Within lexicalist approaches to morphosyntax, the question whether to treat passivisation as an inflectional or a derivational process has often been raised. The traditional treatment of passive considers it to be an inflectional process. This traditional view is followed by certain number of recent studies (Stump, 2006; Hippisley, 2007) that presuppose the existence of one single paradigm for both the active and the passive verb forms of a given lexeme. On the other hand, (Sadler and Spencer, 1998) claim a derivational analysis of passives while (Bresnan, 1982) states that passives may serve as an input for derivational processes. If one accepts that inflected forms cannot function as an input for derivational processes, (Bresnan, 1982) inevitably entails a derivational treatment of passivisation. (Blevins, 2003) proposes a treat- ment for passives as a morphological process called Passive Lexical Rule (PLR) applicable to a whole series of languages. (Sag et al., 2003) argue that the argument-structure-changing operation triggered by passivisation are evidence for a derivational analysis for passives.

Others, like (Kiparsky, 2005) do not explicitly choose between the two approaches. Yet the data Kiparsky presents for Latin rather favours a derivational analysis. Usually, deriva- tion is said to operate semantic change in an unpredictable way: as opposed to inflectional processes, given one derivational process, there is not necessarily one unique predictable se- mantic change that systematically occurs. Based on data from Latin, (Kiparsky, 2005) shows that morphological passive forms may have several distinct values, not only syntactic passive value. These values are lexically specified, i.e. not directly predictable.

In this paper we analyse the morphological passive of the Western Iranian language So- rani Kurdish. We use the more common name Sorani to refer to its standardised dialect, corresponding to what (Haig, 2010) refers to as Suleimani. Sorani mainly distinguishes it- self through its verbal morphology and its intricate system of "endoclitic" person markers (Samvelian, 2007). Sorani verb forms roughly consist of a set of prefixes and suffixes clus- tered around a given stem (cf. Table 2). Most traditional descriptions of Sorani morphology concur in stating the existence of two distinct verbal stems: one for the present tense forms and one for the past and non-finite forms. Clustered around these stems, prefixes mostly con- vey tense, aspect and mood (TAM) and polarity features, whereas suffixes may encode TAM and person information. Sorani also displays three sets of personal endings (Table 1): PE1 for the present verb and the perfect subjunctive , PE2 for imperfective, preterite and past perfect forms and PE3 for the remaining perfect forms. In past tenses, Sorani verbs display remnant features of split ergativity (Haig, 2010): if a verb is transitive, the normal person markers are replaced with the endoclitic person markers as subject-verb agreement markers in the past tenses. They are inserted in verb-internal second position (Samvelian, 2007).

Passives are formed by inserting the sequence -rê/-râ between the stem and the other suffixes (shaded column in Table 2). Descriptive grammars implicitly present -rê/-râ as just another set of inflectional suffixes for transitive verbs (McCarus, 1958; MacKenzie, 1961; Blau, 2000; Thackston, 2006).

In order to get a more precise view of Sorani passive, we have developed a full Paradigm Function Morphology (PFM, (Stump, 2001)) account for both possible analyses of Sorani passivisation -- inflection or derivation. We argue that passivisation in Sorani Kurdish is a derivational process and propose four arguments supporting this claim. Instead of inflectional Voice Realisation Rules, we propose a Lexical Passive Derivation Rule.