 317-322). 3 Thus, an explanation for the distribution of these "personal endings" should be sought elsewhere.
In order to determine the morphological and syntactical status of this set of morphemes, let us first investigate their precise distribution.
First, they are usually absent when the subject of the sentence is overt, unless it is marked by the topic marker -ⲉⲓⲟⲛ -eion. 4 Compare the following two phrases from a text known as The Miracle of Saint Mēnas: 5 (1) dbmnt 713.4.16 ⲁⲡⲟⲅⲅⲗ̄ ⲡⲉⲥⲁⲣⲁ· apogg-il pes-ar-a skipper-det say-pst1-pred 'The skipper said.' (2) dbmnt 713.5.14-15 ⲁⲡⲟⲅⲅⲗ̄ⲗⲟⲛ ⲡⲉⲥⲥⲛⲁ· apogg-il-lon pes-s-n-a skipper-det-top say-pst2-2/3sg-pred 'The skipper said.' Apart from the difference in preterite marker, which we will not address here, 6 the verbal forms in examples (1) and (2) differ in that the latter features a personal ending, whereas the former doesn't. In his analysis of this text, Browne (1994: 32) suggests this variation may be "because of similar variation in the Greek Vorlage," where pesara 'say-pst1-pred' would correspond to the Greek present tense legei and pessna 'say-pst2-2/3sg-pred' to Greek aorist eipen. Evidence from bilingual psalms, however, shows that this matching is not consistent throughout Nubian translations from Greek and should therefore be discarded. I will argue below that the decisive difference between (1) and (2) is the presence of the topic marker -eion in (2), which shows that the subject has moved leftward out of its original position. Second, "personal endings" are usually present when the subject is null.

Topicalization
In van Gerven Oei (2015a: 319) I argued that the distribution of the personal endings can be partially correlated with the presence of the topic marker -eion, as on (2) apoggillon 'sailor-det-top.' In the context of the current paper this should be reformulated as follows: personal endings can be correlated with an empty subject position, which is either the result of leftward movement (often topicalization) or pro-drop.
The topic marker -eion is a second-position clitic (cf. Spencer & Luís 2012: 48) and etymologically related to the conjunction ⲟⲛ on 'and.' It is traditionally translated as 'and' or 'but,' or is otherwise left untranslated. It marks the topic of the sentence and is a common occurrence in literary Old Nubian texts, although it remains virtually unattested in the documentary, non-literary evidence.
When we look at some of the contexts in which it is frequently employed, we may notice that its presence is often connected to scrambling and quantifier raising, allowing word order patterns that are not allowed in a regular Old Nubian SOV sentence.
As to the motivation of this type of leftward movement in literary texts, it often seems to occur in contexts in which a Nubian translator attempts to emulate the original Greek sentence order. 9 This was often difficult, because with the emergence of Hellenistic koinē Greek had transitioned from an SOV to an SVO language (Taylor 1994). In practice, this meant that Nubian translators needed to invent strategies to move the verb to the left of the object without being ungrammatical. The liberal employment of the topic marker -eion attached to a constituent (often the subject) moved to the left edge appears to have been one of such strategies.
We fortunately have a small number of fully bilingual Old Nubian-Greek texts, 10 which allow us to test this hypothesis. Consider the following Old Nubian translation of Ps. 61:11, mē elpizete ep' adikian kai epi harpagmata mē epipotheite 'Trust not in unrighteousness, and lust not after robberies.' injustice.nmlz-pl-pl-acc and sourt-in-gou-k-ende-eion j grasping-pl-pl-acc-conj-top eik-ij-men-na-sō k [ TP t j t k ] be.near-plact-neg-imp.2/3pl.pred-comm 'Do not hope for injustices nor be eager for graspings. ' (Ps. 61:11) Old Nubian imperative forms always move to the left of their objects, as in the first part of the translation, where the verb form teeirmenasō 'hope-neg-pred-comm' precedes its object okkdirtingouka 'injustice.nmlz-pl-pl-acc.' This word order naturally matches the Greek word order mē elpizete ep' adikian.
After the conjunction on 'and,' we find the inverse order imitating the Greek epi harpagmata mē epipotheite with the object sourtingoukendeeion 'grasping-pl-pl-acc-conj-top' preceding the imperative eikijmennasō 'be.near-plact-neg-imp.2/3pl-comm.' But as we know that the imperative verb form must have left its original position and moved to the left, this means that the object has moved even further leftward. This is confirmed by the presence of topic marker -eion. We therefore may assume a double movement: leftward movement of both imperative forms and topicalization of the object in the second clause.
The same topic marker -eion is frequently found together with the universal quantifier, which in Old Nubian often displays overt quantifier raising. 8 The precise distribution, meaning, and use of this morpheme are still unclear. There is however a strong correlation between its appearance and scrambled constituents. See Browne (1997: 28-37) for an overview of its distribution. 9 It is well established that preserving Greek word order was one of the more important aspects of the Bible translation. See Askeland (2012: 10-11 kipt-a miššan-non i kaskase-l-dō [ TP t i timm-is-an-a ] people-pred all-top baptistery-det-sup assemble-pst2-3pl-pred 'All the people assembled in the baptistery.' Both the scrambling data in (4) and (5) and the quantifier raising in (6) show that the topic marker -eion should be associated with leftward movement.

Null subjects
Whenever the subject is moved leftward out of the core clause and into a topic position, leaving a subject gap, we find "personal endings" on the verb.
say-pst2-2/3sg-pred 'And the skipper said.' baptistery-det-sup assemble-pst2-3pl-pred 'All the people assembled in the baptistery.' This analysis is also consistent with our data suggesting that such cross-referencing morphology is present whenever the subject is null in pro-drop contexts: -sin k 2sg-acc-top say-pred give.2/3-inten-prs-1sg.pred-emp [ TP pro i ŋokkor ai-dō auou-tak-o-k-ka t j t k ] miracle 1sg-sup do-pass-pst1-det-acc 'And I will tell you of the miracle that happened to me.' (7) dbmnt 713.1.11-2.2 ⲉⲓⳡⲧ̄ⲧⲟⲩ ⳟⲟⲩⲕⲧⲗⲱ ⲇⲓⲉⲓⲕⲟⲛ ⲕⲟⲁ̄ⲗⲟ ⲉⲛⲟⲛⲁ· ⲥⲉⲩⲁⲧⲧⲕⲟⲛ ⲕⲟⲛⲙⲉⲛⲛⲁⲗⲱ· eiñittou ŋoukt-lō [ TP pro i diei-k-on ko-a-lo wealth.nmlz glory.nmlz-loc be.many-acc-top have-pred-foc en-o-n i -a ] seu-att-k-on j be-pst1-2/3sg-pred inherit-nmlz-acc-top [ TP pro k t j kon-men-n k -a-lō ] have-neg-prs.2/3sg-pred-foc 'In wealth and glory she possessed much, but an heir she did not have.' (8) dbmnt 713.9.14-16 ⲁⲟⲩⲕⲕⲟⲛ ⲙⲁⲛ ⲇⲡ̄ⲡⲛ̄ ⲅⲁⲇⲇⲱ ⲕⲉⲛⲟ̄ⲟⲥⲥⲁⲛⲁ· aou-k-kon i man dipp-in gad-dō ship-acc-top dem.dist village-gen shore-sup [ TP pro j t i ken-oos-s-an j -a ] place-asp-pst2-3pl-pred 'And they beached the ship on the shore of that village.' Data from non-coreferential (non-subject) relative clauses (van Gerven Oei 2015b: 18-21) confirm the fact that null subjects are correlated with "personal endings." Note that overt subjects in Old Nubian non-coreferential relative clauses are marked with the genitive case, as in for example Altaic languages and Japanese (see, e.g., Krause 2001 dem.prox 3sg-gen eat-pst2-det suddenly fowl añ-r-aŋ-a live-det-inch-pred 'this egg that he had eaten suddenly coming to life as a fowl' The relative clause in (9) is non-coreferential, with the relativized object koumpou 'egg.' Its overt subject tanna '3sg-gen' is marked with the genitive case, and the participial form kipsil 'eat-pst2-det' shows no "personal ending." (10) dbmnt 1391.5.4-7 ⲉⲗⲟⲛ ⲙⲩⲥⲧⲏⲣⲟⲩ ⲉⲕ̄ⲕⲁ ⲉⲅⲓⲇⲣⲟⲩⲕⲁ ⲟⲩⲕⲁ ⲡⲗ̄ⲗⲓ̈ⲅⲣⲁ̄ ⲇⲉⳡⳝⲉⲥⲟ· el-on mustērou [ RC pro i eik-ka egid-r-ou i ]-ka ou-ka now-top mystery 2sg-acc ask-prs-1/2pl-acc 1pl.excl-acc pill-igr-a deñ-j-e-so shine-caus-pred give.1-plact-imp.2/3sg.pred-comm 'And now reveal to us the mystery which we ask you.' The non-coreferential RC in (10) has an implied subject 'we' and a subject clitic on the verb egidrou-'ask-prs-1/2pl.' The distribution in non-coreferential relative clauses (9) and (10) thus parallels the distribution in main clauses (1) and (7), in spite of the fact that the latter feature nominative subjects and the former genitive subjects. Fuß (2005: 130-39) provides a typological framework that allows us to interpret the data on Old Nubian "personal endings" presented above, offering several syntactic and morphological criteria (based on Zwicky & Pullum 1983) 11 to distinguish clitics from agreement markers. Not all syntactic and morphological criteria listed by Fuß are applicable or testable in Old Nubian, for lack of living speakers. I have only listed those that can be verified based on the extant material. Moreover, because agreement is completely absent in Old Nubian, it is difficult to test the morphological criteria language-internally. Comparative evidence with Nobiin and Mattokki, however, gives us some indication of their validity.

Subject clitics or agreement
(11) Syntactic criteria: • Clitics have complementary distribution with DP arguments; • Clitics don't show up in subject gap environments; • Clitic doubling usually only appears with definite DPs.
(12) Morphological criteria • Clitics exhibit a lower degree of host selectivity than agreement; • Clitics exhibit fewer morphological and semantic idiosyncracies than agreement; • Clitics exhibit fewer arbitrary gaps in their paradigm than agreement. (1) vs. (7) and (9) vs. (10), there appears to be in general a complementary distribution between overt subject DPs and the series of "personal endings." This is a first indication that we are in fact dealing with a series of subject clitics and not agreement. Compare this for example with subject clitic data from Zapotec (Broadwell 2005):

Complementary distribution with DP arguments As we have seen in examples
Zapotec (Broadwell 2005: ex. 1) Ù-dàw réé=bííny gèhèht compl-eat pl-person tortilla 'The people ate tortillas.' Zapotec (Broadwell 2005: ex. 2) Ù-dàw=réhby gèhèht compl-eat-3pl tortilla 'They ate tortillas.' (15) Zapotec (Broadwell 2005: ex. 3) *Ù-dàw=réhby réé=bííny gèhèht compl-eat-3pl pl-person tortilla 'The people ate tortillas.' Similarly, other languages with subject clitics exhibit clitic doubling behavior such as noticed in (2) and (6), where the subject has moved and left a trace. Compare, for example, data from Rumantsch (Anderson 2006), which allows clitic doubling in case of subject inversion. In Rumantsch, subject clitics appear when the subject is inverted or when it is absent. Regular sentence order: Rumantsch ( In these examples, subject clitics appear in conjunction with a null or moved subject. A similar situation can be observed in Old Nubian, where subject clitics appear in contexts where the subject is either null, or has moved to the left edge of the clause. Absence in subject gap environments Unlike agreement, subject clitics are absent in subject gap environments, for example in conjoined clauses such as (5) and the following example: (20) dbmnt 713.15.10-15 ⲥⲟⲣⲧⲟⲗⲗⲟⲛ ⲇⲟⲩⲙⲙⲓⳝⲁ· ⲧⲉⲇⲇⲱ ⲥⲓⲗⲉⲗⲁ· ⲧⲁⲇⲇⲉ ⲧⲁⲛ ⲟⲅⳝⲗ̄ⲇⲉ. ⲧⲁⲛ ⳟⲁⲇ⳿ⲇⲉ. ⲧⲁⲛ ⲙⲉⲇⳝⲟⲩⲛⲓⲅⲟⲩⲗⲇⲉⲕⲉⲕⲕⲁ ⲡⲥ̄ⲧⲁⲣⲓⲥⲛⲁ sorto-l-lon doumm-ij-a ted-dō silel-a tad-de tan priest-det-top take-plact-pred 3pl-sup pray-pred 3sg-conj 3sg.gen ogj-il-de tan ŋad-de tan husband-det-conj 3sg.gen son.det-conj 3sg.gen medjou-ni-gou-l-dekek-ka pist-ar-is-n-a servant-pl-pl-det-conj-acc splash-tr-pst2-2/3sg-pred 'And the priest took them, prayed for them, and baptized her, her husband, her son, and her servants.' In (20) only the final verb in the series of conjoined clauses pistarisna 'splash-tr-pst2-2/3sg-pred' is marked with a subject clitic, whereas the other verbs doummija 'takeplact-pred' and silela 'pray-pred' are not. If the series of "personal endings" would be a form of agreement, we would have expected these morphemes to appear on all verbs.
Clitic doubling only with definite DPs The clitic doubling examples (2) and (6), in which the subject had moved to the left edge and was marked with topic marker -eion, already showed that in both cases the subject was definite. In both these cases and in the case of pro-drop clauses, the clitic licenses pro or the trace.
There are a few examples in which we find clitic doubling in which it is not immediately apparent that the subject has moved to the left. All of these examples, however, involve a complex syntax with embedded clauses or overt movement, and it is difficult to reconstruct the precise position of the subject. Referring to Swiss Rhaeto-Romance dialects, Fuß (2005: 192) observes that "clitic doubling fulfils certain stylistic (or, rather, discourse) functions in which a full element is added to reinforce an enclitic pronoun for reasons of emphasis." ouel-lon pes-imm-a mijirk-il one-top say-aff.prs.3sg-pred disobedience-det di-a-r-ka-lo ŋek-n-a-a die-pst1-det-acc-foc produce-prs.2/3sg-pred-quot 'And again he says: "Disobedience produces death."' In (21) we find both a definite subject marked by the determiner -(i)l (van Gerven Oei 2011: 256-62) and a verb marked by a subject clitic. This suggests that mijirkil 'disobe-van Gerven Oei: Subject Clitics Art. 7, page 9 of 19 dience-det' has moved to the left of the quotation, leaving a trace. Because it is illicit to have a topic marker in a quotation environment, this move is invisible on the surface.
(22) dbmnt 687.113.5-6 ⲥⲗⲟ ⲡⲓⲛⲁ ⲓⲟⲩⲇⲁⲓⲟⲥⲅⲟⲩⲛ ⲟⲩⲣⲟⲩⲟⲩ ⲟⲩⲛⲛⲟⲩⲧⲁⲕⲟⲗ⳹ is-lo pi-n-a ioudaios-gou-n ourouou wh-loc exist-prs.2/3sg-pred Jew-pl-gen king ounn-outak-o-l bear-pass-pst1-det 'Where is the born king of the Jews?' In (22) we find a syntactically rare construction in which the main verb in a question has inverted without showing any special morphology (usually we find an affirmative suffix as in (23)). The constituent order follows exactly the Greek Vorlage and it may well be that underlyingly the subject has moved to a higher position in the clause. But again this is not visible.
(23) dbmnt 713.9.16-10.3 ⲕⲩⲣⲓⲁ̄ⲕⲉⲛ ⲟⲩⲕⲟⲩⲣ⳿ ⲉⲓⲛⲛ̄ⲛⲟⲛ· ⲁⲡⲟⲅⲅⲗ̄ ⲇⲡ̄ⲡⲇ̄ⲇⲱ ⲕⲉⲇⲁ ⲕⲓⲥⲛⲁ· kuriake-n oukour ein-in-non apogg-il dipp-id-dō Sunday-gen day be-prs.2/3sg-top sailor-det village-det-sup ked-a ki-s-n-a ascend-pred come-pst2-2/3sg-pred 'And as it was Sunday, the sailor went up to the village.' In (23) we find a highly uncommon construction in which an entire subordinate clause is marked by the topic marker. Again we would have to assume that the subject has moved out of its original position, with the subject clitic licensing the empty position.
Finally, we have also one example of clitic doubling within a non-coreferential relative clause, with a genitive-marked subject: (24) dbmnt 1391.6.4-7 ⲉⲗⲟⲛⲇⲉⲉⲓⲟⲛ ⲇⲉⲕⲕⲓⲅⲣ̄ⲙⲉⲛⲇⲣⲉ̄ⲗⲟ· ⳟⲉⲉⲓⲟⲩ ⲉⲓⲛ ⲟⲩⲛⲛⲁ ⲁⲓⲕⲁ ⲉⲕⲓⲇⲣⲟⲩⲕ elon-de-eion dekk-igir-men-d-r-e-lo ŋeeiou ein now-conj-top conceal-caus-neg-inten-prs-1sg.pred-foc thing dem.prox oun-na ai-ka ekid-r-ou-k 2pl-gen 1sg-acc ask-prs-1/2pl-acc 'And now I will not conceal the thing that you asked me.' Non-coreferential relative clauses do not allow for a topic position, and on the surface it appears as if the subject has not moved to a higher position. A possible explanation for the presence of the explicit subject ounna in the relative clause could be to disambiguate the verbal form.
More data are needed to explain the precise distribution of clitic doubling beyond the clear-cut topicalization cases. Fuß (2005: 132) indicates that clitic doubling constructions can easily be misinterpreted as agreement configurations and are therefore often a hinge in the historical development from clitics to agreement. As we will argue below, this is precisely what happened in the transition from Old Nubian to contemporary Nile Nubian languages.

Morphological criteria
Low degree of host selectivity (promiscuous attachment) Agreement affixes are usually more rigid in their host selectivity than clitics. Old Nubian subject clitics are generally only found on verbal roots or nouns to which a verbalizer has been attached. There is one attestation of a subject clitic that appears to attach to a question word, but the form, which is a hapax, may very well have a different analysis: (25) dbmnt 1385.1.7-9 ⳟⲁⲉⲓⲁ̄ ⲟⲩⲕ ⲟⲕⲓⳝⲁⲣⲣⲉ· ⲧⲓⲕⲁⲛⲉⲅⲟⲩⲉ̄ⲙⲁ ⲥⲉⲛⲁ· ⲙⲉⲛⲉⲛⲛ[ⲟⲛ] ⲕⲟⲣⲟⲥⲉⲅⲟⲩⲉ̄ⲙⲁ· ŋaei-a ou-k ok-ij-ar-r-e tikan-e-goue-ma who-pred 2pl-acc call-plact-inten-prs-1sg.pred sheep-pl.pred-pl.pred-aff s-en-a menennon koros-e-goue-ma wh-prs.2/3sg-pred and.not shepherd-pl.pred-pl.pred-aff 'What shall I call you, are you sheep and not shepherds?'

Morphological and semantic idiosyncracies Old Nubian subject clitics have no morphological effect on their host words and show no semantic idiosyncracies.
Arbitrary gaps Old Nubian subject clitics used in regular declarative sentences show no arbitrary gaps. However, on verb forms featuring the so-called affirmative suffix (used in emphatic contexts) there may be a gap with 2sg and 2pl verb forms. For first and third person affirmative verb forms we find the regular subject clitics (with assimilation to the affirmative suffix): (26) dbmnt 713.6.8-10 ⲁⲗⲉⲥⲛ̄ ⲟⲩⲛⲣⲉ ⲉⲛⲉⲛⲛⲟⲛ· ⲭⲣⲓⲥⲧⲓⲁ̄ⲛⲟⲥⲁⳟⲁⲇⲙ̄ⲙⲉ· ale-sin oun-r-e en-en-non true-emp bear-prs-1sg.pred be-prs.2/3sg-top khristianos-aŋ-ad-imm-e Christian-inch-inten-aff.prs-1sg.pred 'And if it be that I give birth, I will become a Christian.' (27) dbmnt 1004.i.11-12 ⲁⲓ̈ⲁⲅⲗ̄ⲗⲉ ⲡⲥ̄ⲉⲗ̄ ⲇⲓⲉⲛⲕⲉⲧⲁⲗ· ⲁⳡⲁⲇⲙ̄ⲙⲁ· ai-a-gil-le pis-eil di-en-ketal añ-ad-imm-a 1sg-dat-dir-int believe-det die-prs.2/3sg-also live-inten-aff.prs.3sg-pred 'Even if someone who believes in me dies, he will live.' For second persons, however, we find suppletive forms without a subject clitic from the same series, ending in -ⲗⲏ -lē for 2sg and -ⲗⲕⲉ -lke for 2pl. The origin of these suppletive forms is unknown, but 2pl -ⲕⲉ -ke also appears on certain vetitive forms and in vocative contexts.

Focus
Old Nubian exhibits effects of what Siewierska (2004: 159) calls "focus hierarchy," the restriction of cross-referencing to non-focus-marked constituents. She provides several examples, including from the Chalcatongo Mixtec language (Macaulay 1996). Chalcatongo Mixtec is an VSO language, where a focus-marked subject precedes the verb. Topics move even further to the left.

(31)
Chalcatongo Mixtec (Macaulay 1996: ex. 203b) ñãʔã wãã xĩ́nũ=ñá woman the run=3f 'The woman is running.' (30) shows a full subject DP in preverbal focus position and a verb without subject clitic. In (31), the subject is the topic of the sentence, and therefore a subject clitic is allowed.
We find a similar distribution in Old Nubian. Although clitic doubling has been attested with definite DPs, it is absent when the DP in question is marked with the focus marker -ⲗⲟ/-ⲗⲱ -lo/-lō. 12 joun-jour-t-lo sal-lō dou-ar-a sal-l-on go∼int-nmlz-loc word-foc exist-pst1-pred word-det-top till-ila-lo dou-o-n-a God-dat-foc exist-pst1-2/3sg-pred 'In the beginning was the Word, and the Word was with God.' (Jn 1:1) In the first clause of (32), the subject sallō 'word-foc' is marked with the focus marker -lō, while the main verb douara 'exist-pst1-pred' remains without subject clitic. In the second clause, sallon has become the topic, leaving a subject gap, while the focus has shifted to tillilalo 'God-dat-foc.' As a result, we find a subject clitic on douona 'exist-pst1-2/3sg-pred.' This example shows clearly how topic and focus marking and subject clitics interact.
In general, it appears that the presence of the focus marker -lo is correlated with the absence of movement: Thus the behavior of the Old Nubian focus marker complements our other data. Focusmarked subjects do not move leftward and are therefore correlated with the absence of subject clitics on the verb. Fuß (2005: 2) remarks that "it is a long-standing observation in historical linguistics that verbal agreement morphology develops from (originally independent) personal pronouns" and that clitics play a key role in the transition from pronoun to regular agreement. Based on previous literature he proposes the following grammaticalization scheme:
(37) Old Nubian subject clitics deriving from demonstrative pronouns • ⲉⲓⲛ ein 'this' > 2/3sg clitic -n; • ⲙⲁⲛ man 'that' > 3pl clitic -an. 13 The different origins of first/second person and third person clitics, personal and demonstrative pronouns respectively, is not uncommon. Oswald Szemerényi argues that Proto-Indo-European person suffixes developed along the same lines (see Fuß 2005: 18n4). It is unclear at which stage in the history of Old Nubian the development from weak pronoun to clitic pronoun took place, but it appears that the incorporation of pronominal elements in the verbal complex was not restricted to cross-referencing with the subject. For example, Smagina (2017: §46) argues that the passive marker -tak, derives from an incorporated 3sg.acc pronoun tak(ka).
The impetus for the reinterpretation of these subject clitics as agreement morphology in the Nile Nubian languages Nobiin and the closely related language varieties Mattokki (Kenzi) and Andaandi (Dongolawi) may have very well been the clitic doubling constructions with specific DPs discussed above. In Nobiin, the subject clitics have fused with tense morphology, and have become a stable agreement paradigm that is obligatory for any main verb (Bechhaus-Gerst 2011: 75-84). 14 Mattokki, on the other hand, seems to have preserved the Old Nubian system, with only little change (Table 2). 15 Fuß (2005: 233) claims that "[n]ew verbal agreement morphology arises only for those slots of the agreement paradigm where the existing verbal inflection is non-distinctive." The data from Nobiin confirm this. Whereas the Old Nubian subject clitic paradigm had several syncretic forms (2/3sg and 1/2pl), Nobiin features innovative 2sg and 2pl forms. As may be observed in Table 2, the 2sg and 2pl agreement suffixes differ considerably from the Old Nubian tense + subject clitics, creating new, distinctive forms. The -k in the 2pl form can be traced back to the Old Nubian 2pl clitic -ⲕⲉ -ke (see (29)). 16 (38) dbmnt 714.21.15-22.1 ⲙⲛⲁ ⲟⲩⲗⲅⲣⲓⲙⲉⲛⲁⲕⲉ ⲡⲁⲣⲧⲁⲕⲟⲕⲕⲁ· mn-a oulg-r-imen-a-ke par-tak-o-k-ka wh-pred ear-tr-neg-pred-2pl write-pass-pst1-det-acc 'Don't you hear what is written?' Although the appearance of the clitic -ke is uncommon in Old Nubian texts, its nascent development should most probably be correlated with the grammaticalization process described above, with its usage in Nobiin becoming generalized up to the point that it became part of the 2pl agreement suffix. As for the second part of the innovative Nobiin agreement suffixes, -(a)m, we see a reflex of the Old Nubian finite predicate marker -a and the affirmative marker -ⲙ(ⲁ) -m(a).

Old Nubian Nobiin Mattokki
van Gerven Oei: Subject Clitics Art. 7, page 14 of 19 (39) dbmnt 714.20.3-6 ⲧⲁⲛ ⲁⲉⲗ̄ⲗⲁ ⲡⲉⲥⲉⲛ ⲓⲉⲣⲉⲟ̄ⲥⲓⲗⲟⲧⳝⲱ ⲕⲙ̄ⲙⲥ̄ⲥⲟⲙ ⳟⲉⲓⲙⲥ̄ⲥⲟⲙⲁ· tan aeil-la pes-en iereos-ilo-tjō 3sg.gen heart-dat say-prs.2/3sg priest-loc-with kim-mis-s-o-m ŋei-mis-s-o-ma eat-neg-pst2-1/2pl.pred-aff drink-neg-pst2-1/2pl.pred-aff 'If he says in his heart that we did not eat or drink together with the priest.' Again this morpheme rarely appears in Old Nubian in this form, but as is clear from (25), it appears in contexts where disambiguation may have been necessary, which led to a grammaticalization of the form in both 2sg and 2pl agreement markers in Nobiin. 17 We may therefore posit the following historical development: Innovative agreement markers in Nobiin • -nam < *n-a-m(a) • -rokom < *ro-ke-a-m(a) As is clear from Table 2, Mattokki appears to be more conservative than Nobiin in its preservation of the neutral tense paradigm. The same appears to be true of the past tense, where Mattokki shows a preservation of the -s from the Old Nubian second preterite, whereas Nobiin has a syncretic paradigm combining forms from the Old Nubian first and second preterite (Bechhaus-Gerst 2011: 82-83). 18 The development of agreement in contemporary Nile Nubian languages has led to the complete collapse of the Old Nubian discourse marking morphology. Both the topic marker -eion and focus marker -lo have disappeared without any trace. Considering the interaction and interdependence of topic and focus marking and the series of subject clitics that we described above, this may perhaps not be surprising, but begs the question about how the distribution of these morphemes in Old Nubian, and their eventual disappearance, can be modeled. Fuß's approach only allows us to explain the development from subject clitic to agreement, but lacks an explanation for the larger collapse of the Old Nubian discourse marking system. At the same time, it appears that Nobiin and Andaandi/Mattokki have developed a new series of clitics based on several Old Nubian auxiliary verbs. The first are the future tense clitics fa(a)-(Nobiin) and b(i)-(Andaandi/Mattokki) from the Old Nubian verb ⲡⲁⲗ/ⲡⲉⲗ pal/pel 'to come out': Mongolian, allows for subject pronouns to appear in post-verbal positions, which would allow for development of subject clitics and later subject agreement in Buryat. Although we have no evidence of earlier stages of the Old Nubian language with full-fledged pronouns following verbal forms, it is evident that its series of subject clitics developed from pronominal elements.
He proposes the following syntactical reanalysis (Fuß 2005: 213): Aside from the fact that we have no evidence of such a complicated syntactical movement in Old Nubian, it seems to conflict with our scrambling data, which suggest that scrambling is only possible with a topicalized subject, as in (5). If indeed the entire TP moves to SpecCP in order for the verb to adjacent to the subject clitic adjoined to C 0 , we lose the structural motivation for this correlation between scrambling and topicalization. Moreover, such an approach encounters difficulties when there appears to be cliticized material on the right of the subject clitic, as in the case of Old Nubian. According to Zwicky & Pullum (1983: 504), only clitics can attach to material that already contains clitics. In the case of Old Nubian, this implies that both the predicate marker -a, as well as a number of discourse markers, including the focus marker ( §4), should be analyzed as clitics as well.
Indeed, the predicate marker shows a low degree of host selectivity, marking whichever is the main predicate of the clause, whether nominal or verbal (van Gerven Oei 2015a). The following example shows this quite dramatically: tōek-katt-a alesk-att-a tan ŋakter-k-on true-adj-pred power-adj-pred be.patient-adj-pred 3sg.gen anger-acc-top oukou-n dourtal-lō os-a der-r-a min-n-a-lō day-gen each-loc take.out-pred apply-neut-pred not.be-prs.2/3sg-pred-foc 'As David, bearing witness, says, God is a judge, truthful, powerful, patient. And he does not apply his anger each day.' In (48) we find the predicate marker attached to nominal predicates such as amiskada, 'judge-pred' and tōekkatta 'power-adj-pred,' but also the verbal predicate osa derra minnalō 'take.out-pred apply-neut-pred not.be-prs.2/3sg-pred-foc.' It also present on the adjunctive VP mataraŋa 'witness-inch-pred.' Additionally, the predicate marker appears within the scope of universal quantifiers, as in (6) kipta miššannon 'people-pred all-top,' and functions as a vocative marker.
To assume that the predicate marker and the focus marker, as well as a number of other discourse markers, are structurally autonomous, which Fuß's approach seems to imply, would lead to an explosion of the number of projections and movement involved to allow for adjacencies of the type proposed in (47). Moreover, we would encounter issues with the ordering of specific clitics, as for example in (26) where the subject clitic precedes the predicate marker and (29) where subject clitic appears to follow it.

Conclusion
In this paper, I have argued that the Old Nubian "personal endings" can best be analyzed as subject clitics, based on their syntactical distribution and morphology. They interact with other discourse markers in the sentence. Topicalization of the subject, marked by -eion, is correlated with the presence of a subject clitic on the main verb in order to license the subject trace. Conversely, focus marking of the subject by -lo signals that the subject has stayed in situ and is therefore correlated with the absence of a subject clitic. Subject clitics also appear in pro-drop constructions, and definite subjects that are neither topicalized nor focus-marked may be doubled by a subject clitic.
Although overall the diachronic syntactical model developed by Fuß (2005) is helpful to think through the transition from the Old Nubian subject clitic system to the agreement system of contemporary Nile Nubian languages, his purely syntactic approach clashes with several aspects of the Old Nubian clitic system, perhaps most problematically the distribution of the predicate marker -a. Conversely, a morpho-phonological approach such as proposed by Anderson (2005) may allow us to properly model the behavior of Old Nubian clitics, but lacks explanatory power on a diachronic scale.
In their broad overview of the different theoretical approaches to clitics, Spencer & Luís (2012: 231-32) suggest that a morphological approach à la Anderson may be more productive than a purely syntactic approach, "[w]hat seems clear, though, is that grammatical theory has to have some way of accommodating the fact that some clitics are exponents just of morphosyntactic properties while other clitics are exponents of representations with more semantic content, possibly even full-blown semantic predicates." To this important observation we should add that such a grammatical theory should come up with a convincing approach toward diachronic data as presented in this paper. If indeed a fully syntactic approach à la Fuß is too cumbersome, a morphological approach should be able to account for the forms of reinterpretation discussed above. Moreover, on a more general level, we need to continue to seek a complete account in which the reanalysis of subject clitics as agreement has been accompanied by the complete collapse of a rich discourse marking system.