From derivation to inflection: the case of the Turkish nominalizer (y)Iş

1 Introduction

In this article, I will argue that the synchronic plurifunctionality of the Turkish nominalizer -(y)Iş mirrors different stages of grammaticalization, with the attested starting point as a derivational suffix and the preliminary endpoint as an inflectional marker with functions including those of a bound complementizer, albeit in an early stage of development, as in its function as an inflectional marker the item does not seem to be obligatory, i.e., it can generally be exchanged with another, more established item. The complementizer function of the item under investigation parallels in some respects the behavior of how in complementizer function in English (cf. Legate 2010). Like the English item, it bears a relationship to manner semantics in its development. In other Turkic languages, especially in Modern Uyghur and Uzbek, the cognate verbal noun -(X)š is a fully fledged and highly pervasive inflectional marker in nominal clauses that frequently functions as a bound complementizer, combinable with a large set of complement-taking matrix elements,^[1] a fact that suggests that the situation in Turkish represents an earlier stage of grammaticalization. Although the developments leading to the situation in Modern Turkish on the one hand and in Modern Uyghur and Uzbek on the other are insufficiently investigated,^[2] comparison of the modern languages with the situation in Old Turkic, which will be summarized below (Section 4), reveals that in the modern languages mentioned, verbal nouns in -(y)Iş and -(X)š, respectively, are much more versatile than in Old Turkic, and that Turkish seems to occupy an intermediate position between Old Turkic and, e.g., Modern Uyghur.

The article will focus on the following research questions:

1. What are the synchronic functions of this polyfunctional item?

2. How can these functions be ordered diachronically, i.e., be evaluated in the light of grammaticalization?

3. What is the place of -(y)Iş in the Turkish verbal noun paradigm,^[3] and what are the semantic features in contrast to, especially, the competing verbal nouns -mA and -DIK?

The item -(y)Iş is prolific enough to have entered virtually every grammar of Turkish (cf. Banguoğlu 1986: 261–263, 422; Brendemoen and Hovdhaugen 2004: 130; Ergin 1958: 165–166; Ersen-Rasch 2004: 204–205; Gencan 1975: 213–214; Göksel and Kerslake 2005: 427–430; Kononov 1956: 116–117, 462–465; Korkmaz 2007: 106–107, 172, 899–909; Kornfilt 1997: 450–451; Lewis 1967: 172–173; van Schaaik 2020: 484–489, 693–695). However, its precise function is still not completely clear. Among the studies which comment on the functions of -(y)Iş in detail, Erdal (1998), van Schaaik (1999), Göksel and Kerslake (2005: 427–430), Karakoç and Herkenrath (2016: 641–646), and Coşkun (2019: 15–20) deserve special mention. Erdal’s article (1998) offers the most comprehensive analysis of the functions of -(y)Iş to this day; however, his article is more or less indigestible for general linguists without an advanced command of Turkish as the author does not translate his Turkish examples.^[4] In the present article, I shall draw on Erdal’s insights and analyses for my argument, while also discussing new examples and shifting the focus to the aspect of grammaticalization, and the impact of the manner semantics.

In the following sections, I will first briefly introduce the central terminology as it applies to Turkish (Section 2) and the system of verbal nouns in Modern Standard Turkish (Section 3). After that, I shall give a detailed survey of the functions of -(y)Iş in Modern Standard Turkish (Section 4) before I try to evaluate the position of the complementizer uses and related functions of this item within the Turkish system of verbal nouns (Section 5). Finally, I will propose a path of grammaticalization and reason on the possible contribution of the manner semantics in the evolution of -(y)Iş from a derivation marker into an inflectional marker with complementizer functions (Section 6).

2 Terminology

While most of the terms used in this article are established standard terms, there is considerable variation and/or vagueness in the literature about what concepts they refer to precisely and how they should be applied to Turkish data.

I will use the term grammaticalization in a way that covers not only changes from the “lexicon box” into the “grammar box” (thus polemically expressed by Himmelmann 2004), but also includes developments within grammar, e.g., the assumption of new grammatical functions by items, or the organization in new paradigms (i.e., “processes leading to greater grammaticality of linguistic items” [Himmelmann 2004: 25]).

The term derivation will be used as an umbrella term comprising the word-formation types denominal noun (DNN), denominal verb (DNV), deverbal noun (DVN) and deverbal verb (DVV), in other words, strategies different from composition that create lexemes, i.e., stems with a modified meaning. These strategies may or may not be productive. For this article, only DVN derivations are relevant.

By contrast, inflection will be used to label fully productive processes which do not affect the semantics of the stem itself, i.e., a strategy to create word forms, not new lexemes. Inflectional items are usually applicable to any element of a given word class unless specific semantic or combinational restrictions apply. One criterion that helps identify inflectional markers is the participation of the resulting word form in a paradigm of competing word forms (cf. Matthews 1991: 43–44; Haspelmath and Sims 2010: 16–17, 85–86). There is a grey zone between derivation and inflection, a precondition for the development described in this contribution.

The term complementizer will be used for subordinators which make clauses accessible to complement-taking matrix elements (CTM),^[5] in other words, items that enable clauses to enter argument slots, independently of their morphological status. This view is compatible with Noonan’s characterization of complementizers (“Complement types often have associated with them a word, particle, clitic or affix, one of whose functions it is to identify the entity as a complement. Such forms are known as complementizers”, Noonan 2007: 55),^[6] and conforms to Kehayov and Boye’s conception of complementizers as complementation markers (2016: 7; see also Karakoç and Herkenrath 2016: 625). It differs from the equally widespread conception of complementizers as “words” or “particles”, i.e. (more or less) free items (thus probably intended, although not explicitly stated, by Dixon 2006: 24). Thus, the term complementizer in the present article refers to a specific syntactic property of a given element, regardless whether it is bound or free, and regardless whether the embedded clause is finite or non-finite.^[7]

Contrastingly, the term nominalizer will be used to refer to a part-of-speech-related concept, i.e., an item that forms nouns. (The nominal nature of the resulting item can be tested by adding morphemes of noun inflection). In other words, the term nominalizer is associated with its function to form a specific kind of part of speech, i.e., a noun, while the term complementizer is associated with its syntactic function, i.e., the faculty of making clauses accessible to government by a CTM. Consequently, items may be both nominalizers and complementizers. In the case of Turkish -(y)Iş, we can speak of a nominalizer, which in some of its uses is also a complementizer. Nominalizers comprise derivational and inflectional markers. In the present article, nominalizers which are inflectional markers will be labeled verbal nouns (VN), while nominalizers which are derivational markers will be labeled deverbal nouns (DVN; cf. van Schaaik 1999: 88–89). This usage follows a firmly established terminology in Turkic and Mongolic studies.^[8]

Evaluating the classes complementizer and nominalizer together, it must be added that Turkish verbal nouns can appear in complement clauses, i.e., clauses that fill argument slots of CTMs, but they are not confined to this function: They also occur in nominal slots that do not have argument status, including the predicate slot of a sentence (as nouns can function as predicates in Turkish),^[9] and as the host of case suffixes not governed by CTMs, which then have adjunct status in the sentence.^[10] These cases are not complement clauses sensu proprio (although they must often be translated into English as complement clauses), but other embedded nominal clauses with various syntactic functions. I consider these cases equivalent to, but not identical with complement clauses, as they behave similarly. They will be included in the discussion in this article but pointed out as special cases. The verbal nouns in these clauses function as inflectional nominalizers with a functional scope that ranges beyond that of prototypical complementizers. Thus, complementation is a frequent function of Turkish verbal nouns, but it is not the only one.

In the opposition finite versus non-finite, the term finite will be used for the capacity of an item to form predicates of independent sentences, while the term non-finite will be used to designate the absence of this capacity. In this respect, Turkish markers such as -Iyor, -sIn, and -DI are unambiguously finite markers, while markers such as -mA, -DIK and -(y)Iş are unambiguously non-finite markers.^[11] There are items in Turkish that may be either finite or non-finite, where the concrete status is determined by the paradigm to which it belongs (e.g., the 1.sg.fut item -(y)AcAğIm may be a verbal noun, participle [=non-finite] or finite form, while the forms of -(y)AcAK inflected for other persons are unambiguous in this respect, e.g. 2.sg.fut -(y)AcAğIn [non-finite] vs. -(y)AcAksIn [finite]).^[12]

I will use the term proposition to designate an abstract truthbearer, i.e., a covert pragmatic entity that need not be immediately formally identifiable but can be retrieved from an utterance (“coercion”, Boye 2012, see Example (3) below; cf. Rentzsch 2015: 18–20). Propositions can be evaluated as true or false, certainly, likely or unlikely true, etc. Contrastingly, the opposition factual versus non-factual will be reserved in this article for semantic qualities encoded by morphological means. In other words, factuality as understood here is directly associated with a form. Thus, it can be said that the Turkish nominalizer -DIK encodes the feature [+factual], while the nominalizer -mA is [−factual] (see, among other things, Kornfilt [2007: 315]; Csató [2010: 114, 116–117] uses the terms assertive versus non-assertive in more or less the same sense; compare also the factive versus action distinction by Lees [1965: 115], and the fact versus act distinction by van Schaaik [1999: 90–91]).

The relationship between factuality and the proposition, as distinguished in this article, needs some explanation. Propositions can be said to be true or false, and they can be evaluated for their truth value less straightforwardly, e.g., unlikely true, likely true, certain, etc. If a proposition is (said to be) true, it is presented as a fact. On the other hand, while propositions, i.e. abstract truthbearers, are covert, i.e., hidden within an utterance but retrievable by pragmatic means,^[13] there are also items which encode factuality in their semantics, such as the above-mentioned nominalizer -DIK. Other, paradigmatically contrasting items, may be lacking this feature, thus representing the negative value in an opposition [±factual]. This does not mean that [−factual] items necessarily deny factuality.^[14]

The intricate relationship between propositions and factuality in Turkish is illustrated in the following examples. In (1), the SoA^[15] <reading the book> is presented as a possibility (in the non-terminological, general sense, i.e., an act in van Schaaik’s [1999] use of the term rather than a fact); hence it is formally marked by the nominalizer -mA with the semantic value [−factual]:

Conversely, in the following example, the reading of the book is presented as a fact and marked thus by the factual marker -DIK. At the same time, the sentence includes a proposition <I have read the book> (which can be said to be either true or false).

On the other hand, the following construction, although using the non-factual verbal noun -mA, also includes the proposition <I have read the book>, which is not encoded semantically, i.e., in terms of a direct form-meaning relationship, but can be retrieved from the utterance by pragmatic means.

Example (3) shows that the use of a [−factual] item does not deny the notion of factuality, it just does not actively mark it (privative opposition).^[16] The notion of proposition, on the other hand, is not necessarily directly associated with formal marking at all: from Example (3), the proposition <I have read the book> can be retrieved.

3 The system of verbal nouns in Turkish

In Turkish, clausal complements are most commonly constructed by means of verbal nouns, i.e., inflectional nominalizers which then function as bound complementizers. These inflectional markers are suffixed to the stem of the predicate of the complement clause and produce nominal forms that accept possessive suffixes, case suffixes, etc. Consequently, this kind of complement clause is essentially a complex nominal phrase, involving at least a predicate, usually represented by a verb stem,^[17] which is nominalized by an inflectional suffix, here labelled verbal noun, and potential arguments such as subject, direct and indirect objects, as well as adjuncts. The nominal clause can basically enter any nominal slot within the syntagma, including argument slots and also the predicate slot, as Turkish permits nominal predicates. Those nominal clauses that occur in the predicate slot of the matrix sentence, as well as case-marked nominal clauses that are not governed by a CTM, are not complement clauses in the proper sense (see above, Section 2).

Examples of a simple and a complex nominal clause in object position are given in (4) and (5), respectively. As can be seen from Example (5), the complement clause may contain further subordinate clauses, in the present case three adverbial clauses. Example (6) demonstrates that clausal complementation is potentially recursive, i.e., a complement clause may contain further complement clauses.^[18]

There are also “free”^[20] complementizers in Modern Standard Turkish such as postponed diye, originally a converb (“gerund”) from the verb demek ‘to say’, with quotative and other functions (Example (7)), or the particle ki, copied from Persian, one of the functions of which is to introduce finite complement clauses (Example (8)). These items shall not be considered in this article, but one example each shall be given for the sake of completeness (cf., e.g., Csató 2010: 108–110; Deny 1921: 848–850, 915–917; Johanson 1975: 105–106; Kornfilt 1997: 439–440, 443).

The most frequent verbal noun markers that can assume complementizer function in Turkish are -DIK, -(y)AcAK, and -mA. These items can be considered members of the Turkish paradigm of verbal nouns. The difference between -DIK and -mA can be characterized as factual versus non-factual (see above, Section 2). This terminology can be translated in functional terms to the effect that -DIK forms proposition-type complement clauses, while -mA forms SoA-type complement clauses. Both verbal nouns occur in the following example, in which the first nominal clause functions as the subject argument to the predicate belli ‘obvious’, while the second nominal clause does not function as a complement, but as an adjunct. Although the syntactic status of the two nominal clauses within the matrix sentence is different, the two verbal nouns can be compared on a semantic level.

Note that in the second half of this example -mA does not deny the factuality of the SoA, it just does not encode the feature [+factual] (see above, Section 2). On the contrary, we must assume that the food is actually fresh. The use of the [−factual] verbal noun must probably be explained with the contrast to the corresponding form with -DIK, ol-duğ-undan, which would convey a meaning of causality (‘because it was fresh’).

-mA has a mainly combinational variant -mAK, which occurs with the locative and ablative cases as well as in unmarked form, while according to contemporary Turkish spelling rules -mA is used with the genitive, accusative, and dative. Locative and ablative can combine with either -mAK or -mA. In those positions where both items can occur, constructions with -mA are said to be “less abstract in meaning than those formed with -mAK” (Göksel and Kerslake 2005: 418).

It is important to note that the selection of either -DIK or -mA is in many cases, although not always, governed by the CTM. For example, belli ‘clear’ and kesin ‘certain’ select -DIK,^[21] while muhtemel ‘probable’, rağmen ‘in spite of’ or lazım ‘necessary’ select -mA. Some combinational issues of this nature have been described by Csató (2010) and van Schaaik (2014, 2020: 603–631). The combinational rules can in part be explained from the semantic design of the CTM, but historical developments – conventionalization and grammaticalization – play a role as well.

The third common verbal noun, -(y)AcAK, has a prospective meaning. Like -DIK, it forms propositional-type complement clauses as in (10).

The verbal noun -(y)AcAK will not be dealt with in this article, while -DIK and -mA will be considered for the sake of comparison with -(y)Iş. There are further VNs in Turkish with a less prominent position in the verbal noun paradigm such as -(y)AsI, -(V)rlIK, -(mA)mAzlIK, -mIşlIK, and -mAklIK (cf., e.g., Rentzsch 2016: 62–64). These will be left out here as they do not contribute significantly to the establishment of the status of -(y)Iş in the verbal noun paradigm.

4 The functions of -(y)Iş in Modern Standard Turkish

-(y)Iş is a frequent word formation suffix in Turkish that forms DVNs with concrete or abstract semantics such as

Another prominent meaning of nominal derivations in -(y)Iş in Turkish is the manner in which the action is performed, e.g.,

There are ambiguous cases; e.g., it can be debated whether gülümseyiş (from gülümse- ‘to smile’) signifies ‘the way of smiling’ (i.e., the manner) or just ‘the smile’ (i.e. the abstract noun). It can convey either meaning, the appropriate reading being retrieved from the context. Similarly, it can be argued that anlayış and görüş in (11) are not just plain abstract nouns but also encode or imply manner (‘way of understanding’, ‘manner of seeing’). The context-dependence of the readings of the item -(y)Iş is elucidated by the following examples by Gencan (1975: 214).

Yağış in Example (13) is a plain, concrete noun without any manner implications while in (14), the formally identical item yağış obtains a clear manner reading.

The items in (11)–(13) above are plain nouns which do not even designate an SoA as they stand. They can be said to be “nominal” in that they do not normally preserve the government features of the corresponding verb stems. Contrastingly, Example (14) already typifies a usage of -(y)Iş which transcends the function of a plain noun (see below).

The derivational function of -(y)Iş is historically old; for its cognate, -(X)š, it is documented since the Old Turkic era (8th–12th century), where, according to Erdal (1991: 266), it is mostly used to form abstract nouns (such as ėniš ‘descent’ from ėn- ‘to descend’; Erdal 1991: 269) but may also be used to form concrete nouns (such as adrïš ‘fork of a road’ from adïr- ‘to separate’; Erdal 1991: 265).

Apart from the derivational function of -(y)Iş, Modern Standard Turkish has developed inflectional uses of this item, capable of forming nominal clauses, including complement clauses, like -mA, -DIK, and -(y)AcAK, but described as “more restricted in its functions” by Göksel and Kerslake (2005: 427–430; cf. van Schaaik 1999: 103–104). Inflectional uses of -(y)Iş are said in the literature either to expresses the manner in which the action is performed (like the DNVs in Example (12) above; see Erdal 1998: 57), or a single instance of an event, a usage which makes the event countable (Göksel and Kerslake 2005: 428; van Schaaik 1999: 100; cf. Erdal 1998: 58). Differently from the lexicalized derivations mentioned before, these formations are fully productive and can be formed from every verb stem (Erdal 1998: 54). As already mentioned, formally identical items may serve either as derivational markers (DVNs) or as inflectional markers (VNs), and the function becomes clear from the specific context in which the item occurs. The manner reading can be seen in the following example:

In Example (15), it is not feasible to analyze utanmayış as a lexicalized noun as it is modified by an adverbial clause (gülerken).

Contrastingly, the countable instance reading is typified by the following two examples, which include the attributive modifiers ilk ‘first’ and son ‘last’, respectively:

In Example (15), -(y)Iş forms part of an adjunct, while in Example (16)–(17) it forms part of the predicate. Hence, these examples do not typify complement clauses sensu stricto, but other types of embedded nominal clauses.

However, there are also occurrences of -(y)Iş as an inflectional marker that cannot be explained in terms of manner or countable events, and which are difficult to describe. These occurrences seem to be capable of occurring in any nominal slot within the syntagma and thus cover the full range of sentence-internal combinational possibilities of Turkish nominal clauses, like nominal clauses with -mA and -DIK. Note that, different from how-clauses in English, Turkish -(y)Iş-clauses may occur in positions not assigned case (cf. Legate 2010: 124).^[22] An example for a -(y)Iş-clause in the subject slot:

An example of a -(y)Iş-clause in the predicative noun slot (cf. Example [25] below):

Example (18) can be analyzed as a complement clause with the predicate tuhaf ‘strange’ as CTM, while Example (19) is a nominal clause that functions as a predicate. The constructions in Examples (18) and (19), which do not demonstrate any conspicuous reading in terms of manner or of countable instances, seem to be “optional” in a sense that -(y)Iş can be interchanged either by -mA or by -DIK. This fact has been noted and supported by examples by Karakoç and Herkenrath (2016), whose contribution has its own section on -(y)Iş (2016: 641–646), with a quite detailed discussion of the qualities that distinguish -(y)Iş from -mA and -DIK. While both Erdal (1998) and Karakoç and Herkenrath (2016) study the VN uses of -(y)Iş in detail, they differ in their findings. Erdal considers the item a “factive imperfective” VN “clearly linked to topicalization” (1998: 63), whereas Karakoç and Herkenrath do not accept the value of factivity and imperfectivity (2016: 642) but contend that the item encodes a “direct reference to the inner process of an action” (2016: 643, 645) “without reference to the knowledge about it” (2016: 642).

The function of -(y)Iş in the Turkish system of verbal nouns will be re-investigated in the subsequent section, and the opinions about this item will be discussed in light of the findings presented there.

5 The function of -(y)Iş within the Turkish system of verbal nouns

The function of a grammatical item can only be fully understood in contrast to competing items, in other words, from the oppositions that are realized within a given paradigm. In this section, occurrences of -(y)Iş in primary sources will be investigated, and their exchangeability with the factual VN -DIK and the non-factual VN -mA will be tested (commutation test). It will turn out that in nearly all instances, -(y)Iş can be replaced either by -mA or by -DIK, or sometimes by both.^[24] -mA and -DIK, on the other hand, are mutually less liberally interchangeable and, if they can be replaced by the other item, usually produce highly different semantic readings.^[25] Thus, -(y)Iş has not (yet?) become an obligatory member of the Turkish VN paradigm.^[26] On the other hand, native speakers of Turkish confirm that certain differences in meaning can be perceived when -(y)Iş is replaced by one of the other items, although the semantic difference seems to be difficult to grasp.

Let us first consider two examples in which -(y)Iş can be replaced by the non-factual VN -mA, but not by the factual VN -DIK:

The CTM iyi ‘good’ conventionally governs -mA, not -DIK, in Turkish (Csató 2010: 117). Thus, in the present example, -(y)Iş enters a slot where otherwise a non-factual VN would be expected. However, neither -(y)Iş nor -mA deny the factuality of the complement clause: On the contrary, the days’ being filled with activities is presented as a reality.^[27] What -(y)Iş and -mA have in common is that in the present structure, the SoA is highlighted, not the proposition.

Example (21) is especially intricate as the complement clause formally functions as a possessor which is governed by a possessee, in the present case boşluk ‘emptiness’. Possessees are a special case of CTMs and may, in principle, govern either complement clauses with the non-factual VN -mA, as the alternative in Example (21), or complement clauses with the factual VN -DIK.^[28] Differently from iyi in Example (20), the selection of the verbal noun thus cannot be motivated by conventionalized government rules of the CTM. Clearly, the choice of the VN must be determined by purely semantical properties here. In Example (21), -(y)Iş and its possible alternative -mA do not assign factuality to the complement clause, but they do not inhibit the interpretation p is true for p = <the concrete person that is being talked about in this context is not understanding anything>. At the same time, the example sentence is not about a concrete situation, but about a type of emptiness characterized by <not understanding anything>, which is more suitable for encoding by a [−factual] complementizer.

Contrastingly, in (22), -(y)Iş can be replaced by the factual VN -DIK:

Note that in this example, izleyiş and tırmanış do not designate ‘the manner of watching’ and ‘the manner of climbing’, but the concrete event of watching and climbing;^[29] nonetheless it is appropriate to translate these instances into English with a how-clause (cf. van Schaaik 1999: 100). Replacement with -mA is not possible in this example. -(y)Iş can be replaced by -DIK, but then the sentence – or more precisely: the CTM – acquires a modified reading: ‘She thinks [that İbrahim watched them and that she climbed the hill with Ömer].’ In other words, the CTM düşün- combined with -(y)Iş means ‘to think about’, while with -DIK it means ‘to think that’, with an epistemic meaning component, which results from the marked factuality of -DIK and its faculty for forming propositional-type complement clauses.^[30] This behaviour of -(y)Iş is well reconcilable with Karakoç and Herkenrath’s analysis of the item in terms of “a direct reference to [the] process of an action without reference to the knowledge about it” (2016: 642).

The fact that -(y)Iş can be replaced sometimes by -mA, sometimes by -DIK, might lead to the idea that -(y)Iş is both factual and non-factual, an assumption which is, of course, hardly possible. Rather, it must be the case that the semantic value of -(y)Iş allows interpretation in terms of factuality and non-factuality, depending on the context. In this respect, -(y)Iş sides with -mA, which is [−factual] but does not necessarily exclude factual readings, while -DIK, positively marked for factuality, regularly produces a propositional reading.

In some instances, -(y)Iş can be replaced both by -mA and by -DIK. These are the most interesting and complex cases, as the semantic differences between these options may be more or less obvious. Native informants tend to be able to say that the choice of this and that VN is possible, acceptable, or “good”, but often cannot explain the precise difference. This issue needs more investigation with a large number of informants and carefully selected examples.^[31] Here, only preliminary results can be presented on the basis of Examples (23)–(25); a detailed study of the semantic subtleties is still to be done and must probably take into account the transitionary nature of the status of -(y)Iş.

In this example, the semantic implications of the selected verbal nouns are not very clear. It seems that ol-duğ-undan puts emphasis on the reason of being able to endure the girl,^[32] while ol-uş-undan and ol-ma-sından rather focus on the feature of being blue-eyed itself (i.e., the SoA). The reason is encoded in the ablative case suffix -DAn, but the selection of the verbal noun -DIK seems to shift the focus from the event (or the property) to the reason. Ol-uş-undan, on the other hand, seems to shift the focus to the girl’s specific property of being blue-eyed, although the difference from ol-ma-sından is difficult to grasp. Note that in this example, the nominal clause occupies the predicate slot in the sentence, as opposed to the following example, in which it enters the subject slot:

In (24), the variant with -mA implies that the pure event of traveling and disappearing is talked about, a reading which corresponds to an SoA-type complement clause, while the variant with -DIK implies that the fact of Ayhan’s traveling and disappearance is discussed, a reading that corresponds to a propositional-type complement clause. The variant with -(y)Iş implies that the whatabouts and whereabouts of Ayhan’s traveling and disappearance are discussed, in other words, not the manner of traveling and disappearance, but the concrete circumstances of the event.^[33] The last example is an extended version of Example (19) above (the nominal clause is in the predicate slot):

In this example, according to informants, the variant with -DIK signifies that the problem lies in the fact that we misunderstand some things, while in the variant with -mA, the event of misunderstanding itself is focused on. The variant with -(y)Iş, which appears in the original text, seems to imply that the problem lies in the character of the concrete event. The semantic subtleties are not immediately obvious, but a difference can be perceived. Example (25) is also a counterexample to Erdal’s claim that -(y)Iş is “clearly linked to topicalization” (Erdal 1998: 63), as -(y)Iş forms part of the comment.

So, how can the complementizer uses of -(y)Iş and its related functions in nominal clauses be analyzed, to be more precise, those instances that can neither be explained in terms of manner in a narrow sense nor in terms of countable events? The main opinions about -(y)Iş in literature – factive imperfective (Erdal 1998), single instance of an event (Göksel and Kerslake 2005; van Schaaik 1999), direct reference to the inner process of an action (Karakoç and Herkenrath 2016) – can all be supported by some of the examples cited but do not seem to account for all readings: How can “being blue-eyed” (Example (23)) be reconciled with imperfectivity or reference to an inner process? Is “misunderstanding some things” (Example [25]) a single instance of an event?

It is obvious that factuality itself, which is the distinctive criterion in the opposition between -DIK and -mA, is not coded per se in the semantics of -(y)Iş, otherwise, replacement with -mA, like in Examples (20)–(21), would be impossible. Thus, the semantic value of -(y)Iş is [−factual.] On the other hand, -(y)Iş (like -mA) allows readings in terms of proposition or SoA, depending on the context; this is why -(y)Iş is interchangeable with both -mA and -DIK in Examples (23)–(25). -DIK with its semantic value [+factual], on the other hand, regularly evokes propositional readings.

The examples presented in this study support the following claims:

1. -(y)Iş as an inflectional marker forms SoA-type nominal clauses.

2. -(y)Iş does not assign marked factuality.

3. -(y)Iş does not deny the notion of factuality. Thus, like -mA, it represents the value [−factual] in a privative opposition to [+factual].^[34] In other words, -(y)Iş (like -mA) represents either non-factuality or the neutral value in the opposition [±factual] and may produce factual readings in suitable contexts.

4. A suitable context provided, a proposition can be retrieved from a nominal clause marked with -(y)Iş (“coercion”, Boye 2012).

5. -(y)Iş itself does not encode a positive viewpoint aspect value, thus it is principally combinable with aspect markers such as the progressive in Examples (21) and (25). This feature is shared with -mA and -DIK.

6. Differently from -mA, -(y)Iş is semantically more focused on the character (type) of the SoA or on concrete circumstances, a quality which, a suitable context provided, may produce readings in terms of an internal process, while -mA focuses on the SoA as such, without special attention to its character. This quality of -(y)Iş is related to the manner reading of the DVN.

Thus, the verbal noun -(y)Iş forms SoA-type nominal clauses which are [−factual] and present the character or the concrete circumstances of an event. Depending on the content of the SoA itself and its context, the falling of rain (Example (14)), not being ashamed while laughing (Example (15)), falling in love (Example (16)), everybody’s having an occupation (Example (18)), misunderstanding some things (Example (19)/(25)), Ibrahim’s watching somebody from the boat (Example (22)), etc., may be interpreted more or less in terms of (i) manner, (ii) whatabouts and whereabouts, and (iii) concrete circumstances.

These claims can be reconciled with the opinions in literature as follows: Events envisaged with their concrete circumstances are potentially countable, a fact that corresponds to one of the generally accepted usages of this item, which often co-occurs with expressions denoting first, last, or every. They are performed in a specific way, which agrees with the widespread manner reading. They are often presented as actually valid, or as representing valid types, which allows reading as a fact and interpretation in terms of a proposition. But they can also be talked about with their nature and character, a possibility that allows interpretation in terms of an SoA. The proposal is thus that -(y)Iş in its use as a verbal noun is non-factual and encodes the character or circumstances connected to an event (or a state or a quality), with context-specific readings covering the range demonstrated in this article, and in literature. The vacillating use of -(y)Iş can be motivated by the fact that it is an emerging inflectional marker and not yet fully paradigmatized in contrast to other, more widely used verbal nouns, such as -mA, -DIK, and -(y)AcAK. Its nature as an emerging – i.e., not yet firmly established – inflectional marker also accounts for some combinational restrictions mentioned in the literature, such as the fact that -(y)Iş clauses rarely contain adverbial modifiers (van Schaaik 1999: 103).^[35] The development of -(y)Iş on the derivation-inflection cline is not (yet) advanced enough to favor all combinational properties of fully developed verbal nouns.

6 A possible path of grammaticalization

As has been demonstrated in this article, Turkish -(y)Iş shares with -mA the semantic value [−factual] and the syntactic capacity to form SoA-type nominal clauses, including complement clauses. In order to offer a possible explanation for the functional and semantic ambiguity of this item, both historical data, especially from Old Turkic, and the more far-going functions in modern Turkic languages such as Modern Uyghur and Uzbek must be taken into consideration.

In light of the Old Turkic data, it can be said that -(X)š/-(y)Iş started as a DVN, i.e., a derivation suffix mostly forming abstract or concrete nouns. This function survives in Turkish to this day (cf. Example (11)). We furthermore observe highly prominent readings for this item in terms of manner, which occur both with fully lexicalized nouns (Example (12)) and in more productive derivations (Example (14)). Probably departing from this productive DVN, the morpheme has developed more flexible functions in the transitional zone between derivation and inflection, encoding the character or circumstances connected to an SoA, with a relatively broad spectrum of context-specific interpretations, including readings in terms of manner, whatabouts and whereabouts, countable events, etc.^[36] In a similar way to how in English, which has developed complementizer functions (Legate 2010), -(y)Iş can function quite productively as a bound complementizer in Turkish, with uses overlapping with either -mA, -DIK, or both, but generally more restricted in function than the latter two VNs (cf. Examples (20)–(25)). Differently from how in English, nominal clauses in -(y)Iş can function as the subject of a sentence, and they can even occur in the predicate slot and as hosts of case markers such as the locative and the ablative, which may produce adjuncts. In these respects, the functionality of -(y)Iş exceeds that of classical complementizers.

Already Erdal has drawn attention to the high versatility of the cognate nominalizer -(X)š in Uzbek, Modern Uyghur, and Kirghiz, where it is frequently involved in auxiliary constructions with CTMs such as mumkin ‘possible’ and kėrek ‘necessity’ (cf. Erdal 1998: 64 and Endnote 14). In Modern Uyghur, this item has developed into the general non-factual verbal noun, which covers almost all functions that are served by -mA (and -mAK) in Turkish, in other words, the Modern Uyghur VN -(X)š has a much broader range of usage than Turkish -(y)Iş. Compare the following two examples from Uyghur, (26a) and (27a), with translation both into Turkish in (26b) and (27b)^[37] and into English.

Compared to the situation in Modern Uyghur, the development of -(y)Iş into a full-fledged inflectional marker and bound complementizer is at a less advanced stage in Turkish. Rather than a general non-factual VN, Turkish -(y)Iş serves the more specialized task of marking the character or circumstances connected to an SoA, with readings such as manner, whatabouts and whereabouts, and concrete circumstances. It seems reasonable to assume that the manner reading of the DVN -(y)Iş constitutes the point of departure for this item’s evolution into a more versatile VN. The variability in function, the shifting exchangeability with other VNs, and the vacillating readings of -(y)Iş must be evaluated in light of its transitional stage on the path from derivational marker to inflectional marker.