Adverbial and attributive modification of Persian separable light verb constructions

Abstract

Persian makes extensive use of light verb constructions (LVCs) consisting of a non-verbal preverb and a semantically light verbal element. The current paper concentrates on LVCs with nominal preverbs (e.g. sedâ dâdan ‘produce a sound’, lit. ‘sound give’) which license an attributively used adjective intervening between the two components of the construction. Such LVCs are idiomatically combining expressions, in the sense of Nunberg, Sag & Wasow (1994: 496). The individual components of idiomatically combining expressions have an identifiable meaning and combine in a non-arbitrary way. Thus, they are conceived as being formed compositionally. Evidence for this view can be taken from the fact that the attributively used adjectives function as internal modifiers, targeting only the nominal component of the LVC.

As adjectives can also be used adverbially, two modification patterns emerge: The nominal preverb is modified by an attributive modifier, or the same adjective can be used as an adverbial modifier of the whole LVC. Two corresponding interpretation patterns arise: Attributive and adverbial modification either both result in the same, or in different interpretations.

The paper makes the following claims: First, only compositionally derived LVCs license attributive modification of their nominal preverb; and second, different interpretations of the two modification patterns only result if the light verb and the preverb each license a suitable property as a target for the modifier. If, on the other hand, such a property is only licensed by the preverb, adverbial and attributive modification result in the same interpretation.

1 Introduction

In a paper on idioms, Nunberg et al. (1994) argue that although idioms have a conventionalized and sometimes figurative meaning, they can be compositionally derived. The authors introduce the term ‘idiomatically combining expressions’ for idioms such as English pull strings, take advantage and take a bath. The different components of such idiomatically combining expressions have identifiable meanings, which are combined in a non-arbitrary way. One piece of evidence supporting this view is that a number of idiomatic expressions allow for internal modification (Ernst 1981, Gazdar et al. 1985, Nunberg et al. 1994). Thus, a modifier – which is not an obligatory component of the expression – is inserted into the idiom, as in pull political strings. Some authors (e.g. Everaert & Hollebrandse 1995, Nicolas 1995, Glatz 2006, Megerdoomian 2012) argue that the internal modifier always modifies the whole idiom and not just one of its components. This holds true for pull political strings, which can be paraphrased as ‘politically pull strings’ (Ernst 1981: 51). But such an analysis is doubtful for cases like those in (1). The two examples in (1) are instances of so-called light verb constructions (LVCs), which are complex predicates formed by a semantically light verbal element and a second predicational component. In the English example in (1a), warm functions as an internal modifier, only targeting the nominal component of the LVC. Thus, the adjective specifies the temperature of the water of the bath and (1a) resists paraphrasing into an expression like ‘bath warmly’. In the Persian example in (1b), the adjective boland ‘loud’ functions as an internal modifier of the nominal preverb sedâ ‘sound’. The adjective modifies the loudness of the produced sound, but not the loudness of the event of producing a sound. Thus, (1b) does not mean ‘producing a sound loudly’ or ‘doing something loudly which produces a sound’.

Various examples that Nunberg et al. discuss under the notion of ‘idioms’ and ‘idiomatically combining expressions’ are in fact LVCs. In the current paper, we explore Persian LVCs from the perspective of attributive and adverbial modification. We essentially argue that (i) internal (attributive) modifiers do not modify the whole LVC but only the nominal preverb, and (ii) interesting asymmetries can be found by comparing adverbial and attributive modification of LVCs. In (2), the adjective boland ‘loud’ is used as an adverbial modifier of the LVC sedâ dâdan ‘produce a sound’; the interpretation of (2) is the same as for (1b). The adverbial modifier is again interpreted as indicating the loudness of the resulting sound.

Thus, there is no semantic difference in the interpretation of adverbial and attributive modification in this particular case. As shown in (3) with respect to the LVC mesâl zadan ‘give an example’ (lit. ‘example hit’), attributive and adverbial modification can result in different interpretations. In (3a), xub ‘good’ specifies the manner of giving an example, whereas the attributive modifier in (b) specifies the quality of the example.

A natural question resulting from the brief discussion of these examples is the following: Under which conditions do adverbial and attributive modification result in the same interpretation, and when do different interpretations arise? The paper addresses this question for a number of Persian LVCs.

Nunberg et al. take the possibility of internal modification of idiomatically combining expressions as one piece of evidence in favor of a compositional analysis of certain idiomatic expressions. We follow this route and propose a compositional analysis of those Persian LVCs licensing internal modifiers. The possibility of the internal modification lends support to such a compositional treatment of at least a subset of Persian LVCs. We are not the first to advocate for a compositional analysis of Persian LVCs (e.g. Karimi 1997, Müller 2010, Samvelian & Faghiri 2014), but our approach strengthens this proposal by demonstrating the usefulness of modification as a diagnostic of compositionality.

The paper is structured as follows: In Section 2, we introduce the notion of LVCs with a special emphasis on their compositionality. The section has a special focus on Persian LVCs and introduces the distinction between separable and non-separable ones. In Section 3, we focus on functional morphology realized at the nominal preverb. It will be shown that nominal preverbs behave essentially like other nouns but resist case marking. This is, as we will argue, due to the fact that preverbs are not the object of the light verb. Section 4 focusses on adverbial and attributive modification and introduces the different interpretation patterns mentioned above. A basic claim we defend in the paper is that attributive modification of the nominal preverb of LVCs proves that separable LVCs are semantically compositional. In Section 5.1, we develop a compositional analysis of LVCs. Section 5.2 introduces the data we use in our case study on the modification of Persian LVCs. In the case study, we investigate verbs of emission (e.g. English bleed, drone, shatter). The relevant background on this semantic verb class is introduced in this section as well. Section 6 presents our analysis of how the different interpretation patterns arise with respect to the examples introduced in Section 5. Finally, we will provide a general discussion of the approach presented in the paper in Section 7.

2 Light verb constructions

Light verb constructions are a special type of complex predicates (see, e.g. Amberber, Baker & Harvey 2010 for various types of complex predicates). Butt (2010: 49) defines the notion ‘complex predicate’ as a construction consisting of ‘two or more predicational elements (e.g. nouns, verbs and adjectives) which predicate as a single unit, i.e. their arguments map onto a monoclausal syntactic structure’. In LVCs, the complex predicate consists of a semantically light verb and a second component, which provides the major semantic contribution to the joint predication. This second component, as it precedes the light verb in Persian, is termed ‘preverb’. Light verb constructions have been extensively discussed from an argument structure perspective (see, for an overview, the literature cited in Butt 2010, 2013, Butt & Lahiri 2013). With respect to the semantics of LVCs, a central question is whether the light verb is semantically bleached or not. In the latter case, it is assumed to add a semantic contribution to the complex predicate. The first position is, for example, advocated in Jespersen (1942), Cattell (1984) and Grimshaw & Mester (1988). Butt especially (e.g. Butt & Geuder 2001; Butt 2010, 2013) has been explicit in assuming that light verbs are not semantically empty, but contribute a, probably hard to clearly define, semantic meaning (e.g. volitionality, agentivity). A similar view is adopted by other authors as well, e.g. Isoda (1991) and Brugman (2001). This view is supported by diachronic analyses (Bowern 2008, Butt & Lahiri 2013) which demonstrate that light verbs do not result from a process of semantic bleaching. In Section 5, we present evidence that Persian light verbs make a semantic contribution to the complex predicate.

Persian makes extensive use of LVCs as the language only has a very limited set of lexical verbs; according to Mohammad & Karimi (1992: 195) only 115 full verbs exist. In Persian, the preverb of an LVC can either be a noun (4a), a participle (b), an adjective (c) or a prepositional phrase (d). The light verbal element is taken from the set of lexically full verbs, but not all full verbs are (productively) used as light verbal elements (see Dabir-Moghaddam 1997 for an overview of Persian light verbs).

To avoid confusion, a note on the terminology used in the paper is in order. We are conceiving of LVCs as complex predicates consisting of a verbal head, which is called ‘light verb’, and a preverb. We use the term ‘full verb’ to denote elements of the lexical category ‘verb’. The terms ‘light verb’ and ‘heavy verb’ designate uses of full verbs. The term ‘light verb’ designates a semantically reduced usage of a full verb. In such a use, the verb requires a preverb to express a full-fledged predication. ‘Heavy verb’, on the other hand, signifies a usage of a verb, in which it has full predicational content. In its heavy use, dadân ‘give’ denotes an event of giving (a transfer of a theme from an agent to a recipient). In its light use, the verb does not denote an event of giving. Rather, the denoted event is dependent on the preverb. For example, the LVC jâvab dadân literally means ‘answer give’ and is interpreted as meaning ‘to answer’. The LVC does not denote a literal transfer of an answer from an agent to a recipient but an event of answering. The type of denoted event crucially depends on the preverb but not on the light verb.

Two major issues with regard to LVCs are (i) their idiomaticity and (ii) whether they show lexical or phrasal properties. Light verb constructions have a conventionalized and often figurative interpretation. In an influential paper, Nunberg et al. (1994: 499ff) argue that idiomatically combining expressions are compositional, even though they have conventionalized and possibly figurative meaning. One of Nunberg et al’s (1994: 504) examples is the English LVC pull strings, which, as a complex predicate, cannot simply be interpreted literally. Pull is not used in its ‘heavy’ sense and strings is interpreted metaphorically. Nevertheless, as Nunberg et al. argue, the two components of the LVC have an identifiable meaning and combine in a non-arbitrary way. One piece of evidence in favor of such a compositional approach is internal modification (Ernst 1981, Gazdar et al. 1985, Nunberg et al. 1994). An English example is given in (5) below. In it, the adjective immediate modifies the nominal component (point) of the idiomatically combining expression beside the point. Nunberg et al. reason that if a modifier can target only one component of an idiomatically combining expression rather than the whole idiom, the parts need to have an identifiable meaning.

Various authors (e.g. Everaert & Hollebrandse 1995, Nicolas 1995, Glatz 2006, Megerdoomian 2012) propose that in cases of internal modification, it is always the whole expression which is in the scope of the modifier rather than just one of its parts. These authors claim that although immediate is syntactically realized as a modifier of the noun point, it scopes over the whole idiom, rather than just the noun. Under such a view, an idiom like beside the point shows syntactic flexibility – elements can be placed within – but it does not necessarily follow that its parts have an identifiable meaning. With respect to (5), Ernst (1981: 52) states that immediate ‘adds the information that the point which someone is not addressing is the one immediately at hand’. External modifiers, like political in He came apart at the political seams (Ernst 1981: 51), modify the whole idiom. Ernst (1981: 51) calls them ‘domain delimiters’ as they specify the domain to which the idiom applies. Thus, the above mentioned idiom can be paraphrased as ‘as far as politics is concerned, he came apart at the seams’ (cf. Ernst 1981: 55). A similar paraphrase does not work for the internal modifiers; immediate does not specify a domain with respect to which the idiom is restricted. This demonstrates, on the one hand, that an interpretational difference between internal and external modifiers exists, and, on the other hand, that immediate does not behave like an external modifier. Thus, we follow Ernst’s analysis and propose that internal modifiers do not modify the whole expression but just part of it. One of the paper’s goals is to demonstrate that attributive modifiers of nominal preverbs are internal modifiers and therefore present evidence in favor of a compositional analysis of (at least some) Persian LVCs. The property of internal modification only shows that the preverb has an identifiable meaning; it does not show that the components not targeted by the modifier (e.g. the light verb) do as well.

A property showing that the light verb has a regular contribution is that compositional LVCs come in families (Nunberg et al. 1994, Sag et al. 2002). Members of a family are headed by the same light verb, whereas only the preverb differs. The different members of a family exemplify the same interpretational pattern. Two LVC families headed by the same light verb keʃidæn‘pull’ are (Family 2011: 13): (i) the combination of the light verb with a noun denoting a smokeable substance results in the interpretation ‘to smoke N’ (e.g. sigar keʃidæn ‘smoke cigarettes’, lit. ‘cigarette pull’), (ii) in combination with a noun denoting a building or another type of built object, the LVC means ‘to build N’ (e.g. jâde keʃidæn‘build a road’. lit. ‘road pull’; divâr keʃidæn ‘build a wall’, lit. ‘wall pull’). The existence of such LVC families, as Nunberg et al. claim, would be surprising, if the LVCs were not the outcome of a compositional process. In the current paper, we will not go into the discussion of LVC families (see Family 2006, 2011, Samvelian & Faghiri 2014 for that issue), but rather only concentrate on the modification of LVCs.

The issue of internal modification is related to the debate of whether LVCs have lexical or phrasal properties (see, for example, Ghomeshi & Massam 1994, Goldberg 1996 and Müller 2010 for this topic). Light verb constructions licensing internal modification are called ‘separable LVCs’ since a lexical element separates the two components of the LVC. In the Persian example in (6), an adjective separates both elements and crucially, the adjective does not take part in the formation of the complex predicate. The complex predicate is sedâ dâdan ‘produce a sound’ and boland ‘loud’ is used as an attributive modifier of the preverb sedâ ‘sound’. Its status as an attributive modifier is clearly indicated by the ezâfe-construction, which is triggered by attributive modification, but not by adverbial (we will discuss this issue in Section 4). Thus, we have attributive modification within an LVC, rather than an LVC consisting of two preverbs (a nominal and an adjectival one) and a light verb. Evidence for this view is that sedâ ye bolandi ‘a loud sound’ always receives the same interpretation, irrespective of whether sedâ functions as the preverb of an LVC or is used outside of one.

Non-separable LVCs, such as bayân kardan ‘to state’ (lit. ‘expression do’), do not license an attributive modifier of the nominal preverb. Separable LVCs show a greater syntactic flexibility than non-separable ones, as elements like modifiers can be placed between the preverb and the light verb. LVCs licensing internal modification are always separable and therefore display phrasal properties (e.g. Müller 2010 and references therein). Thus, compositionality of LVCs (in the sense of Nunberg et al. 1994) goes along with displaying phrasal properties.

Karimi-Doostan (2011) argues for a relationship between the separability of Persian LVCs and the respective type of nominal preverb. He distinguishes three types of nominals, as shown in Table 1. An eventive noun like taxrib ‘destruction’ refers to an eventuality – the destruction of something – and displays nominal features. As (7a) reveals, the noun is compatible with a demonstrative determiner. Verbal nouns (e.g. anjâm ‘performing’) also refer to eventualities – the performance of something – but do not show nominal features as they are not compatible with demonstrative determiners (7b).

Table 1 Classification of Persian nouns (based on Karimi-Doostan 2011).

Karimi-Doostan mentions further properties with respect to which verbal nouns differ from ‘ordinary’ nouns: verbal nouns cannot be pluralized, they cannot be selected by determiners and preposition and they cannot function as subject or object (Karimi-Doostan 2011: 83). One property verbal nouns share with non-eventive nouns is that they license attributive modifiers (8), at least as long as they are not realized as the preverb of an LVC. Non-eventive nouns, finally, do not refer to an eventuality – rather they refer to (abstract) objects – and do show nominal features. ^[3] An example is guš ‘ear’, which licenses a demonstrative determiner (7c) but clearly does not refer to any eventuality. In Section 3, we turn to a deeper discussion of the nominal morphology licensed by preverbs.

Fabregas & Marín (2012: 36) argue that the eventuality denoted by eventive nouns can be located either in space or time. Thus, such nouns can be subject to predicates like English take place or its Persian correspondent surat gereftan. In addition, eventive nouns take temporal or aspectual modifiers like English constant. We make use of the first criterion to illustrate that Karimi-Doostan’s eventive nouns and verbal nouns are eventive. ^[4] In (9) and (10) the nouns taxrib and anjâm are realized as the subject of the predicate surat gereftan which temporally locates the eventuality denoted by them. The noun guš ‘ear’, on the other hand, cannot be subject of that predicate and therefore it cannot be temporally located. Thus, guš does not refer to an eventuality.

Karimi-Doostan claims that a nominal preverb is separable if it is an eventive noun (i.e. if it refers to an eventuality and shows nominal features). The reason he proposes is that eventive nouns independently refer to an eventuality and therefore are more independent from the light verb than non-eventive nouns are. Following Karimi-Doostan’s (2011: 91) analysis, verbal nouns cannot be separated from the light verb since they cannot function as a lexical head of a DP. Karimi-Doostan relates separability basically to the nominal type of the preverb, but not to the LVCs’ compositionality. If he is right, it follows that only LVCs which have an eventive noun as preverb are compositional in Persian. The separable LVCs investigated in the current paper do have eventive nouns as preverbs. But we would like to be cautious with respect to the claim that only eventive nouns can be separated from the light verb. One particular reason is that Mohammad & Karimi (1992: 197) present an example in which a non-predicative nominal preverb is separated from its light verb by the future tense auxiliary xâhad (11). The auxiliary is placed immediately in front of the main verb, which results in the separation of preverb and light verb in examples like (11).

The example in (11) demonstrates only that LVCs like guš kardan ‘listen’ (lit. ‘ear give’) also show some degree of syntactic flexibility, which indicates that the LVC does not function as a single word. The example does not show that the different components of the LVC have an identifiable meaning since xâhad does not modify one of the LVC’s components individually. This shows that not every element intervening between the two components of an LVC can be counted as evidence for the compositional nature of that LVC. In the remainder, we only conceive of LVCs as being compositionally derived if they can be separated by internal modifiers. A question, which we cannot address within the current paper, is, whether separability and compositionality always go hand in hand or whether some LVCs are separable (by e.g. the future auxiliary) but not compositionally derived. In the next section, we will keep discussing functional morphology used within an LVC, but therein turn to grammatical categories of the nominal acting as preverb.

3 Preverbs and nominal morphology

In the discussion of the separability of LVCs, it is usually mentioned that functional morphemes such as the negation marker ne-, the imperfective aspect marker mi-, auxiliaries, modals and object clitics can be placed between the preverb and the light verb. One example (11) of the future auxiliary separating the preverb and the light verb was shown in the previous section. These functional morphemes realize grammatical categories of the verb like tense, aspect and modality, but also agreement. Such data figure crucially into the discussion of the phrasal status of LVCs (Müller 2010). The debate usually focusses on functional morphology realizing grammatical categories of the verb, but the preverb licenses its own functional morphology as well. The current section presents a brief discussion of some aspects of Persian nominal morphology and how it is employed in LVCs.

In Persian, nouns can be marked for number (singular vs. plural), indefiniteness and case. Before turning to a discussion of nominal preverbs, we will briefly mention some general characteristics of Persian nouns. In contrast to languages like English and German, all Persian nouns can be easily used bare. Bare nouns in object position are transnumeral (e.g. Lazard 1992: 62; Wiese 1997: 137; Ghomeshi 2008: 90f), which means that count nouns, which are not marked for plurality, allow (at least in certain contexts) a plural interpretation (12a). Depending on the context, a bare noun can be non-referential, which frequently gives rise to pseudo-incorporation (Dayal 2011, Modarresi 2015). In such cases, the noun acts like an event modifier (12a). We will argue later in Section 5.1 that pseudo-incorporation and LVCs result from different types of compositional processes.

Adding a plural marker to a noun results in an unambiguously referential and plural interpretation (13), e.g. Ghomeshi (2003: 58).

Persian has two indefinite articles, a free form yek – identical to the numeral for ‘one’ – and a phrase-final suffix -i. The two markers can be used alone ((14a) and (b)) but they can also co-occur within the same NP (c).

Lyons (1999: 90f) argues that yek as well as -i are quasi-indefinite articles, which indicate cardinality. Ghomeshi (2003: 60, 65), on the other hand, posits that -i does not express cardinality, whereas yek can do so. Since yek is identical to the numeral ‘one’, the expression of cardinality is not surprising. Support for Ghomeshi’s view that -i does not express cardinality is its compatibility with plural morphology (15).

Ghomeshi (2003: 65) – similarly Paul (2008: 322) – shows that -i and yek do not have the same distribution, since -i can, for example, be used in negative contexts, whereas yek cannot (16). In (16a), it is negated that Ali bought any car, whereas in (b) yek is interpreted as the numeral ‘one’ rather than an expression of indefiniteness. It is negated that Ali bought just one car, rather, as (16b) shows, he bought several. A specification that Ali bought more than one car is only compatible with the use of yek (16b) but not with -i (16a).

A second relevant difference between yek and -i is that only the first, but not the second, can be used with generic nouns (Paul 2008: 314, 322). The sentence in (17) requires a generic interpretation of the noun morabbaʔ ‘square’. Yek, although it is optional, can be used with the noun, whereas -i cannot.

Ghomeshi (2003: 63) claims that -i marks referentially specific nouns. Such an analysis is supported by (18a), in which -i is suffixed to the postnominal adjective modifying the noun zan ‘woman’. The noun refers to a specific woman which left Ahmad.

As (18b) shows, -i can also be used in referentially opaque contexts. In (18b), Ahmad wanted to marry a specific woman but could not find her. The specific woman Ahmad wanted to marry does not need to exist at all and therefore zan e puldâr-i ‘a [specific] rich woman’ is not (necessarily) referential. In this particular example, -i signals epistemic specificity (for a general discussion of the different types of specificity see, e.g. von Heusinger 2011). ^[5] We treat -i as a marker of specificity, in line with Ghomeshi (2003), but reject the view that the respective noun has to be referential. Further support for the analysis that the noun marked by -i does not need to be referential is the fact that -i is also used under negation (cf. example (16)). In the remainder, we keep glossing -i as expressing indefiniteness but meaning ‘indefinite specific’. As far as we can determine, a standard analysis of indefiniteness in terms of choice functions is compatible with the Persian data. For details of such an analysis, we refer the reader to the relevant literature (e.g. Heim 1982; von Heusinger 1997, 2000, 2011).

Turning now to nominal preverbs, there is some debate about whether they are always indefinite and non-referential (e.g. Mohammad & Karimi 1992, Goldberg 1996, Megerdoomian 2012), or whether they can be referential and even definite (e.g. Karimi-Doostan 2011: 74). Nominal preverbs of LVCs are not different from other nouns in licensing functional morphology. As (19) shows, plural marking of nominal preverbs is admissible. The use of the plural marker -hâ on sedâ ‘sound’ results in the interpretation that the car is not only producing one (type of) sound but different (types of) sounds.

Also shown in (19) is that the nominal preverb sedâ ‘sound’ licenses the specificity marker -i. It has the same effect on nominal preverbs as it has on other nouns: The preverb receives a specific and sometimes also referential interpretation. The referentiality of sedâ-i is shown in (20). Adding the specificity marker -i to the nominal preverb licenses a discourse anaphor referring back to the preverb’s referent (20a). Without -i, the preverb does not license such an anaphor (20b). One could assume that the anaphor refers back to the event rather than to the referent of sedâ. But then, the licensing of the anaphor should be independent of the presence of the specificity marker, which – as (20) shows – is not the case. Thus, we have reason to claim that the anaphor truly does pick up the referent of sedâ-i and the nominal preverb is referential.

A further piece of evidence showing that the anaphor refers back to the preverb is shown by the agreement facts in (21). The preverb takes the plural marker, and the verb in the second sentence shows plural agreement. Plural agreement can only be triggered by the plural marker on the preverb since no other nominal component is marked for plurality.

The difference in specificity in (20) is corroborated by the fact that sedâ-i allows adding a specification on the type of sound produced within the event (22a), whereas sedâ does not (b).

It seems that modified preverbs preferentially take the indefiniteness suffix -i, although it is not mandatory, as seen in (23). Without -i, a generic reading obtains. A functional motivation for the use of the indefiniteness marker in the context of modification is that the modifier narrows down the reference of the nominal concepts, which very likely results in a specific interpretation of the modified noun.

Persian does not have a grammaticalized definite article, but one way of expressing definiteness is by the use of a demonstrative determiner. ^[6] As (24) shows, a nominal preverb can take a demonstrative determiner. The LVC is be natije residan ‘to conclude’ (lit. ‘to conclusion arrive’), the preverb is a PP and the determiner is positioned in between the preposition and its complement. The sentence in (24) expresses that the government arrived at a unique conclusion, which licenses the demonstrative’s use.

A further example is given by Karimi-Doostan (2011: 89) and shown in (25). The preverb râhnamâʔi ‘advice’ is modified by the demonstrative. In addition, the preverb takes the accusative case marker -râ. Persian displays definiteness-based differential object marking and basically restricts the case marker -râ to direct object arguments which are conceived as having definite reference (see Bossong 1985, Lazard 1992, Ghomeshi 1997b; Aissen (2003) shows that definiteness is not the only feature relevant in determining the use of -râ).

The occurrence of the case marker allows us distinguishing between NPs functioning as preverbs and NPs functioning as the object of a verb. As we will show below, -râ does not show up on preverbs – as also stated by Megerdoomian (2012: 194) – and therefore in râhnamâʔi-râ kardan ‘give this/the advice’ is not an LVC. The native speakers we consulted also expressed the view that in râhnamâʔi-râ kardan ‘give this/the advice’ is not a complex predicate. The reasoning is based on the native speakers’ intuition that in râhnamâʔi-râ kardan ‘give this/the advice’ has a different predicational meaning than râhnamâʔi-râ kardan ‘give (an) advice’. A somewhat clearer example is given in (26), which indicates that the use of the accusative marker requires a non-light interpretation of the verb. Šir dâdan, in (26a), is an LVC meaning ‘to breastfeed’. The LVC puts selectional restrictions on the subject argument, as only women can breastfeed a child. If šir is case marked, the predication as well as the selectional restrictions on the subject referent change. In the second sentence, dâdan ‘give’ is used as a heavy verb, the interpretation is simply that the father gives the milk to his daughter, but not that he is (necessarily) feeding the milk to her. Thus, šir functions as a preverb of the LVC šir dâdan in (26a) but it is the object of the verb dâdan in (b).

That (26b) does not mean ‘feeding the daughter with milk’ is indicated by the example in (27). In this sentence, it is explicitly stated that the father gave the milk to the daughter, but she did not drink it; rather using it for making a cake. A similar example based on the sentence in (26a) is odd, which is expected given that šir dâdan means ‘to breastfeed’.

Thus, in difference to, for example, Vahedi-Langrudi (1996), we do not analyze the preverb as being an object of the light verb for the following reasons: First, use of -râ enforces a heavy interpretation of the verb (as discussed above). Second, an element other than the preverb takes the accusative marker in examples like (28). This – as also mentioned by Megerdoomian (2012: 198) – indicates that došmæn ‘enemy’ (a) and tækalif ‘homework’ (b) function as direct objects of the respective LVCs. ^[7]

Third, the preverb does not become the subject under passivization (29). Sentence (29a) is the passivized form of (26a), whereas (29b) is the passivized form of (26b). Šir dâdan is used as an LVC in (26a) and accordingly, the nominal preverb does not become the subject of the passive sentence. In (26b), dâdan is used as a heavy verb and šir is its direct object. In this configuration, šir does become the subject of the passive sentence.

The preverb is also not an indirect object (in opposition to a claim made by Tabaian 1979: 199), since those require the preposition be ‘to’ (30).

We are now in a position to give a summary on the behavior of nominal preverbs with respect to nominal functional morphology. Nominal preverbs behave like other nouns in the expression of number and (in)definiteness. A crucial difference is only found with respect to case marking, which can be explained by the fact that nominal preverbs cannot be realized as the object of the verb. This aspect is crucial for the analysis of the semantic composition of LVCs to which we turn in Section 5. In the next section, we turn to the attributive modification of nominal preverbs, which we will compare to adverbial modification of LVCs.

4 Modification

Persian has a lexical class of adjectives and most adverbials are either overtly or non-overtly (zero derivation) derived from them. Examples of adjectives and corresponding adverbials are given in (31). The adverbial šadidan ‘severely’ is derived from the adjective šadid ‘severe’ by the addition of the adverbializer -an. In the case of boland, the adverbial is not overtly derived from the adjective. We speak of šadidan as a derived adverb, but assume that boland is an adjective, irrespective of whether it functions attributively or adverbially. Thus, we clearly take the syntactic function (attributive vs. adverbial modifier) and lexical class (adverb vs. adjective) as distinct. Nevertheless, we propose different semantic representations for the attributive and the adverbial use of adjectives, as we will make clear below.

Attributively used adjectives are usually realized postnominally and trigger the ezâfe-morpheme. The ezâfe-morpheme is a linking element, which links the modifier and its head (32a). ^[8] Adverbial modification, on the other hand, does not require the ezâfe-morpheme and the adverbial – irrespective of whether it is a (derived) adverb or an adjective – mostly precedes the LVC (32b).

For the analysis of the ezâfe-morpheme, we follow Ortmann (2002: 66) in viewing it as an overt realization of the operation of argument extension (Wunderlich 1997: 98). ^[9] The abstract representation for argument extension is shown in (33). arg (for argument extension) operates on a noun and introduces a further predicate which shares its argument with the nominal predicate.

The ezâfe-morpheme is a language-specific instantiation of the argument extension operator. ^[10] The semantic composition for a simple example like sag-e bozorg ‘big dog’ (lit. ‘dog-ezâfe big’) is shown in (34). The combination of the noun with the adjective – after argument extension – proceeds via function composition.

Whereas morphology speaks in favor of argument extension in case of intersective attributive modification, the morphological evidence points to a different licensing process as far as adverbial modification is concerned. Adverbial modifiers – at least in some cases – are overtly derived from non-adverbially used adjectives. Thus, adverbial modification is licensed by the process of modifier extension (Wunderlich 1997: 98). Modifier extension is similar to argument extension, but operates on a modifier and introduces a predicate variable, which shares an argument with the modifier (35). In the process of creating an adverbial modifier – by adding the mod operator – the individual argument is converted into an event argument. In cases of zero derivation, there is no morphological exponent of the process and we simply assume that it takes place if two expressions can otherwise not combine (see Wunderlich 1997: 97).

For illustration, we use the example amiq nafas kešid ‘breathe deeply’ from (32b). In (36), the adjective amiq is converted into an adverbial modifier by the covert process of modifier extension, and after that combines with the LVC nafas kešid. Note that for the moment, we left out the individual argument of the LVC to keep the representation as simple as possible. But we will include the individual arguments in the later representations.

Besides the different licensing of attributive and adverbial modification, we also assume different semantic representations for attributive and adverbial modifiers. There is, for example, no morphological difference between the adverbial and attributive use of boland; nevertheless, we propose two semantic representations. There are two reasons for this assumption: (i) Not all LVCs which take a certain adverbial modifier license the corresponding attributive modifier, and (ii) attributive and adverbial modifiers contribute differently to the predication. (i) can be illustrated by the examples in (37). Eʔterâz kardan ‘to protest’ licenses boland as an adverbial (37a), but not as an attributive modifier (b).

Generally, eʔterâz kardan is a separable LVC, as (38) shows. The use of boland as an attributive modifier in (37a) is prohibited due to the fact that the noun eʔterâz ‘protest’ does not license it as such, irrespective of whether the noun is used as the preverb of an LVC or not. Thus, the modification construction eʔterâz e bolandi ‘loud protest’ is invalid.

The second point, that the adverbial and attributive modifiers contribute differently to a predication, can be seen by comparing (39a) and (b). The property specified by sarii ‘fast’ in (39a) is speed and it can be said that speed is a property of the individual denoted by xodro ‘car’. But speed is not a property of the event denoted by kâr kardan ‘work’ (39b). This can be illustrated by the fact that (39b) cannot be paraphrased as ‘the event of working was fast’. Thus, speed is not a property of the event directly but it relates to the manner of working. The sentence in (39b) can be paraphrased as ‘he worked in a fast manner’.

An appropriate semantic representation for attributive sarii is given in (40a), whereas the semantic representation for the adverbial use is given in (b).

The semantic representations in (40) differ from those given above; we analyze sarii not as a predicate of an individual or an event, but rather as a measure function assigning an individual (or event) a value with respect to a certain property. The relevant property in case of sarii is speed and we represent measure functions by small caps in the semantic representation. In (40a), speed is a property of an individual, whereas in (b) speed is not a property of the event. Rather it is related to the event by some mediating function f. ^[12] Thus, adverbial sarii comes with a functional argument f which has to be contributed by the event description. An appropriate function could for example be the attribute manner, if adverbial sarii specifies the speed of executed activity. An analysis of sarii kâr kardan ‘work fast’ is sketched in (41). The activity predicate has a manner component, which is analyzed as ‘manner(e)=working’; the value ‘working’ is clearly a shortcut for the actual analysis of the manner component, but it is sufficient for illustrating this instance of adverbial modification (see Morzycki 2016: Chapter 5.4 for a discussion of the notion of ‘manner’). manner is a mediating function between the event and speed, and saturates the f argument.

To keep attributive and adverbial sarii as similar as possible, we also introduce a mediating function f in case of the attributive modifier – $\unicode[STIX]{x1D706}$ f $\unicode[STIX]{x1D706}$ x(speed(f(x))=high) – and assume that f can simply be the identity function (e.g. Partee, ter Meulen & Wall 1990: 34). For reasons of simplicity, we ignore the functional argument in cases where it is instantiated in this way.

After discussing the modification strategies, we now turn to the interpretation of adverbial and attributive modification of LVCs. We only focus on modification constructions where an LVC licenses one and the same adjective both as attributive and as adverbial modifier. This, for example, has been illustrated for nafas kešidan ‘breathe’ shown in (32) and is repeated for convenience in (42). With the adjective amiq as a modifier of the LVC nafas kešidan, attributive modification (42a) and adverbial modification (b) result in the same interpretation. Both sentences in (42) convey the meaning that the subject referent took a deep breath.

A further example exemplifying this interpretation pattern is shown in (43). Irrespective whether surii is used adverbially (43a) or attributively (b), it is specified that the marriage was an official one. Which, for example, consists in signing of the relevant documents.

Megerdoomian (2012: 196) mentions the same fact, but states that attributive modifiers always modify the whole LVC. According to Megerdoomian, the attributive constructions in (42a) and (43b) thus receive an interpretation as adverbial modifiers as well. However, this is too simplistic, since the attributive modification of a nominal preverb can result in a different interpretation than the adverbial modification of the whole LVC. In (44a), the adjective xub is used adverbially and precedes the LVC mesâl zadân ‘give an example’ (lit. ‘example hit’). The sentence means that the teacher gave the example in a good way, so it is the manner of giving the example which is specified by xub. In (44b), on the other hand, xub is used as an attributive modifier of the preverb mesâl ‘example’, and it specifies the quality of the example but not the manner of giving it.

That the two sentences in (44) have different interpretations is shown in (45). Adding a subordinated sentence, expressing that the manner of giving the example was bad, results in a contradiction for the sentence in (45a), but it is non-contradictory for the sentence in (b).

We take this as clear evidence that attributively used adjectives truly act as attributive modifiers and do not – contrary to Megerdoomian’s assumption – function as adverbial modifiers. Following our analysis, adverbial and attributive modifiers have different scope. The modifier’s scope depends on its syntactic use (46).

If used as an adverbial modifier, the adjective has scope over the whole LVC. In its attributive use, it has scope over the preverb only. This is in line with the syntactic fact that the attributive adjective is licensed by the noun it modifies via the ezâfe-morpheme. Licensing of a modifier is possible if head and modifier are semantically compatible. In the examples under discussion, the attributive modifiers behave as internal modifiers (in the sense of Ernst 1981) and modify a single component of the LVC rather than the LVC as a whole. This gives support for a compositional analysis – in the sense of Nunberg et al. (1994) – for the LVCs under discussion.

Based on the brief discussion above, we arrive at two different possible interpretation patterns for adverbial and attributive modification of separable LVCs (with nominal preverbs):

1. (a) The attributive and the adverbial modifier can result in the same interpretation;

2. (b) the attributive and the adverbial modifier can result in different interpretations.

Since an attributive modification of the nominal preverb is possible, the attributive adjective needs to combine with the noun before the LVC is composed. Syntactically, this requires an analysis in which the complex NP, which functions as the preverb, is build up before it combines with the light verb. A syntactic analysis of Persian LVCs compatible with that requirement is presented in Müller (2010). We do not aim at providing a syntactic analysis of the data under discussion but turn to a deeper discussion of the compositional semantics of Persian LVCs. In the next section, we propose a compositional analysis of Persian LVCs, based on which we will show how the patterns in (a) and (b) are achieved in Section 6.

5 Composing LVCs

The last section has demonstrated that attributive modifiers function as internal modifiers and target the preverb only. Nunberg et al. (1994: 503) propose that modification of just a part of an idiomatic expression is ‘powerful evidence that the pieces of idioms have identifiable meanings’ which interact with each other. Thus, the modification data gain evidence for a compositional analysis of those LVCs, which license internal modifiers.

Light verb constructions have been extensively analyzed from a syntactic perspective within different grammatical frameworks. Among the questions addressed in this strand of work are: (i) whether and how the argument structure of the complex predicate is derived from its components (e.g. Grimshaw & Mester 1988, Butt 2010, Müller 2010), and (ii) whether LVCs show lexical or phrasal properties (e.g. Goldberg 1996, Müller 2010). The semantic composition of LVCs has only rarely been addressed explicitly in the semantics literature. Notable exceptions come from work on the composition of the complex predicate’s event structure (e.g. Karimi 1997, Folli, Harley & Karimi 2005, Pantcheva 2009). The analysis presented in the current section is only partial as it leaves out some aspects (e.g. event structure and argument structure) which are not crucial for our examination of the attributive and adverbial modification in Section 6. The schematic analysis in Section 5.1 is applied to the formation of LVCs expressing events of substance and sound emission in Section 5.2. This set of LVCs will serve as a case study in Section 6.

5.1 The semantic composition of light verb constructions

Various authors working on Persian (e.g. Ghomeshi & Massam 1994, Vahedi-Langrudi 1996, Ghomeshi 2008) do not distinguish between LVCs and other types of complex predicates. The mentioned authors explicitly claim that the two constructions in (47) are of the same type.

Šafâ dâdan ‘to cure’, in (47a), is an LVC. In (47a), dâdan ‘give’ does not denote a process of giving, rather the main predicational content is provided by the noun šafâ ‘cure’. The light verb adds information to the main event predication (Butt & Lahiri 2013: 23). (47b), on the other hand, is an instance of pseudo-incorporation. The verb xordan ‘eat’ retains its full lexical meaning in (47b), qazâ ‘food’ functions more like a modifier of the verb.

LVCs and pseudo-incorporation have in common that the grammatical head of the construction is a verb. It is either the incorporating verb in case of pseudo-incorporation or the full verb in light use in case of an LVC. The verbs take, for instance, tense, aspect and agreement morphology. Thus, the two constructions are superficially syntactically similar. ^[13] The crucial difference between LVCs and pseudo-incorporation is that the main predication is contributed by the noun in (47a) but by the verb in (b). From a semantic perspective, the two constructions cannot be analyzed analogously. Chung & Ladusaw (2004) argue that functional application is not the right mode of composition in case of incorporation constructions. Instead, they propose a compositional mode they call ‘restrict’. ‘Restrict’ is defined as an operation, which composes a predicate with a property denoting expression. The result of this process is that the domain of the original function denoted by the predicate is restricted to a subdomain. (48a) shows the composition of the predicate feed and a predicate denoting indefinite noun. ‘Restrict’ is a non-saturating process, the y argument is saturated by existential closure (48b). The meaning of (48b) is that John fed a dog or that John was dog-feeding.

Such an analysis does not work for LVCs. The nominal preverb of an LVC does not restrict the domain of the predicate represented by the light verb. An analysis for šafâ dâdan ‘to cure’ in terms of ‘restrict’ would require that šafâ ‘cure’ provides a restriction on the domain of the function denoted by dâdan ‘give’. But this is not the case as šafâ dâdan does not denote a specific type of giving.

Our compositional analysis of LVCs builds on Butt & Geuder’s (2001) examination of LVCs in English and Urdu. The authors mainly assume that light verbs differ from their corresponding heavy verbs in not being able to denote events of their own. One piece of evidence supporting this view is that light verbs cannot be used to refer anaphorically to an LVC. Thus, light verbs cannot be used to pick up an event introduced before. In (49), the light verb dâdan ‘do’ cannot be used to refer back to the LVC safâ dâdan ‘to cure’. If a light verb is used outside of an LVC, as in the second sentence in (49), it is interpreted as a heavy verb.

Butt & Geuder conceive of light verbs as being semantically similar to modifiers. Within the framework of neo-Davidsonian event semantics, Butt & Geuder treat light verbs as predicates of events, which are conjoined with the main event predication. The analysis sketched by Butt & Geuder (2001: 356) is shown in (50). The eventive noun wash introduces an event predication, whereas the light verb give contributes an event predicate that is conjoined with the main event predication. In the case of light give, Butt & Geuder (2001: 356) postulate the event predicate GIVE-TYPE(e), which is intended to cover the meaning contributed by the light verb. The event predicate GIVE-TYPE(e) can be expanded as outlined in (50b).

The semantic contribution of light verbs is often very subtle and a given light verb shows a high degree of meaning flexibility. Thus, the exact semantic contribution a light verb makes is determined – following Butt & Geuder (2001: 356) – through contextual factors. Nevertheless, the meaning contributed by the light verb should somehow be constrained by the meaning of the corresponding heavy verb. Butt & Lahiri (2013: 23) propose that a light verb and its heavy verb are synchronically united within a single underspecified lexical entry. The authors propose that the light verbs predicate ‘a subset of lexical semantic information associated with the main verb’, thus meaning components active in the verb’s heavy use are deactivated in the light use of a verb. ^[14] We argued above that the light verb does not have the same meaning than its corresponding heavy verb since the light verb does not denote an event of its own. However, it is a still unresolved question which meaning components – associated with the heavy verb – are deactivated in a particular light use and which not. Answering this question requires a deep going investigation of the meaning contributed by a light verb (in a particular LVC) and the meaning of its corresponding heavy verb. This question goes beyond the limits of the current paper but we turn to a discussion of the semantic contribution of some Persian light verbs in the next section.

We adopt the view that ‘[l]ight verbs make no independent reference to a class of events’ (Butt & Geuder 2001: 358) and require the composition with an event-denoting expression to yield a full-fledged event description. The eventive noun, as the main predicational component of the LVC, provides the event predicate. (51) shows the basic semantic representation we propose for a light verb like dâdan ‘give’.

The light verb does not denote an event of its own, which we cover by the fact that it also introduces a property variable P, representing a property of events. The variable needs to be saturated by an eventive expression, which is contributed by the preverb. As the preverb specifies the event property, it does not end up as an object of the LVC, which is in line with the grammatical facts discussed in Section 3. Furthermore, we assume that the light verb also introduces an individual argument, but the exact thematic relation between the event denoted by the verb and the individual argument is underspecified. tr is a variable for an attribute of the type ‘thematic role’. As we show in the next section, the variable is saturated within the compositional process. The lexical contribution of the light verb is – in line with Butt & Geuder’s account – represented as a ‘GIVE-TYPE’-event predicate. A more detailed analysis of particular LVCs should be able to provide a more nuanced representation of the light verb’s lexical contribution.

We present a detailed discussion of the composition of the light verb and its nominal preverb after having introduced the relevant data in the next subsection.

5.2 Case study: Verbs of emission

The current subsection focusses on Persian LVCs denoting the emission of substances (e.g. blood) or sounds. The focus is on this particular class of predicates as they are well analyzed and we can thus build on previous work.

Verbs of emission, quite generally, denote a process of emitting a stimulus out of an emitter. The stimulus can either be a smell (e.g. English smell, German duften ‘emit a pleasant smell’), light (e.g. sparkle, light), a sound (e.g. drone, bark) or a substance (e.g. bleed, fester). Verbs of emission do not only vary concerning the emitted stuff but also with respect to aktionsart. Following Rappaport Hovav & Levin (2000: 283), English verbs of smell and light emission are rather stative-like, whereas verbs of sound and substance emission are more dynamic (see Fleischhauer 2016b: Chapter 7 for an extensive discussion of German verbs of emission). Emission verbs are usually, at least in English and German, intransitive and their single argument is the emitter, which is the entity emitting the stuff. Only a few verbs of emission allow a transitive use, such as English clatter as in I clattered the tea-cups (Potashnik 2012: 263; see also Levin & Rappaport Hovav 1995 on the issue of transitivity of emission verbs). The transitive uses have a causative interpretation and the added argument is interpreted as a causer of the emission process. The emittee, which is the emitted stimulus, is – at least in English and German – semantically incorporated into the verb (Goldberg 2005: 20ff). Many verbs of emission – especially verbs of substance emission – are derived from nouns denoting the emitted stimulus (blood – bleeding, rain – to rain).

The various processes of emission are denoted by LVCs in Persian. Two such examples are shown in (52) and as they reveal, there is not a single light verb uniquely related to the expression of emission processes. Like in English and German, the single argument of the predicates is the emitter, whereas the emittee – the sound or substance emitted – is contributed by the preverb.

Our analysis of the two LVCs in (52) will be only programmatic and some aspects need to be fleshed out in much more detail in future work. Especially the composition of argument structure deserves closer discussion but cannot be done within the current paper. Before we proceed to a compositional analysis of the examples, it needs to be noted that the nouns sedâ and xunrizi also combine with other light verbs forming semantically similar LVCs. Examples of different LVCs using these two nominal preverbs are given in (53) and (54). The LVCs differ in various respects. The examples in (53e) and (54d) are causative, whereas the others are not. The light verb oftâdan ‘fall’ contributes the meaning component of ‘suddenness’ and ‘unexpectedness’ to the LVCs in (53d) and (54c). The LVCs in (53a) and (53b), on the one hand, and those in (54a) and (54b), on the other, seem to be very similar to each other. Focusing on the first two, the LVC means ‘to produce a sound’. But the two show subtle semantic differences, as shown in (55). They differ with respect to the licensing of the adverbial amdan ‘intentionally’, showing that only sedâ kardan (lit. ‘sound do’) can be used for situations in which a sound is intentionally produced, but sedâ dâdan (lit. ‘sound give’) cannot.

The examples in (53) and (54) indicate that the choice of the light verb affects the meaning of the resulting LVC and therefore it is reasonable to conclude that the light verb makes a semantic contribution to the complex predication. In the compositional analysis, we represent kardan ‘do’ and dâdan ‘give’ simply as DO-TYPE(e) and GIVE-TYPE(e) respectively. This is merely a shortcut for the meaning contributed by the respective light verbs.

Our analysis of compositional LVCs starts with xunrizi kardan ‘to bleed’. We basically assume that the Persian LVC has the same semantic representation as the corresponding English or German verbs to bleed and bluten respectively. The representation of German bluten ‘bleed’, taken from Fleischhauer (2016b: 254), is shown in (56). The verb is decomposed into an event predicate ‘emit’ and two thematic roles, the emitter and the emittee. Small caps are intended to represent functional attributes. A functional attribute assigns a unique value to the bearer of the attribute (see Löbner 2014: 26). ^[15] The y argument – the emittee – is existentially bound, as it is lexically specified as being blood.

With respect to Persian, the crucial question is how we arrive at a semantic representation like the one in (56) from the two components xunrizi ‘bleeding’ and kardan ‘do’. A representation for the light verb kardan ‘do’, based on our discussion in Section 5.1, is shown in (57).

That xunrizi is eventive is indicated in (58), which shows that the nominal referent can be temporally located. Thus, xunrizi refers to an event of bleeding. The emitter can be realized as an adjunct to the nominal (xunrizi ye jân ‘bleeding of John’), as shown in (58) as well.

The emittee – blood – cannot be realized as an adjunct; a construction such as xunrizi ye xun ‘bleeding of blood’ is odd. This is due to the fact that the emittee is already lexically specified. Support for this view is gained by the fact that a cognate object construction comparable to English The insect bleeds green blood is judged as being unacceptable by Persian native speakers (59).

We propose the semantic representation in (60) for the eventive noun xunrizi. The referential argument is e, whereas the emittee argument is existentially bound. The emitter argument is optional, it can be realized – in which case it requires the ezâfe-morpheme – but can also be left out. In such a case, we assume that it becomes existentially bound.

In the composition of kardan and its nominal preverb xunrizi, the preverb saturates the predicate variable (61). As the eventive noun only provides one suitable argument – the emittee argument is lexically specified – there is no question of which argument gets unified with the tr argument of the light verb. It is this argument which gets unified with the thematic role argument of the light verb. ^[16] Since the tr argument of the light verb is non-optional, the emitter becomes a non-optional argument of the LVC as well and finally gets realized as its subject argument.

Our second example is the LVC sedâ dâdan ‘produce a sound’. Sedâ ‘sound’ is also an eventive noun, as (62) indicates. It seems that the noun does not only refer to the emitted sound but also denotes the process of emitting it.

The composition of sedâ ‘sound’ and dâdan ‘give’ (63) works like the composition of xunrizi kardan sketched above. Sedâ dâdan and xunrizi kardan mainly differ with respect to the emittee argument and the lexical contribution of the light verb, sketched as GIVE-TYPE(e) in the representation. The resulting semantic representation for the LVC is shown in (63).

Having discussed the relevant background on verbs of emission and the compositional derivation of the relevant verbs, we turn now to an analysis of the modification constructions in the next section.

6 Modified light verb constructions

In Section 4, we showed that two different interpretational patterns with respect to modification arise. In the first, adverbial and attributive modification result in the same interpretation, whereas in the second, different interpretations are obtained. We start our discussion with the first pattern in Section 6.1 and turn to the second one in Section 6.2.

6.1 Getting the same interpretations

To illustrate the pattern of achieving the same interpretation for adverbial and attributive modification, we use the sound emission LVC sedâ dâdan ‘produce a sound’ in (64). In the examples, the adjective boland ‘loud’ is used as a modifier; in (64a) this use is attributive and in (b) it functions adverbially. Both sentences have the same interpretation: The sound produced by the subject referent is loud. Both sentences can be used to describe the same situation: Someone was holding a cup and let it fall down onto the floor. Boland indicates the loudness of the sound produced by the cup crashing on the floor. In its adverbial use, boland does not indicate the loudness of the action producing the sound. Such a scenario makes it obvious that adverbial boland does not specify the loudness of the sound’s production, but only that of the resulting sound.

We start with discussing the attributive use of boland. The attributive modifier combines with sedâ before the LVC is formed. The semantic composition of sedâ ye bolandi is shown in (65). ^[18] In (65a), the representation of the eventive noun sedâ is provided. In the next step, the process of argument extension is applied. This introduces a predicate variable Q which is saturated in the third step by boland. The adjective introduces the loudness specification and the resulting representation for the attributive modification construction is shown in (65c). ‘high’ is used for indicating the value of loudness, meaning that it is a context-dependent high degree of the respective property. Boland modifies the loudness of the emittee since (a) as an attributive modifier it predicates about an individual rather than an event and (b) the only obligatory individual is the semantically incorporated emittee.

The LVC is now formed by the combination of the light verb dâdan and the modified preverb sedâ ye boland. The composition of the (modified) preverb and the light verb proceeds as described in the last section, resulting in the representation shown in (66).

The adverbial modification by boland proceeds differently, as adverbial boland modifies the whole LVC and not just the preverb. Before the adverbial modifier applies to the LVC, the process of modifier extension is required. Modifier extension of boland has already been discussed above but is repeated in (67) to sketch the complete compositional process.

After modifier extension, adverbial boland can combine with sedâ dâdan, which saturates the predicate argument of the modifier (68). The functional argument f – mediating between the loudness attribute and the event – is saturated by the emittee attribute.

Comparing (66) and (68) shows that in both cases, the same interpretation results as loudness is an attribute of the emittee. The way the interpretation is achieved differs in both sentences. In (66) boland directly modifies the individual, whereas in (68) loudness indirectly modifies the individual by being applied to the emittee attribute, which in turn has the individual as its value. The crucial question is why emittee is an appropriate function saturating the f argument introduced by the adverbial modifier, but emitter is not. Boland sedâ dâdan means producing a loud sound and not that someone who is loud produces a sound. As we are dealing with predicate modifiers, it is not surprising that the emittee rather than the emitter is targeted by the adverbial modifier, since the emittee is lexically encoded in the complex predicate. The emitter, as the subject of the construction, is not part of the complex predicate and therefore not in the scope of predicate modifiers.

6.2 Getting different interpretations

After showing how the same interpretation for adverbially and attributively used modifiers is obtained, we now turn to cases where uses of the same adjective result in different interpretations. The relevant examples we are going to analyze are given in (69). In both sentences, the LVC is xunrizi kardan ‘to bleed’. The modifier used in the examples is ziyâd ‘much’, which specifies the quantity of emitted blood in its attributive use (69a). The sentence can be paraphrased as ‘s/he emitted a large quantity of blood’. In its adverbial use, ziyâd indicates the frequency or temporal duration of the event. One possible paraphrase of (69a) is ‘s/he bled often’.

That (69b) really does not mean ‘S/he emitted a large quantity of blood’ but only ‘S/he bled often’ is indicated by the example in (70). The subordinated sentence in (70) specifies the quantity of the emitted blood as being small. Thus, if (69b) meant that a large quantity of blood has been emitted, the sentence would be contradictory. (70) is not judged as being contradictory by native speakers; instead it has the interpretation that the subject referent bled often but each time s/he emitted a small quantity of blood. The interpretation that a large quantity of blood has been emitted is only an implicature in the case of adverbially used ziyâd. It is an expectation that bleeding often results in the emission of a large quantity of blood. Nevertheless, in each single event of bleeding, it can be a small quantity of blood that is emitted (see Fleischhauer 2015, 2016a, b, 2018 and Fleischhauer, Gamerschlag & Petersen 2017 for a discussion of related data from German, Russian and French).

The semantic representation of attributive and adverbial ziyâd is given in (71). We only assume one attribute quantity, which is reinterpreted as frequency or duration with respect to events.

Our analysis starts with the adverbial use of ziyâd. After the process of modifier extension, the adverbial modifier can combine with the LVC xunrizi kardan (72a). The LVC saturates the P argument of the modifier; in the next step, the f argument needs to be saturated. Since the interpretation is not that a lot of blood has been emitted but that the subject referent bled often, f can only be the identity function. This results in the interpretation in (72b). In this case, ziyâd specifies the quantity of the event, which is exactly what (69b) means.

The application of the identity function is licensed by the fact that all non-stative predicates allow for a quantity specification (this quantity specification is also called extent gradation by Bolinger 1972, Löbner 2012 and Fleischhauer 2016b; see also Doetjes 1997 on this issue).

Next, we turn to an analysis of ziyâd as an attributive modifier of xunrizi. As discussed in Section 5, xunrizi is an eventive noun, for which the semantic representation is repeated in (73).

The noun is referring to a bleeding event and therefore it could be expected that attributive modification of xunrizi by ziyâd would result in the same interpretation as adverbial modification. However, xunrizi ye ziyâd does not mean ‘the frequent bleeding’ but ‘the bleeding of a large quantity of blood’. As (74) shows, xunrizi ye ziyâd can be combined with a sentence negating a high frequency of bleeding events (a), but not with a sentence expressing that the amount of blood has only been small (b). ^[19]

One needs to explain why ziyâd specifies a property of the existentially bound emittee argument, but not of the referential argument of the eventive noun. A straightforward explanation is that ziyâd only scopes over the preverb but not the whole LVC. This is plausible, given that the adjective modifies the preverb before it combines with the light verb. With respect to the interpretation of the adverbial use of ziyâd, we assumed that quantity as a property of events is licensed by the eventiveness of the LVC. Following this assumption, only quantity as a property of the existentially bound emittee argument is a suitable target for the attributively used modifier. Going this route gives us the semantic representation in (75) for the modified preverb.

After applying the attributive modifier to xunrizi, the composition of the LVC goes straightforwardly as discussed above. The resulting semantic representation is shown in (76).

7 Generalizing the patterns

The current section aims to generalize the interpretation patterns of adverbial and attributive modification. Our analysis is based on the following – as we believe – uncontroversial assumptions:

1. (a) An attributive modifier within an LVC has scope over the nominal preverb to which it is linked by the ezâfe-morpheme.

2. (b) An adverbial modifier has scope over the whole LVC.

Assumption (a) is justified by the fact that an attributive modifier within an LVC always specifies a property of the nominal preverb, but not a property of the event. Since the attributive modifier applies to the noun, it only has access to properties licensed by the noun. Even further, the attributive modifier cannot have access to eventive properties of the LVC (e.g. frequency or temporal duration) since it modifies the noun before the LVC is formed. This means that the respective eventive properties are not licensed when the attributive modifier and the nominal preverb combine.

Assumption (b) is justified by the fact that adverbial modifiers do have access to eventive properties. The nominal preverb – as part of the LVC – is also within the scope of an adverbial modifier, as the discussion of boland sedâ dadân in Section 6.1 revealed. Since adverbial modifiers modify an LVC and not just one of its constituents, they have access to all properties licensed by the LVC, including those licensed by the preverb. A further example illustrating this is shown in (77). Xeili roughly corresponds to German sehr ‘very’ and can be used for modifying adjectives as well as verbs. The interpretation of the sentence in (77) is that the subject referent emitted a large quantity of blood, which is the same interpretation we get with attributively used ziyâd. Unlike adverbial ziyâd, xeili does not specify a property of the event such as its frequency or duration. ^[20]

Xeili could not directly modify xunrizi, as xunrizi is a noun and does not fulfill xeili’s selectional requirements. Thus, we get the same interpretation, namely ‘emitting a large quantity of blood’, by use of two different morphosyntactic constructions. It is either possible to modify the preverb directly, as with attributively used ziyâd, or to modify the LVC as a whole by use of xeili.

One issue in need of an explanation is under which conditions adverbial and attributive modification result in the same interpretation and when they lead to different ones. Attributive modifiers modify properties of individuals; adverbial modifiers are derived from the attributive use by the process of modifier extension. The process turns a modifier of individuals into a modifier of events. Thus, having different interpretations for attributive and adverbial modification should be the default, as each type of modifier applies to a different ontological type. An explanation is needed for those cases in which both modifiers result in the same interpretation.

Compare the semantic representations for adverbial ziyâd in (78a) and adverbial boland in (b). Adverbial ziyâd applies to an event description like xunrizi kardan and looks for a function f which relates the quantity attribute to the event argument. As any non-stative event description licenses quantity specification, f can be instantiated by the identity function. In its attributive use, ziyâd only scopes over the nominal preverb and therefore does not have access to properties of the event description. Thus, different interpretations for adverbial and attributive ziyâd result.

In case of modification constructions like boland sedâ dâdan ‘lit. loudly produce a sound’, loudness is not an attribute of the event itself. Thus, f cannot be instantiated by the identity function. In this case, a different attribute mediating between the loudness attribute and the event is required. In the example under discussion, the emittee licenses modification by boland – as evidenced by the fact that attributively used boland is licensed by sedâ – and can therefore mediate between the loudness attribute and the event.

Based on our analysis, we predict that if the adverbial and the attributive use of an adjective result in the same interpretation, the attributive use should be preferred. This preference should be reflected by a higher token frequency of attributively modified preverbs compared to the (token) frequency of adverbially modified LVCs. A second indication should be that constructions with attributively modified preverbs should be processed easier than constructions with adverbially modified LVCs. Both predictions are empirically testable by corpus analysis or psycholinguistic experiments. The predictions are based on the assumption that the attributive modifier directly modifies the nominal preverbs, whereas the attributive modifier does only indirectly by modifying the LVC as a whole.

Based on our discussion, we would like to propose the claims in (79). They predict under which circumstances the adverbial and the attributive use of an adjective result in the same or different interpretations. It is relevant to mention that we are always referring to pairs of adjectives and LVCs, and we do not want to claim that the adverbial and attributive use of an adjective such as ziyâd always result in different interpretations, whereas in the case of boland they never do.

It is a future task to explore which attributes are licensed by the different light verbs. In Section 6, we argued that event properties like frequency are licensed by the light verb, depending on whether it is an eventive or a stative one. In Section 5, we have shown that volitionality is also a property licensed by the light verb as the two LVCs sedâ dâdan (lit. ‘sound give’) and sedâ kardan (lit. ‘sound do’) differ with respect to that feature. Quite generally, manner seems to be a property licensed by the light verb as well. In Section 4 we discussed the example repeated in (80). In its adverbial use, xub ‘good’ specifies the manner of giving an example, which is in contrast to its attributive use.

We think that exploring the adverbial modification of LVCs in more detail will help to answer the question which meaning components of an LVC are contributed by the light verb. This seems to be a promising way of determining the exact semantic contribution of the event predicate introduced by the light verb in our semantic representations.

8 Conclusion

In this paper, we proposed an analysis of attributive and adverbial modification of Persian LVCs. Our approach is intended to account for two cases; the first being that the adverbial and the attributive modifier result in the same interpretation, and the second occurring when both result in different interpretations. We argued that whether the first or the second option holds depends on whether the respective attribute is licensed by the preverb only (which results in the first interpretation pattern) or by the LVC as well as the preverb (resulting in the second interpretation pattern).

Our analysis of separable LVCs builds on Gazdar et al.’s (1985) and Nunberg et al.’s (1994) proposal that compositionally combining idiomatic expressions show greater syntactic flexibility than non-compositional ones. Thus, we analyze separable LVCs as being compositional and propose compositionality as being the crucial factor determining whether a preverb and a light verb can be separated by attributive modifiers – meaning internal modification is licensed – or not. In this regard, our analysis differs from Karimi-Doostan’s (2011), who relates separability to the nominal type of the preverb and who has a broader notion of separability than we do. For us, only attributive modifiers intervening between the nominal preverb and the light verb count as evidence for the compositionality of the LVC, rather than any intervening functional element.

A fundamental question in the semantic analysis of LVCs is which part of the overall meaning is contributed by which component. A related question is whether the meaning of the LVCs is only dependent on the (probable figurative) meaning of its components, or whether the construction has a particular semantic contribution as well. We propose that these questions can – at least partially – be answered by systematic analysis of adverbial and internal modification, as we have done in the analysis presented in this paper. We would like to conclude with the programmatic claim that adverbial and internal modification can serve as a way of getting access to the meaning of the different components of an LVC.