3.1 Word-like properties

This subsection is devoted to a discussion of word-like properties of Persian complex predicates. The following phenomena will be discussed: stress, changes in argument structure, transitive complex predicates, separation of light verb and preverb, and morphological derivation. It will be shown that only the last phenomenon provides conclusive evidence for a morphological analysis.

3.2 Phrase-like properties

Putative word-like properties of Persian complex predicates have been discussed in the previous section. It was shown that most of the arguments for wordhood are not conclusive. The only exception is the data from derivational morphology, if one assumes Lexical Integrity (Bresnan & Mchombo Reference Bresnan and Mchombo1995). Instead of providing evidence for a treatment as words, the data reviewed above suggests that Persian complex predicates should be treated in the lexicon (if one assumes that argument-structure change is a lexical process). In what follows I discuss phrasal properties of Persian complex predicates.

4.1 Goldberg's approach to Persian complex predicates

Goldberg (Reference Goldberg, Francis and Michaelis2003) suggests representing Persian complex predicates in a constructional hierarchy that uses the concept of default inheritance. The default property of Persian complex predicates is that they are words (V0), but this default property can be overridden if a complex predicate is used together with the future auxiliary. Goldberg assumes a special construction for this case. The Future Complex Predicate Construction inherits from both the Complex Predicate Construction and the Future Construction. The property of the Complex Predicate Construction of having a bar level of zero is overridden by the value of the bar level of the Future Construction, which is one. In what follows I will explain the analysis in more detail. (28) displays the Complex Predicate Construction in the box notation that is typical for CxG.Footnote ⁹

1. (28) Complex Predicate Construction (CPV0)

   

The ‘<’ means that X0 has to immediately precede V0. Goldberg argues for a constructional status of such X0–V0 combinations, since the combination of X0 and V0 has a meaning that differs from the meanings of the components.Footnote ¹⁰

The future auxiliary xâstan is placed directly in front of the simplex verb, which has to be in the past form. Goldberg expresses this as follows:

1. (29) Future Construction

   

If complex predicates are used in the future, the auxiliary is placed between the X0 and the V0 of the complex predicate. Goldberg (Reference Goldberg, Francis and Michaelis2003: 128) argues that this auxiliary should not be treated as an infix, since it shows agreement and agreement information is usually regarded as belonging to inflectional morphology, which is normally applied outside of derivational morphology. Because of this we have a non-predictable morphological fact, which – according to Goldberg – justifies the stipulation of a Future Complex Predicate Construction. The commonalities between this construction and both the Future Construction and the Complex Predicate Construction are captured by a default inheritance hierarchy, which is shown in figure 3.

Figure 3 Combination of the Complex Predicate Construction with the Future Construction via multiple inheritance with defaults.

As is clear from looking at the figure, the Future Construction and the Complex Predicate Construction differ in important features: first, they differ in syntactic category (V0 vs. V′) and second, the linear sequence of construction parts is incompatible as well (xâstan-agr<V0 vs. X0<V0) given that ‘<’ stands for immediate precedence in Goldberg's notation. The respective values are overridden by the inheriting Future Complex Predicate Construction.

For speakers who allow for both clitic positions in (30), Goldberg (Reference Goldberg, Francis and Michaelis2003: 134) assumes the inheritance hierarchy in figure 4.

1. (30)

   1. (a) masxareh=aš kard-and ridicule=3sg did-3pl ‘They made fun of him.’

   2. (b) masxareh kard-and=aš ridicule did-3pl=3sg

(30a) can be analyzed with the CP+Clitic Construction, while (30b) can be analyzed with the V0-CL Construction, since the Complex Predicate Construction in (28) licenses a V0, which can enter the V0-CL Construction.

Figure 4 Clitic constructions.

The interaction of the Clitic Construction with the Complex Predicate Construction is shown in figure 5.

Figure 5 Part of the hierarchy that includes Future, CP, and Clitics.

Having sketched Goldberg's proposal, I now discuss several problematic aspects of this analysis.

4.2 Problem 1: Semantics needs embedding

The first problem is a formal problem: the meaning of the complex predicate has to be embedded under the meaning of the future auxiliary, but this cannot be modeled by inheritance since the meaning of the Future-CP sub-construction overrides the meaning of the CPV0 construction. Consider figure 6.

Figure 6 Inheritance hierarchy for Persian Complex Predicate Constructions including semantics.

The Future Construction has the sem value future(X), where X stands for the semantic contribution of the verb that is embedded under the future auxiliary. The semantics value of the CPV0 construction is Sem_CPV0, a value that is different from the sem value of the V0. This value has to be identified with the X in the Future Construction, but this is not possible since the sem value of the CPV0 construction is overridden in the Future-CP construction. This problem can be solved by introducing an additional bookkeeping feature, a so-called junk feature, as suggested by Kathol (Reference Kathol and Nerbonne1994: 262) and Koenig (Reference Koenig1999: section 3.3) for the inheritance-based analysis of German adjectives and for simulating the effects of argument-adding lexical rules for extraposition, respectively. Figure 7 shows an inheritance hierarchy that is augmented by bookkeeping features that make it possible to specify the future semantics.

Figure 7 Inheritance hierarchy for Persian Complex Predicate Constructions including semantics and auxiliary features.

In the Future Construction the value of sem-h is identified with the argument of the future functor (the identity is signaled by identical numbers preceded by ‘#’, this identification is also called structure sharing). In the CPV0 construction the semantics of the complex predicate is identified with the sem-h value. The sem-h values of the sub-construction are inherited from both super-constructions and by this the two sem-h values are identified. Since the sem value of the Future-CP differs from the sem value of CPV0, the structure sharing #2 is overridden. However, the sem-h value is inherited from CPV0 and therefore the sem value of the Future-CP construction is future(Sem_CPV0).

In the selection-based account that will be developed in section 5, no stipulation of auxiliary features and overriding of structure sharings is necessary.

4.3 Problem 2: Interpretation of negation

Goldberg (Reference Goldberg, Francis and Michaelis2003: section 5.2) assumes that combinations of the negation prefix and light verbs are stored in the lexicon due to their high frequency. She writes that ‘in a usage-based hierarchy, more specific stored forms preempt or block the creation of forms based on a more general pattern. Therefore, the existence of the forms mi-kardan, and na-kardan block the possibility of adding the prefixes directly to the zero level CP as a whole. This observation is based on the simple assumption that the light verb involved in a CP is recognized as the same verb as its corresponding main verb’. It is unclear how this can be formalized since the meaning of a negated complex predicate is entirely different from the meaning of the negated main verb. In addition, in examples like (31), kardan is a part of kâr kardan and it is not clear why a stored construction that contains a negated part of another construction (na-kardan) should block the negation of the whole construction.

1. (31) kâr na-kard-am (Barjasteh Reference Barjasteh1983: 62) work neg-did-1sg ‘I did not work.’

To account for examples like (31) and idiomatic complex predicates like (32), Goldberg seems to need additional constructions that take care of the embedding of the idiosyncratic semantics under the negation.

1. (32) u mard-râ dust na-dâšt she/he man-râ friend neg-have.pst.3sg ‘She/he did not love the man.’

4.4 Problem 3: Interaction of cliticization and negation

If we agree that we need additional constructions for negated complex predicates, we need further sub-constructions to deal with the interaction of cliticization and negation as demonstrated in (33):

1. (33) u dust=aš na-dâšt she/he friend=3sgneg-have.pst.3sg ‘She/he did not love him/her.’

4.5 Problem 4: Interaction of negation and progressive/subjunctive marking

As was discussed in section 3.2.3, progressive and subjunctive markers are similar to negation in that they can separate the preverb from the light verb. To deal with the progressive and subjunctive markers one could suggest a general X0 Aff-V0 construction that allows a negation affix or the progressive/subjunctive marker in the Aff slot. However, this is not sufficient since the negation affix and the progressive marker can coocur, as is demonstrated by the example in (24a), repeated here as (34):

1. (34) Ali gerye ne-mi-kon-ad Ali cry neg-ipfv-do-3sg ‘Ali does not cry.’

Since both the negation and the progressive have scope over the contribution of the complete complex predicate, another construction will be needed for cases like (34).

4.6 Problem 5: Interaction of future and negation

Yet another construction seems to be needed for the interaction between future and negation since the idiosyncratic semantics of the complex predicate has to be represented somehow.

1. (35) râdio-râ guš na-xâh-ad kard radio-râ ear neg-fut-3sg do.pst ‘She/he will not listen to the radio.’

The idiosyncratic semantics has to be embedded under the future operator and this in turn has to be embedded under the negation.

4.7 Problem 6: Interaction between future, clitics and complex predicates

Goldberg (Reference Goldberg1996: 139) discusses the interaction between future, DO clitics, and complex predicate formation. One of her examples is shown in (36):

1. (36) bâz=aš xâh-am kard open=3sgfut-1sg do.pst ‘I will open it.’

However, the example in (36) is not covered by her 2003 analysis. In order to deal with it, she would have to stipulate one more construction, namely the Future Complex Predicate Clitic Construction. The appropriately extended hierarchy is shown in figure 8. Note that the information regarding the serialization in the newly introduced construction is incompatible with the information in both of the super-constructions.Footnote ¹¹ This means that this information has to be specified stipulatively, thereby overriding the constraints on the superordinated constructions. It is important here to point out that Goldberg assumes that the information in the CPV0 construction is the default information. This information is overridden by both the Future+CP and the CP+Clitic constructions. This means that the latter constructions contain non-default information and it is unclear which construction will win if conflicting information is inherited from both of them.

Figure 8 Extended hierarchy that includes Future, CP, and Clitics.

Note also that the valence of the Future+CP construction and the CP+Clitic construction differs: since the clitic saturates one argument slot, the valence requirements of the respective linguistic objects differ. One must ensure that the valence information is inherited from the CP+Clitic construction and not from the Future+CP construction. There might be ways to formalize this if one assumes formal underpinnings similar to those assumed in LFG; but if one assumes the formalism of HPSG or Sign-based Construction Grammar, the inheritance of valence information causes a conflict which can only be resolved if valence information is default information in general and the result of the inheritance is explicitly represented in the Future-CP+Clitic construction.

According to my informants the order in (36) is not the only one possible. In Classical/Literary Persian, the clitic may also be attached to the future auxiliary. (37) shows an example that corresponds to (36).

1. (37) bâz xâh-am=aš kard open fut-1sg=3sg do.pst ‘I will open it.’

While data from Classical Persian is not relevant for a synchronic grammar, grammatical theory has to be able to account for it in principle. In section 5.6 I show how a small change in the specification of the lexical entry of the future auxiliary can account for the difference between Classical and Modern Persian.

In order to get this example in a pattern-based approach, one would need another construction that is parallel to the Future-CP+Clitic construction in figure 8 but differs in the order of the elements.

But this is not all. Goldberg (Reference Goldberg1996: 139) discusses the examples in (38), where (38b) repeats (36) above:

1. (38)

   1. (a) ?bâz xâh-am kard=aš open fut-1sg do.pst=3sg ‘I will open it.’

   2. (b) bâz=aš xâh-am kard open=3sgfut-1sg do.pst

She rates the first item with two question marks.Footnote ¹² According to my informants (38a) is fine and some even prefer it stylistically. To account for (38a) we have to introduce a new construction, since the complex bâz xâham kard is a V′ and therefore cannot be combined with the clitic via the V0-CL construction. Combining the light verb of the complex predicate with the clitic before the formation of the complex predicate is problematic for two reasons: firstly, complex predicate formation may change the argument structure of the verb so that we do not know about the DO of the whole complex, and secondly, the result of combining say kard and -aš would be V′ and therefore this V′ could not enter into the future constructions, which require a V0 in the past form. This means Goldberg would have to assume a X0 <xâstan <V0_past-CL construction.

In section 4.6, I discussed the interaction between future and negation, but of course there is also interaction between clitics and future and negation, adding three additional patterns:

1. (39)

   1. (a) bâz=aš na-xâh-am kard open=3sgneg-fut-1sg do.pst

   2. (b) ??bâz na-xâh-am=aš kard open neg-fut-1sg=3sg do.pst ‘I will not open it.’

   3. (c) ?bâz na-xâh-am kard=aš open neg-fut-1sg do.pst=3sg ‘I will not open it.’

(39a) is said to be the canonical order, but the other patterns are also attested.

4.8 Problem 7: How is the Complex Predicate Construction related to stems?

As Goldberg (Reference Goldberg, Francis and Michaelis2003: section 3.3) notes, Persian complex predicates enter into nominalizations. But this means that one does not have to list X0 V0 combinations with an idiosyncratic meaning, but rather X0 V-stem combinations, since the nominalization applies to stems, not to fully inflected verbal elements. Since the Future-CP construction needs to inherit from the CPV0 construction, there has to be a CPV-stem construction that is the heir of the CPV0 construction. This means that inflection has to be done in the inheritance hierarchy. While this is possible in principle, it is not possible to do derivation in inheritance hierarchies: Krieger & Nerbonne (Reference Krieger, Nerbonne, Briscoe, Copestake and Paiva1993) observe that recursion as in the German Vorvorvorvorversion Footnote ¹³ ‘prepreprepreversion’ cannot be covered in inheritance networks. Since information about the prefix vor- is contained in Vorversion, inheriting a second time from vor- would not add anything. To go back to our introductory example in figure 1, having two connections between printing device and printer would not change anything. If we say twice that an object has certain properties, nothing is added.

Secondly, in an inheritance-based approach to derivation, it cannot be explained why undoable has the two readings that correspond to the two bracketings in (40), since inheriting information in different orders does not change the result.

1. (40)

   1. (a) [un- [do -able]] ‘not doable’

   2. (b) [[un- do] -able] ‘capable of being undone’

Again referring to figure 1, it does not matter if we first say that a copy machine is a printing device and then add that it is a scanning device too, or if we provide this information in the reverse order. The consequence is that one needs a morphology component in which stem affix combinations form new objects which in turn can be combined with further affixes.Footnote ¹⁴ With such a morphology component the two readings in (40) and the recursion in prepreversion is unproblematic.

Discussing inheritance-based and lexical rule-based approaches to morphology, Koenig (Reference Koenig1999) argues for adopting a uniform analysis of inflection and derivation for reasons of parsimony. A uniform treatment of derivational and inflectional morphology is incompatible with Goldberg's approach.

One way to avoid the problem of relating stems and phrasal constructions is to assume a word-based morphology (Becker Reference Becker, Booij and Marle1993, Haspelmath Reference Haspelmath2002). Since the complex predicates are V0s in Goldberg's proposal they could be related to other V0s by rules that map words to words. But note that such a proposal would have problems for languages like German that allow discontinuous derivations. In German it is possible to derive nouns by circumfixing a verb stem with Ge- -e. For instance, Gerenne ‘repeated running’ is derived from renn-. If the word-based approach took the particle verb herumrenn ‘running around’ as input, we would expect a form *Geherumrenne but the correct form is Herumgerenne ‘repeated instances of aimless running events’. This could be fixed by assuming special word-based rules for derivations involving complex predicates, but that would add to the cost of the proposal under discussion, while the lexical proposal can account for the interaction of discontinuous derivations with complex predicates in a straightforward way (see section 5.8).

4.9 Problem 8: Separation of complex predicates

The examples in (13) – repeated here as (41) – pose a serious problem for Goldberg's analysis, since her construction would state that guš ‘ear’ and nemi-kon-e ‘neg-prog-do-3sg’ are adjacent.

1. (41)

   1. (a) guš be man ne-mi-kon-e (Mohammad & Karimi Reference Mohammad and Karimi1992: 197) ear to me neg-ipfv-do-3sg ‘She/he does not listen to me.’ (ear do=listen)

   2. (b) guš dige ne-mi-kon-e (Mohammad & Karimi Reference Mohammad and Karimi1992: 198) ear no.more neg-prog-do-3sg ‘She/he does not listen anymore.’

Goldberg restricts the focus of her analysis to so-called inseparable complex predicates (122). This class is defined as the class in which the preverb cannot be combined with a determiner. The complex predicate guš kardan belongs to this class. Because the phrasal analysis does not allow for the discontinuous realization of preverb and verb, the examples in (41) pose a problem for it.

Since Goldberg works in the framework of Construction Grammar and since Construction Grammar is a non-transformational framework (Goldberg Reference Goldberg1995: 7, Reference Goldberg2006: 205), Goldberg cannot derive the serialization in (41) by a transformation that turns an underlying phrasal configuration in which guš and nemikone are adjacent into the configuration observed in (41).

The only option that remains (apart from allowing discontinuous realizations of constructions; see footnote 11 above) is to simulate movement/transformations by a threading mechanism such as that used in GPSG (Gazdar Reference Gazdar1981) or by a similar technique. Usually an empty element is assumed and the information about the missing element is percolated up in the syntactic tree and bound off by a filler at a higher node. An alternative is to avoid empty elements and introduce the information about missing elements in a phrasal configuration or lexically. To get the interpretation of complex predicates right, proponents of phrasal approaches would have to assume special constructions for cases in which the preverb is not adjacent to the verb or the future auxiliary. If fronting of the preverb turns out to be possible with all of the constructions discussed so far, this means that one will have to double the number of constructions.

4.10 Summary

Taking stock for a moment, the set of constructions given in table 1 are needed. Table 1 lists the constructions proposed by Goldberg (pp. 1–6) and those that have to be stipulated in addition. I have left out the imperfective and subjunctive constructions, which are parallel to the negation case. For each case that allows a fronted preverb, another construction has to be added (see section 4.9).

Table 1 Complex predicates in interaction with other constructions

Apart from the proliferation of constructions, the problems with morphology (section 4.8) and with the separability of complex predicates (section 4.9) remain unsolved: derivational morphology cannot be dealt with in inheritance networks and in order to account for the separability, one would have to considerably modify the analysis and allow for discontinuous words or stipulate additional phrasal constructions to account for separated complex predicates. As was shown in section 4.2, semantic embedding poses a problem for inheritance-based proposals and so-called junk features would have to be assumed to derive the correct semantic representations.

Having rejected the phrasal inheritance-based proposal, I now turn to my own analysis.

5.1 Basic assumptions

HPSG is a sign-based theory in the sense of Saussure (Reference Saussure1916), that is, all linguistic objects are form-meaning pairs. A certain phonological representation is paired with a semantic representation. HPSG uses typed feature structures for the representation of all linguistic objects (lexical entries, lexical rules, syntactic rules). A lexical head comes with a valence list that contains descriptions of elements that have to be combined with this head in order to yield a maximal projection. For instance, the ditransitive verb dâdan ‘to give’ selects for a nominative NP, an accusative NP and a prepositional object.

1. (42) man ketâb-â-ro be Sepide dâd-am I book-pl-râ to Sepide gave-1sg ‘I gave Sepide the book.’

The valence list of dâdan ‘to give’ can be represented in an abbreviated form as in (43):

1. (43) comps 〈NP[nom], NP[acc], PP[be]〉

NP[nom], NP[acc], and PP[be] are shorthand for feature descriptions. I assume binary branching structures. Figure 9 shows the analysis of the example in (42).

Figure 9 Analysis of man ketâb-â-ro be Sepide dâd-am ‘I gave Sepide the book.’

The boxes with numbers mark identity of the respective information, that is, the description in the comps list of a head has to be compatible with the properties of the object that is combined with the head. The lexical entry for dâdam is combined with all its arguments in binary branching structures. Elements that have been saturated are not contained in the comps value of the mother node. For example be Sepide dâdam ‘to Sepide gave’ is the result of combining dâdam with its PP argument and therefore the PP argument is not contained in the comps list of the node that represents the complete linguistic object be Sepide dâdam. If a phrase (or word) has an empty comps list, it is fully saturated and hence a maximal projection.

Since Persian is a language with rather free constituent order (Karimi Reference Karimi2005), I assume that the subject of finite verbs is represented on the list of arguments, as has been suggested by Kiss (Reference Kiss1995), Pollard (Reference Pollard, Bunt and Horck1996), Kathol (Reference Kathol2000), De Kuthy (Reference De Kuthy2002: section 4.4), and Müller (Reference Müller2002, Reference Müller2007b) for German. This explains why the subject of finite verbs can be realized between other arguments. The authors cited above assume that the arguments in the comps list can be combined with the head in any order, which accounts for their permutability. Hence (44), which is a linearization variant of (42), can be analyzed with no further assumptions.

1. (44) be Sepide ketâb-â-ro man dâd-am (Karimi Reference Karimi2005: 17) to Sepide book-pl-râ I gave-1sg ‘As for the books, TO SEPIDE I gave them.’ or‘As for Sepide, THE BOOKS I gave her.’

See Gunji (Reference Gunji and Poser1986) for a similar approach regarding scrambling in Japanese, and Fanselow (Reference Fanselow2001) for an equivalent proposal for German in the framework of the Minimalist Program.

HPSG immediate-dominance schemata do not constrain the linearization of the daughters. This is taken care of by linearization rules (linear precedence or LP rules). The following rather trivial rules say that complements follow their head if the head is marked as head-initial and that they precede their head if the head is marked as non-initial:

1. (45)

   1. (a) HEAD [initial+]<COMPLEMENT

   2. (b) COMPLEMENT<HEAD [initial–]

Because Persian is an SOV language, verbs are marked as initial – and hence the arguments are realized to the left of the verb as shown in figure 9.Footnote ¹⁵ Since initial is a head feature, it is projected along the head path, thus ensuring that all verbal projections in Persian are initial–.Footnote ¹⁶ Another linearization rule regarding the elements in the predicate complex is discussed in section 5.4.

So far nothing has been said about the representation of semantic information. There are various proposals in the HPSG literature, but here I use Situation Semantics (Barwise & Perry Reference Barwise and Perry1983, Ginzburg & Sag Reference Ginzburg and Sag2000) since it is sufficient for current purposes.Footnote ¹⁷ (46) shows the relevant features of the lexical entries for dâdam ‘give’:Footnote ¹⁸

1. (46) Lexical entry for dâdam ‘give’ first person singular

   

The value of phonology (phon) is the phonological representation of the word. The value of head is its part of speech, and the value of content (cont) is the main semantic contribution. The semantic indices of the NP and PP arguments are referred to by the lowered boxes , , and . These indices are identified with the argument roles of the give′ relation. Figure 10 shows the syntax–semantics interaction. The linking is established in the lexical entry (46). Since the elements in the valence list get identified with the NP and PP arguments, it is ensured that the referential indices of the NPs and the PP are identified with the arguments of the give′ relation. The main contribution of each lexical head is projected to its mother node by the Semantics Principle (Pollard & Sag Reference Pollard and Sag1994: 56) and therefore the main semantic contribution of the whole phrase is identical to the contribution specified in the lexical entry ( in figure 10). The semantic contribution of NPs and PPs is not represented in figure 10 since it is irrelevant for the present discussion. See Pollard & Sag (Reference Pollard and Sag1994) and Copestake, Flickinger, Pollard & Sag (Reference Copestake, Flickinger, Pollard and Sag2005) for a discussion of NP and PP semantics and quantifier scope in HPSG.

Figure 10 Analysis of man ketâb-â-ro be Sepide dâd-am ‘I gave Sepide the book.’ including semantic information.

5.2 The lexical representation of verbs in non-compositional complex predicates

The representation for the verb that is used in the non-compositional complex predicate dust dâštan ‘to like/love’ – as discussed by Goldberg (Reference Goldberg, Francis and Michaelis2003: 139) and Karimi (Reference Karimi1997: 279) – is given in (47) in abbreviated form:Footnote ¹⁹

1. (47) Lexical entry for past stem of dâštan as used in dust dâštan ‘to like/love’

   

The phon value contains information on how the stem is pronounced. The comps list contains descriptions of the syntactic and semantic properties of the elements that have to be combined with dâštan in order to project a complete phrase. I follow Vahedi-Langrudi (Reference Vahedi-Langrudi1996: 14) in assuming that all preverbs are subcategorized arguments of the respective light verbs. Therefore, the comps list contains two full NPs and a lexical noun (lex+) which is identified as dust. lid stands for lexical identifier (Sag Reference Sag and Müller2007a). The specification in (47) ensures that the noun to be combined with dâštan is dust and not something else.

The value of CONT corresponds to the semantic content of the stem when used together with the selected element dust. The first two elements in the comps list are linked to the arguments of love. The boxed numbers convey identity of values. NP is a shorthand for a feature description of a noun phrase with the referential index . Mohammad & Karimi (Reference Mohammad and Karimi1992: 203) show that the noun in complex predicates is non-referential but has a predicate-like character. This is captured in (47): the referential potential of the noun is not used, but the information about the predicate is directly incorporated into the verb.

The lexical entry in (47), and the way it is used in the analyses that will be presented shortly, are basically a formalization of what Goldberg (Reference Goldberg1996: section 8.3) calls the Idiomatic Argument Analysis. Similar suggestions can be found in Krenn & Erbach (Reference Krenn, Erbach and Nerbonne1994), Nunberg et al. (Reference Nunberg, Sag and Wasow1994) for the analysis of idioms and in Müller (Reference Müller2003) for the analysis of particle verbs.

5.3 Simple forms

First, consider the most simple case in which the components of the complex predicate appear adjacent to each other as in (48):

1. (48) u mard-râ dust dâšt she/he man-râ friend have.pst.3sg ‘She/he loved the man.’

In this case the stem entry in (47) has to be inflected, whereupon it can be combined with dust to form dust dâšt. The analysis of (48) is shown in figure 11. The combination of dâšt with its arguments is exactly parallel to the combination of dâdam ‘give’, that was shown in figure 9. The cont value of the lexical item for dâšt is projected along the head path by the Semantics Principle (Pollard & Sag Reference Pollard and Sag1994: 56); therefore the main semantic contribution of the whole phrase is identical to the contribution specified in the lexical entry for dâšt given in (47), or rather the inflected version of this lexical entry.

Figure 11 Analysis of u mard-râ dust dâšt ‘She/he loved the man.’

The account presented here is lexical in the sense that the structure of a clause is determined by the lexical head. This does not entail that the complex predicate is formed in the lexicon, that is, the combination of dust dâšt is licensed by the lexical entry for dâštan. This account thus differs from the one suggested by Barjasteh (Reference Barjasteh1983: 225), who assumes that the combination of dust and dâšt is done in the lexicon. Such lexicalist accounts were criticized by Embick (Reference Embick2004), but Embick's criticism does not apply to accounts of the type presented here (see Müller Reference Müller2006: 874–875 for discussion).

5.4 Future forms

The future forms can be analyzed by assuming a lexical entry for the future auxiliary that makes use of the technique of argument attraction introduced into the framework of HPSG by Hinrichs & Nakazawa (Reference Hinrichs and Nakazawa1989).Footnote ²⁰

1. (49) xâstan ‘will’, future tense auxiliary:

   

This verb xâstan governs another verb in past form and attracts all arguments of this verb (). The specification of the lex value of the embedded verb ensures that the embedded element is not a verbal projection that includes arguments since such projections have the lex value –. This ensures that all arguments are raised to the future auxiliary and are realized to the left of it.

The semantic contribution of the embedded verb () is embedded under the future operator. Figure 12 shows the analysis of (50).

1. (50) u mard-râ dust xâh-ad dâšt she/he man-râ friend fut-3sg have.pst ‘She/he will love the man.’

The future auxiliary functions as the head, both syntactically and semantically: it embeds the verb dâšt and attracts its arguments (). Since only is saturated in the combination xâhad dâšt, all other arguments are represented in the comps list of the mother node. As a consequence the valence information at the mother node of xâhad dâšt is identical to the valence information of dâšt in (47). However, the semantic information is not: since the future auxiliary embeds the semantic contribution of the embedded verb under the future operator and since the semantic contribution gets projected from the head, the semantic contribution of the mother node of xâhad dâšt is future(love(, )). This value is projected along the head path and therefore is also the main semantic contribution of the complete phrase.

Figure 12 Analysis of u mard-râ dust xâhad dâšt ‘She/he will love the man.’

The future auxiliary differs from auxiliaries like budan ‘to be’ in that it precedes the verbal complex that it embeds (Karimi-Doostan Reference Karimi-Doostan1997: 179). This situation is known from Dutch and German, which are verb-final languages as well but which do not have a strictly verb-final verbal complex. While German verbal complexes are verb-final in general, there are certain situations in which verbs that are part of the verbal complex can precede the verbs they embed. For example, the perfect auxiliary hast ‘have’ precedes gewinnen helfen ‘win help’ in (51):

1. (51) daß du uns diese Schlacht hast gewinnen helfen that you us this battle have win help ‘that you helped us win this battle’

   (Haftka Reference Haftka, Heidolph, Fläming and Motsch1981: 723)

Hinrichs & Nakazawa (Reference Hinrichs, Nakazawa and Nerbonne1994) develop an analysis for German Auxiliary Flip constructions that accounts nicely for the rather complex German patterns. Verbs have a feature called flip. The value of this feature is+if the governing verb has to be serialized to the left of the verb and – if the verb is linearized to the right. (52) shows the linearization rules that ensure the proper serialization:

1. (52)

   1. (a) HEAD<[flip+]

   2. (b) [flip–]<HEAD

The flip value of the verb that is embedded under xâstan is+since this is required by the specification in the lexical entry (49). The LP rule in (52a) ensures that the future auxiliary xâstan is serialized to the left of dâšt when it is combined with dâšt. Figure 12 shows an example analysis in which the future auxiliary appears between the parts of the complex predicate.

Thus far nothing has been said about the agreement between the future auxiliary and the subject of the embedded complex predicate. Here, I follow Müller (Reference Müller2002: 369), who assumes that a finite verbs agrees with the first NP with structural case on its valence list, if there is any. In the case of subjectless predicates the verb shows default agreement. This agreement principle can account for the agreement facts in languages like English, German, Hindi, Persian, and Spanish. In sentences with future auxiliaries and complex predicates, the subject of the embedded complex predicate ( in figure 12) is raised to the comps list of the future auxiliary. The agreement principle ensures that the future auxiliary agrees with its subject; thus, since the auxiliary's subject is raised and is therefore identical to the subject of the complex predicate, the future auxiliary agrees with the subject of the complex predicate.

5.5 Imperfective prefix and negation

I assume the lexical rule in figure 13 for the treatment of negation.Footnote ²¹

This lexical rule takes a finite verb as input and licences a lexical item that has the phonology of the input item () prefixed with na-. The syntactic properties of the output () are identical to the syntactic properties of the input of the lexical rule. The semantic contribution of the input of the lexical rule () is embedded under the negation relation.

The lexical rule-based approach makes the same predictions as a word syntax approach that assumes a negation morpheme which is combined with a verb in a binary branching structure. The negation prefix would be the semantic head in the binary branching structure and would contribute the negation relation embedding the contribution of the verbal element. This relation is represented directly in figure 13. Despite their similarity in the case at hand, there are other areas of morphology in which the two approaches differ. In the lexical rule-based approach there is no need to assume subtractive morphemes, abstract morphemes for Ablaut and similar things, or portmanteau morphemes. See, for instance, Pollard & Sag (Reference Pollard and Sag1987: section 8.2), Orgun (Reference Orgun1996), Ackerman & Webelhuth (Reference Ackerman and Webelhuth1998), Riehemann (Reference Riehemann1998), Kathol (Reference Kathol, Levine and Green1999), Koenig (Reference Koenig1999), and Müller (Reference Müller2002) for lexical rule-based approaches to morphology in HPSG and the motivation for these approaches.Footnote ²²

Figure 13 Negation lexical rule depicted as unary branching tree.

The negation lexical rule can apply to simplex verbs and to the verbal part of complex predicates. The interaction between negation and non-compositional complex predicates is the most interesting one, since the negation scopes over the contribution of the whole complex predicate despite being realized in the middle of the predicate. I therefore discuss how the negation lexical rule interacts with the complex predicate dust dâštan. The lexical rule applies to the inflected form of the lexical item in (47) and licenses na-dâšt with the semantics not(love(, )), where and are linked to the NP arguments of dust dâštan. The respective items for na-dâšt may be used both in past and in future sentences. The details of the analyses are shown in figures 14 and 15, respectively. These figures are parallel to figures 11 and 12, the only difference being na- prefixation and the presence of the negation in the semantic contribution of the respective element.

Figure 14 Analysis of u mard-râ dust na-dâšt ‘She/he did not love the man.’

Figure 15 Analysis of u mard-râ dust na-xâhad dâšt ‘She/he will not love the man.’

5.6 Direct object clitics

The various clitic patterns can be accounted for in the following way. A clitic is attached to a verb that selects an argument that corresponds to this clitic. For instance, the verb see in (25) – repeated here as (53) – takes two arguments. The lexical clitic rule maps the verb didam with two arguments onto didam=aš, which does not select a direct object argument.

1. (53) did-am=aš saw-1sg=3sg ‘I saw her/him/it.’

Persian also allows for clitic left dislocation as in (54):

1. (54) Ali-râ did-am=eš Ali-râ saw-1sg=3sg ‘Ali, I saw him.’

Such cases of dislocation have been analyzed in HPSG grammars for Greek (Alexopoulou & Kolliakou Reference Alexopoulou and Kolliakou2002), Spanish (Bildhauer Reference Bildhauer2008), and Maltese (Müller Reference Müller, Comrie, Fabri, Hume, Mifsud, Stolz and Vanhove2009b). Due to space limitations I cannot present the details here.

In case of complex predicates, the clitic is realized either to the left of the verb as in (30a) or to the right of the verb as in (30b). The respective examples are repeated here as (55) for convenience:

1. (55)

   1. (a) masxareh=aš kard-and ridicule=3sg did-3pl ‘They made fun of him.’

   2. (b) masxareh kard-and=aš ridicule did-3pl=3sg

While (55b) is analyzed by the rule discussed above, which attaches the clitic to the right of the light verb, in the analysis of (55a), I assume that aš is combined with the light verb by the Head Complement Schema. In the analysis suggested here, kardand is a light verb that selects masxareh and two NPs. One of the arguments is realized by the clitic.

Similarly, the Head Complement Schema licenses the clitic in cases like (56):

1. (56) bâz=aš xâh-am kard open=3sgfut-1sg do.pst ‘I will open it.’

Since the future auxiliary raises the arguments of the verb it embeds, the arguments of kard are accessible in the lexical entry of xâham. The technique used here is the same as that used by Miller & Sag (Reference Miller and Sag1997) and Monachesi (Reference Monachesi, Hinrichs, Kathol and Nakazawa1998) to account for clitic climbing in French and Italian, respectively.

The lexical entry for the future auxiliary given in (49) requires the embedded verb to be lex+. Since clitics are combined with their head verb via the Head Complement Schema, the lex+specification excludes clitics that are attached to the future auxiliary. However, such examples are possible in Classical/Literary Persian. The respective example in (37) is repeated here as (57):

1. (57) bâz xâh-am=aš kard open fut-1sg=3sg do.pst ‘I will open it.’

This order can be accounted for by relaxing the requirement that the embedded verbal element must be lex+.

Interactions between negation and cliticization are accounted for since I assume that the negation prefix attaches to a verb without changing its valence properties. The clitic can be realized before or after any verb in whose valence list it appears.

5.7 Compositional complex predicates

So far I have mainly discussed idiomatic preverb-verb combinations. However, there is a large class of complex predicates involving a noun as preverb that are compositional and formed according to productive patterns. Karimi-Doostan (Reference Karimi-Doostan1997: 181) subdivides complex predicates with a nominal preverb into three sub-classes: those with non-predicative nouns (guš kardan ‘ear do’=‘to listen’, with verbal nouns (VN) (ʔanjâm dâdan ‘performing give’=‘to perform’), and with process nouns (râhnamâʔi kardan ‘advice do’=‘to advise’). Although I have worked out an analysis for all three cases, due to space limitations it cannot be presented here. It may suffice to say that I assume lexical items for the light verbs that select the nominal preverb. In the case of verbal nouns the arguments are selected for by the nominal preverb. I assume that the light verb attracts the arguments of the preverbal element (compare the argument-attraction analysis of future, above). The semantic content is taken over from the embedded verbal noun and combined with additional information regarding tense and aspect, which is contributed by the light verb. This analysis is similar to the argument-transfer analysis suggested by Grimshaw & Mester (Reference Grimshaw and Mester1985), which is adopted by Mohammad & Karimi (Reference Mohammad and Karimi1992) for Persian. The analysis sketched here differs from Grimshaw andMester's in its being rooted in the lexical entries for the light verbs. The way in which the arguments are attracted is specified in the lexical entry of the light verb and is not due to a rearrangement of arguments by a lexical rule or some other process. The analysis suggested here can account for the syntactic properties of Persian CPs while maintaining the assumption that all arguments are projected from the lexicon.

Instead of going into the details of the analysis of complex predicates with nominal preverbs, I will discuss causative and inchoative constructions with adjectival preverbs in more detail. As Vahedi-Langrudi (Reference Vahedi-Langrudi1996: section 4.1.5) and Karimi-Doostan (Reference Karimi-Doostan1997: section 3.3) point out, the combination of an adjective with kardan should be treated as a causative construction. In addition to the combination with kardan, there is an inchoative variant with šodan:

1. (58)

   1. (a) bâz kardan open make ‘to open’ (transitive)

   2. (b) bâz šodan open become ‘to open’ (intransitive)

Such combinations can be accounted for by the following lexical entries for kardan and šodan, respectively:

1. (59) Lexical entry for past stem of kardan as used in bâz kardan

   

This lexical entry selects an adjective which requires a subject (). This subject is identified with the second element in the comps list of kardan, that is, it is raised to the object position of the verb. In addition to this argument, kardan selects a subject NP which is coindexed with the first argument of the cause′ relation (). The other argument of cause′ is the become′ relation, which in turn embeds the semantic contribution of the adjective ().

The lexical entry for the inchoative variant differs only in not adding an additional argument and not contributing a cause′ relation:

1. (60) Lexical entry for past stem of šodan as used in bâz šodan

   

It is important to note here that neither entry specifies the form of the adjective the verb has to be combined with. This is an important difference between these compositional cases and the cases like dust dâštan ‘friend have’=‘to like/love’) that were discussed in section 5.2.

The lexical entries for light verbs that are part of compositional complex predicate formations and for light verbs that form non-compositional complex predicates are organized in a type hierarchy that captures the commonalities between classes of Complex Predicate Constructions. In this way it is possible to capture the fact that even idiomatic complex predicate formations share properties with other idiomatic complex predicates that involve the same light verb or a light verb belonging to the same class. For instance the fact that complex predicates that are formed with dâštan are stative can be inherited from a general type that describes all stative light verbs. The type for the light verb that is used in dust dâštan is a further specification of a more general, stative dâštan type since the form of the preverb and the idiosyncratic semantic contribution are specified. See Family (2006) for a discussion of verbal islands in Persian and a suggestion for representing the semantic space of complex predicates in networks.

Construction grammarians often assume a continuum between lexicon and grammar. It is unclear what this means formally since every linguistic object is either a phrasal construction or a lexical item, but in the lexical approach presented here, the general idea is captured: there are lexical items in which a lot of information regarding the arguments is specified (phonological information, selectional restrictions, pragmatic conditions) and there are others with fewer restrictions. Lexical items can be compared with regard to those values that are fixed, and hence it makes sense to talk about a continuum.

5.8 Nominalizations

The nominalization facts can be analyzed by letting the derivation affixes combine with the verbal stem, and carrying out the combination with the preverb after this combination. I assume the lexical rule in figure 16, which maps a verbal stem onto a nominal stem with appropriate semantics and adds the respective material to the phon value of the output of the lexical rule.

Figure 16 Lexical rule for agentive nominalizations.

The lexical rule applies to verbs that have an agent argument. I assume that lexical representations of verbs contain a pointer to their designated argument, if they have one. The designated argument is the subject of transitive and unergative verbs. Unaccusative verbs do not have a designated argument. The lexical rule requires the verb it applies to to have a designated argument and identifies the referential index of this designated argument (the index of the agent=) with the referential index of the resulting noun. The semantics of the verbal element () is embedded in the restrictions of the index in the result of the lexical rule application. The valence list is split into two parts, one containing full phrases () and one containing descriptions of signs that take part in complex formation (). The respective constraints on the members of and are not given in the figure. The realization of arguments is optional in general, but the complex-forming dependents have to be realized. This is captured by having as the value of comps. Cases that license other arguments are not covered by the rule, but this is for expository reasons only.

If the verb does not have dependents that take part in complex formation, is the empty list. Therefore the lexical rule applies both to simplex verbs and verbs that take part in complex predicate formation.

Figure 17 shows the analysis of the nominalization in (61).

1. (61) bâz-kon-ande-gân hanuz nay-âmade-and open-do-er-pl yet neg-come-3pl ‘The openers have not come yet.’

The lexical rule takes the semantics of the input stem () and embeds it under the entity-that-Vs′ relation. The other argument of entity-that-Vs′ () is linked to the subject (agent) of the embedded verb. The semantic contribution of the nominalization is the semantic index together with the corresponding set of restrictions that contains the entity-that-Vs′ relation. The arguments that take part in complex predicate formation have to be realized in the nominal environment as well. This is indicated by the in the comps list of the derived noun. The other arguments are optional. In figure 17 they are not presented at the comps list of the noun. Since konande is the head in the structure for bâz-konande, the semantic content () is passed up to the mother node from konande.

Figure 17 Analysis of bâz-kon-ande ‘opener’.

The lexical rule for nominalization shares an important feature with the lexical rules for negation and for imperfective/subjunctive marking. The affixes are combined with the verbal element directly. The preverb is combined with the result of the lexical rule application. As was pointed out in Müller (Reference Müller2003), such a treatment of complex predicates also solves the alleged bracketing paradoxes in the morphology of German particle verbs, which were discussed by Bierwisch (Reference Bierwisch, van der Meer and Hoffmann1987), Stiebels & Wunderlich (Reference Stiebels and Wunderlich1994), and Stiebels (Reference Stiebels1996: 46). The bracketing paradox is especially clear in German since German has circumfixes that attach to the verbal stem but have scope over the complete particle verb. Similar problems exist with regard to Persian complex predicates. As Vahedi-Langrudi (Reference Vahedi-Langrudi1996: 6, 202–203, 211) and Karimi-Doostan (Reference Karimi-Doostan1997) observe,many light verbs do not allow derivation if they are not realized together with a preverb.

1. (62)

   1. (a) pazirâʔi kon-ande entertainment do-er ‘entertainer’

   2. (b) *kon-ande do-er

Karimi-Doostan (Reference Karimi-Doostan1997: 196) therefore suggests the analysis in figure 18b rather than the one in figure 18a. As explained above, I assume the analysis in figure 18a. The problem with the non-existent form konande does not arise since the suffix -ande does not attach to the main verb kon but to a lexical item that selects a noun for later combination. Therefore the fact that kon will form a complex predicate is already contained in the lexical item. -ande attaches to this lexical item and the nominal element is combined with konande in a further step. The information provided by the noun is filled in once the noun is realized.

Figure 18 Alternative structures for pazirâʔi konande ‘entertainer’.

Karimi-Doostan (Reference Karimi-Doostan1997: 198) notes that derivations like the following are impossible:

1. (63) *pazirâʔi xâh-ad kon-ande entertaining fut-3sg do-er

In the fragment developed here, this example is ruled out because xâhad selects a verb and konande is a noun.

Vahedi-Langrudi (Reference Vahedi-Langrudi1996: 204) points out that -i derivation as in (64) is sensitive to the argument structure of the base to which it is added: it requires the verb it attaches to to select a direct object.

1. (64)

   1. (a) didan-i see-worthy ‘worthy to see’

   2. (b) pardâxt-kardan-i payment-do-worthy ‘possible/worthy to pay’

   3. (c) darmân-kardan-i cure-do-possible ‘possible to cure’

Vahedi-Langrudi concludes that this indicates that the affixations apply to the complex verbal heads in Persian. However, an alternative analysis which is in the spirit of the analysis presented above is possible for -i derivation. Müller (Reference Müller2003) suggests an analysis for similar derivations with the German suffix -bar: the suffix attaches to a stem that contains information about the particle and about the arguments that are contributed by the particle. That is, all information about the argument structure of the complete particle verb is already present in this lexical item in an underspecified form. -bar imposes the respective constraints, suppresses the subject of the verb and promotes the object to subject. Due to space limitations I will not provide the analysis of the Persian -i derivation here, but point to the formation of passive participles and -bar derivation that was discussed in Müller's work.

5.9 Summary

In this section I have proposed a lexical treatment of complex predicates in Persian. The preverb is treated as a dependent of the base verb. This treatment can account for the separability of preverb and verb in syntax and morphology: the preverb can be fronted, the future auxiliary can intervene between preverb and verb, clitics can be placed between preverb and verb. Since the syntactic and semantic properties of the whole complex predicate are contained in the lexical item for the verbal stem already (although not fully specified in the case of productive complex predicate formations), an analysis is possible in which morphological processes that are sensitive to information provided by both parts of the complex predicate can affect the verbal stem. Such an analysis covers both the Persian cases and the German examples involving discontinuous derivation morphemes that attach to both sides of the verbal part of a complex predicate.

6.1 Separability

Goldberg (Reference Goldberg, Francis and Michaelis2003: 141) claims that the Idiomatic Argument Analysis predicts that ‘Persian CPs should be generally separable as are general DO+verb combinations in Persian. As we have seen, however, CPs are only separable in certain specific circumstances’. This is not the case, however; as was pointed out in the data discussion, the property of being realized together is a property that many idioms have despite their being clearly phrasal. Linearization constraints ensure that material that is part of an idiomatic expression is serialized together. For SOV languages like Persian, Dutch, or German this means that idiom parts are serialized clause-finally next to the verb. The same holds for complex predicates in Persian and particle verbs in German and Dutch.

Linearization is a complex matter and it is known from German that idiom parts may be extracted, but also rearranged locally under certain conditions. Since none of the constituent order rules for NPs in German is strict, researchers such as Uszkoreit (Reference Uszkoreit1987) suggest violable linearization rules. A grammar with such violable linearization rules can account for the examples in (11) – repeated here as (65) – and even explain the markedness of the first sentence.

1. (65)

   1. (a) ??setâyeš Ali-râ kard-am adoration Ali-râ did-1sg ‘I adored Ali.’

   2. (b) Ali-râ setâyeš kard-am Ali-râ adoration did-1sg ‘I adored Ali.’

Goldberg's account does not assign a structure to sentences like (65a) and hence cannot explain why they exist and why they are marked.

6.2 Selection of certain properties

Goldberg (Reference Goldberg, Francis and Michaelis2003: 141) further argues against the Idiomatic Argument Analysis by claiming that it makes necessary the selection of unusual properties. She writes:

Goldberg's argument is not valid, however, since the two approaches are entirely equivalent as regards the features of preverbs and verbal elements that have to be stated. In Goldberg's analysis the constraints are stated in two boxes inside of a third box as in the case of dust dâštan ‘to love/like’, while in the Idiomatic Argument Analysis the same constraints would be stated inside of the special lexical entry for dâštan that selects dust. Figure 19 shows this graphically in a CxG-compatible form. The value of phon is a list that contains information about the phonological realization of a linguistic object and the value of val is a set that contains all the arguments (Kay & Fillmore Reference Kay and Fillmore1999). The constraints on the parts of the constructions are exactly the same in the two frameworks, but they are stated at different places.Footnote ²³ Similar pictures can be drawn for compositional cases of complex predicates. So everything that Goldberg states in her phrasal constructions can be stated in the lexical construction. If she states restrictions on definiteness in the phrasal pattern, the same can be formulated as a restriction on the kind of argument that is selected. Neither way of formulating the restriction is more ad hoc or motivated than the other.

Figure 19 Comparison of the phrasal and the lexical approach.

If one rejects the approach that assumes multiple related lexical items for light verbs, one can still assume an approach that is equivalent to the approach discussed in section 5. A set of related lexical rules would be applied to light verbs. This can be depicted as in figure 20. The light verb dâšt enters the lexical construction and licences an item that selects for the preverb.Footnote ²⁴ Only this item provides the idiosyncratic meaning of dust dâštan. In a further step the construction in figure 20 is combined with the preverb dust. Such a templatic approach would capture the proposal of Goldberg (Reference Goldberg, Francis and Michaelis2003) and Family (Reference Family2006) that preverb and light verb enter into a certain configuration and by doing so a meaning emerges that differs from the meaning contributions of the parts.

Figure 20 Lexical approach with separate lexical entry for the light verb.

If one compares this approach to the one that was suggested in section 5, it is clear that the facts are captured in both proposals. However, the approach just sketched uses one more lexical item: the one for the light verb. So instead of classifying lexical rules/templates like that in figure 20 in a hierarchy, one can classify lexical entries directly, as was suggested in section 5.

6.3 Learnability

Some proponents of Construction Grammar and Cognitive Grammar argue that the way children learn language is evidence for phrasal constructions since children acquire certain phrasal patterns first (Dąbrowska Reference Dąbrowska2001: 85–86; Tomasello Reference Tomasello2003: 107; Reference Tomasello2006a, Reference Tomasello, Fischer and Stefanowitschb; Goldberg, Casenhier & Sethuraman Reference Goldberg, Casenhier and Sethuraman2005). The same remark as in the previous section can be made in reply to this argument: the phrasal and the selectional approach are indistinguishable as far as learnability properties are concerned (see also Müller in preparation). In the first case the learner learns a certain pattern and a corresponding meaning; in the second case the learner learns valence classes or idioms that select certain other elements and that have a special meaning if they appear together with the selected element. The only thing that is required in the lexical approach in addition to what is required in the phrasal approach is knowledge about the fact that two elements may be combined at all, i.e. that there are functor argument structures. I consider this a very straightforward thing to acquire.

Tomasello (Reference Tomasello2005) argues that semantic and pragmatic information plays an important role in language acquisition but this does not differentiate between the phrasal and the lexical analysis if one assumes that syntactic, semantic and pragmatic information is contained in the representation of linguistic objects. Tomasello's treatment of language acquisition is compatible with the lexical analysis of Persian complex predicates.