The clitic status of person markers in Sorani Kurdish

The ambiguous nature of clitics challenges simple distinctions between morphology and syntax. Clitic properties also feature in theoretical accounts of ergativity. In this paper we analyze the nature of the person markers in Sorani Kurdish. The direct person markers are conventionally treated as verb affixes, while the oblique person markers are assumed to be clitics. A variety of clitic tests show that both the direct and oblique person markers have some affix properties as well as some clitic properties. The results illustrate the difficulty that language-specific features pose for a general theory of clitics.


Introduction
The ambiguous nature of clitics challenges simple distinctions between morphology and syntax (Spencer & Luís, 2012). In this paper we analyze the nature of person markers in Sorani Kurdish. Sorani (Central Kurdish) is a dialect of the Kurdish language spoken in northern Iraq and northwest Iran. Kurdish, with Balochi and Taleshi, constitute the West Iranian branch of the Iranian languages (Haig, 2017). The Iranian language branch separated from the Indo-Aryan languages some 4,000 years ago. The Iranian languages share many typological features including an OV word order, the frequent use of complex predicates based on light verbs and a tense-based alignment split that derives from transitive verb participles (Haig, 2017, p. 465).
The Iranian languages inherited an aspect-based contrast on transitive verbs from Old Iranian. Old Iranian had an accusative alignment system in which all subjects received direct case marking. The Modern Iranian languages, including Kurdish, still preserve this alignment system on intransitive verbs as well as on imperfective transitive verbs to some degree. Atlamaz and Baker (in press)  The Kurdish languages display many variations of the original Middle Iranian tensed-based alignment system. In this paper, we analyze the variety of Sorani spoken in Sulaimnyha in the north of Iraq, and Sinnha in the western part of Iran. The Sorani variety of Kurdish lacks overt case marking on nouns (Haig, 2017, p. 481). It relies solely upon person markers to cross-reference the subject and object in a clause.
The person markers license pro-drop for both the subject and object. Intransitive verbs use a reflex of the Middle Iranian direct suffix to cross-reference its subject in all tenses. We follow the convention of showing the direct cross-reference forms as unmarked as shown in (3). Transitive verbs in Sorani use the direct suffix to cross-reference its subject in the imperfective aspect and its object in the perfective aspect. It uses an oblique form to cross-reference the object in the imperfective aspect and the subject in the perfective aspect (4). Comparing the agreement markers in (3) and (4) shows that Sorani has an accusative alignment system in the imperfective aspect, but an ergative alignment in the perfective aspect. 'I saw you.' Table 1 presents the forms of the independent pronouns, oblique and direct person markers in Sorani. The oblique person markers descend from the pronouns in Middle Iranian and still resemble the Sorani pronouns. The form of the third person oblique allomorphs is conditioned by the preceding sound. The /=i/ follows consonants and the /=y/ follows vowels. The form of the first person direct marker is also conditioned by the preceding sound. The /-m/ follows vowels and the /-im/ follows consonants. Although the direct and oblique markers have overlapping forms, they can easily be distinguished in specific contexts.

Person Pronoun
Oblique Direct Table 1. Sorani person markers.
The oblique person markers in Sorani have many unusual features. The Iranian literature assumes that the direct person markers are affixes and the oblique person markers are clitics. We follow the usual convention of using the equal sign in examples to indicate clitic attachment and a hyphen for affix attachment. The exact nature of the oblique marker has generated considerable debate within the literature on the Sorani dialect. Amin (2012) analyzes this particle as a suffix and claims that it moves from the end of the verb to the end of the object depending on verb tense. Fattah (1997) and Karimi (2010) assume that it is a clitic, while Haig (2017, p. 481) describes it as a mobile pronominal clitic. Haig cites Samvelian's (2007) claim that "despite the evidently clitic nature of the marker itself, functionally, it is an agreement marker" (2017, p. 482). Others, such as Jukil (2015), claim that it falls under the case system and marks ergative case.
Accounts of ergativity for the Indo-Iranian languages focus on the tense-based subject marking on transitive verbs (c.f. Atlamaz and Baker, in press). The Sorani variety of Kurdish differs from the other Indo-Iranian languages in that it does not mark nominal arguments for case. In the absence of case marking, Sorani extended the use of the oblique form to cross-reference the argument that is not cross-referenced by the direct agreement suffix. The grammaticalization of the oblique forms in Sorani has important implications for theories that have been proposed to account for the ergative features in Indo-Iranian languages (Butt, 2017). The use of the oblique marker to cross-reference objects of transitive verbs violates the basic assumptions of ergative theories. Haig (2008: 305) claims that there is no principled way to establish the ergative nature of the oblique The examples in (4) demonstrate a few of the clitic properties of the oblique person markers in Sorani.
The oblique marker attaches to the imperfective aspect marker in (4a) and to the verb root in (4b). Its placement in (4b) is highly unusual since it precedes the direct person affix, which suggests that the affix may also be a clitic. The example in (5) shows another puzzling feature of Sorani person marking. In the case of a third person agent acting on a first or second person object the perfective transitive verb in the interrogative mood has two oblique clitics. In (5), the oblique subject marker follows the oblique object marker. Table 3 shows the full person marking paradigm for transitive verbs in Sorani.  In this paper we review several tests for clitichood and use them to investigate the nature of the Sorani person markers. The interaction of the Sorani person markers with verb transitivity and tense, as well as the movement of the oblique person forms to direct objects and negation provide the challenges for our analysis.

Properties of Sorani oblique markers
The Sorani oblique markers can double the independent pronouns (6). The pronouns are optional in Sorani, whereas the oblique clitics are obligatory in transitive clauses. The obligatory nature of the Sorani clitics contradicts the optional status of many clitics (Kramer, 2014;Samvelian, 2007 The oblique pronominal clitics are hosted by the first available constituent of the verb phrase (Öpengin, 2013;Gharib, 2012). The movement of the clitics to these hosts is the primary evidence for their clitic status. These possibilities are shown in (7-10). In the absence of any other host in the verb phrase, the clitic attaches to the verb root, as in (7). The imperfective aspect marker (8), the negation marker (9), and the direct object (10) can all host the oblique person clitic. Because clitics depend upon a host for their realization Kayne (1975) claimed that clitics cannot be used in isolation. The Sorani oblique clitics cannot be used in isolation as a response to information questions, unlike the independent pronouns (11).
(11) a. ke hat? Min/*m. The examples in (4), repeated in (12) show that the Sorani oblique clitics change from crossreferencing the object to cross-referencing the subject with a change from imperfective to perfective.

Properties of Sorani direct agreement suffixes
Although there is widespread agreement that the direct agreement markers are verb affixes in Iranian languages, the Sorani agreement markers have many surprising properties. As the example in (15) shows, the direct agreement suffix follows the oblique subject clitic rather than preceding it. This order violates the expectation that affixes are more tightly attached to stems than clitics (Zwicky & Pullum, 1983).

Summary
We observed that Sorani Kurdish has two sets of person markers. One set has traditionally been described as a direct agreement marker and descends from the nominative agreement markers in Middle Iranian.
The other set of person markers has been described as an oblique pronominal clitic and descends from Middle Iranian clitics that were derived from independent pronouns. Applying various clitic tests to the Sorani person markers produces mixed results for both the direct and oblique sets of person markers, as shown in Table 4. The numbers in Table 4 refer back to the examples that we presented in the preceding sections.
The results show that both sets of person markers in Sorani have mixed clitic/affix properties. The direct suffixes have some clitic properties: optionality, mobility, specific reference and co-option. The oblique clitics, likewise, have some affix properties: obligatory, tense sensitivity, and non-specific reference. These mixed results add to the puzzle of precisely characterizing the distinction between agreement markers and pronominal clitics discussed in Spencer and Luís (2012). We conclude that tests of clitichood found in the literature have been influenced by features of the languages in which they have been used. In this situation, it is impossible to predict the constellation of results for clitic tests in each language. This outcome shows that linguists lack a general theory of clitics.

Test
Direct suffix (nominative) Oblique clitic (ergative) 1. optionality clitic (16-18) affix (6) 2. mobility clitic (4) clitic (6-10) 3. isolated use clitic (11) 4. tense sensitivity affix (12) affix (12) 5. specific reference clitic (19) affix (13) 6. multiplicity affix clitic (5) 7. co-option clitic (20) affix (14) In the absence of an understanding of the nature of the clitic/affix divide, theories about the nature of other linguistic features based on the clitic/affix distinction remain undefined. Notable among these theories is work on the nature of ergativity. Theoretical accounts of ergativity that rely upon a clitic/affix distinction (e.g. Woolford, 2000) need to be reexamined in light of their assumptions about the nature of clitics. Since clitics result from language-specific processes of grammaticalization, theories of ergativity that presuppose a unified clitic category are suspect. Our analysis of the clitic properties of Sorani person markers advances both the theoretical understanding of clitics as well as an understanding of ergativity in the Indo-Iranian languages.