3.1 Proposal: Two structures

This Section argues that structures like (2a) and (2b), given again as (8a) and (8b) for convenience, have different syntactic structures.

In contrast, previous work on (8b) argue that it has the same structure as (8a), as outlined in the Introduction. There are two key arguments in favor of a One Structure Approach to (8a) and (8b). The first concerns a particular analysis of the “unselected object”. The One Structure Approach observes that the main verbs in both (8a) and (8b) are “unergative verbs” (in a traditional sense), and that unergatives do not license NP direct objects (9). It is then emphasized that both sing and waltz can appear with a direct object only when the verb phrase includes a second XP (on (8b) see especially Hoekstra & Mulder 1990; Levin & Rappaport Hovav 1995; Folli & Harley 2006; Ramchand 2008). The consequent view is that the unselected object is syntactically and semantically dependent on the secondary XP, and not the main unergative verb; and that this is the case in both (8a) and (8b). A second idea emphasized in the One Structure tradition is that (8a) and (8b) are semantically related, as both have causal meanings.

While the details of its analysis remains an active topic of research, (8a) is uncontroversially characterized as a resultative, specifically a resultative built on an unergative primary predicate, known as an Unselected Object or Exceptional Case Marking (ECM) resultative (Levin & Rappaport Hovav 1995; Wechsler 2005). Folli & Harley (2006: 137) develop in particular detail a syntactic, unified analysis of the structures underlying (8b) and (8a) that illustrates the One Structure view (10) (for another influential treatment, see Ramchand 2008).

(10)	Resultative analysis (Based on Folli & Harley 2006) (To be rejected)

On this Small Clause analysis, a secondary predicate is a complement of vP. The secondary XP supplies a specifier in which the unselected object can be syntactically and semantically licensed. The object has no syntactic or semantic relation to the verb. The key point from (10) is that the object John is licensed in relation to the secondary XP (which would also be the case if e.g. a complex predication analysis of resultatives were adopted).

In the remainder of this section I argue that, while a resultative syntax as in (10) (or a complex predicate alternative) is correct for (8a), the structure of (8b) is as in (11). Specifically, I argue that the “unselected object” in (8b) is not a dependent of the secondary XP, in contrast to (8a). As such, the manner of motion verb phrase in (11) has a direct object NP complement, i.e. a transitive vP syntax (where “transitive” is to be understood as in Section 2). The PP is a vP adjunct. The object does not bear a syntactic or semantic dependency to the PP.

(11)	Adjunction analysis (Structure to be partly revised in Section 4)

Given this analysis, I will refer to verb phrases like (8b) as Manner of Motion Transitives (MMT).³

3.2 Evidence from syntax

Syntactic evidence that (8b) has the structure in (11), and does not involve resultative secondary predication, comes from constituency.

In resultatives, the resultative secondary predicate is not a constituent independent of the main verb and object (Simpson 1983; Roberts 1988; Levin & Rappaport Hovav 1995). Tests for verb phrase constituency are given in (12): VP pro-form do so (12a), though-movement (12b), V-fronting (12c), and an adverb intervening between O and PP (12d) (see Williams 2005). In (12) these tests produce ungrammatical strings when applied to resultatives like (8a), confirming that in resultative secondary predicate syntax the verb and object do not form a constituent to the exclusion of the secondary XP.

In contrast, applying these tests to MMT verb phrases like (8b) produces grammatical sentences (13).

As such, the PP in (8b) is an adjunct, as in the structure in (11). The object argument NP in (8b) is not syntactically introduced with, or syntactically dependent on, the PP.

3.3 Evidence from event structure interpretation

This syntactic difference between (8a) and (8b) correlates with the long-standing observation that these two sentences have very different event structural interpretations (among other properties) (see Beavers 2012: 926–927 for discussion).

A typical characterization of resultative semantics is that the secondary XP denotes some state, a state that is a consequence of the event described by the main manner verb (e.g. Beavers 2012). As the state is held by one of the participants, resultatives can be sensibly modified by a temporally-delimiting telic adverb (14a). In contrast MMT clauses can be atelic (14b) (dependent on the semantics of the PP), and are consequently regularly identified as the exception to this generalization (Goldberg & Jackendoff 2004; Folli & Harley 2006).⁵ The object of (14b) is also not a holder of a result state (Folli & Harley 2006; Ramchand 2008; Beavers 2012).

There are further differences in composition. Resultative formation is said to “covertly” derive an eventuality that is independent of those denoted by its primary and secondary predicates (Dowty 1979). One diagnostic is the by-Manner/Means adjunct (by-MM), a type of modifier that describes the Manner/Means by which an event unfolds (Dowty 1979). As shown particularly clearly for ECM resultatives with (15) from Williams (2009: 694), a by-MM adjunct can be introduced in a resultative clause that does not sensibly refer to either of the events described by the overt predicates making up the verb phrase: (15) neither entails that Ozzy sang by not resting between songs, nor that Ozzy is hoarse by not resting between songs. Instead by-MM describes the derived complex eventuality (“sing-hoarse”).

As far as I know, it has never been shown that the derived eventuality is composed with sentences of the kind in (8b). To the extent that by-MM can modify an MMT,⁶ the adjunct describes the event described by the manner of motion verb. In (16), for example, taping it around her feet describes the means by which the dancing takes place, not dance-around-the-room.

In order to accommodate contrasts of these types, previous work proposes revisions to standard analyses of resultatives, severely weakening generalizations about resultative semantics in the process (Beavers 2012). If the analysis here is correct, these revisions are unnecessary. Instead, (8b) does not have resultative semantics because it is not syntactically resultative.⁷ The appearance that (8a) and (8b) have the same underlying structure follows from the fact that vP-related prepositional phrases appear in clause-final position in English when the string is pronounced.

3.4 Against alternative analyses

This subsection clarifies that the key dimension by which (8a) and (8b) differ is syntactic.

First, and most importantly, (8a) and (8b) should not be distinguished by the lexical semantics of the unergative predicates involved. Manner of motion unergative verbs occur in “unselected object” resultatives (17) (examples based on Rappaport Hovav & Levin 2001). It is therefore only the waltz(-type) clauses of the kind in (8b) that are not resultative. I develop this point in detail through the remainder of the paper.

Evidence that the examples in (17) are structurally resultative come from the same constituency tests. The relevant contrasts between (8a) and (8b) are therefore in the structure of (2b), rather than root semantics.

Another perspective might be that resultative secondary predication exclusively involves APs, and not PPs, as suggested in Rothstein (2004); Kratzer (2005) and Mateu (2005). However, PPs can form resultative syntactic structures according to the constituency tests from the last Section, as shown in (20) and (21). The category of the secondary XP in (8a) and (8b) is therefore irrelevant.

Third, the difference between (8a) and (8b) does not concern (a)telicity, and whether the secondary XP is bounded or unbounded. The examples like (8b) discussed so far have an unbounded Path PP. Constituency tests with the bounded Goal PP in (22) show the bounded Goal is also an adjunct, again in contrast to the Result XP in resultatives.

Finally, Ettlinger (2008) and Matushansky et al. (2012) suggest another potentially relevant lexical semantic approach to resultative formation, whereby only property-denoting phrases can be complements and produce resultatives, while any Path or Directional phrases merge as adjuncts and so are not resultatives. It is not clear that this generalization is correct. The examples in (23) involve syntactic complementation with Path PPs.

Moreover, property-denoting PPs, like to a stunning shade of black in (24), can be adjuncts (example from Embick 2004: 382 (62)), as shown in (25) with do so (Embick 2004: 382 (63)), though-movement, V-fronting, and post-object adverbs tests.⁸

Incidentally, if this characterization of (e.g.) (20) and (24) is correct, then the interpretation of the directional P in relation to the verb (phrase) it combines with is sensitive to the syntactic position of the preposition. This suggestion is consistent with what is known about locative adverbials, whose composition has long been recognized to correlate with clause position (see especially Maienborn 2001). In particular, when locatives attaches low, the PP specifies a location relating to the object (26). In contrast, when the locative attaches high(er) it specifies the region in which the event takes place. The locative in Argentina in (27a), for example, specifies the region in which the whole event takes place; it does not locate the contract (Maienborn 2001). Similarly, along his arm in (27b) describes the region occupied by the tickling-of-John, and not a trajectory of John.

The sentences in (28) provide initial reason to think structure has a similar effect with directional Ps: in (28a), to heads a property-denoting XP, that (I have argued (25)) has an adjunction relationship to vP; in (28b), to heads a property-denoting XP that is a resultative complement; and in (28c) to can head a Path XP that looks like a (resultative) complement. I leave full investigation of these effects to future work.

Returning to the main point, there may be a range of superficial differences between the sentences in (8a) and (8b), but the key to explaining their differences is the syntax of their verbal structures.

4.1 Background: Previous work

This section turns to the question of object interpretation in sentences like (2b), as well as (29) and (30). Since the earliest work on the topic, it has been observed that these particular unselected objects have the remarkable property of being interpreted as an “agent”, or “the participant (at least in part) responsible for the unfolding of the (manner of motion) event”.

The apparent agentivity of the object has been central to study of these unergative “transitive alternations”, in work that has made significant contributions to questions as diverse as selection, causation, and the determinants of argument interpretation, among other topics (Cruse 1973; Jackendoff 1990; Hale & Keyser 1993; Ritter & Rosen 1998; Levin & Rappaport Hovav 1999; Folli & Harley 2006; Ramchand 2008). This literature has centered on the observation that as agents, the object in (e.g.) (29b) is a “causee”, in contrast with objects in inchoative-causatives or instrumental-causatives (31), which are not. The agent object is thus taken to derive an intuition that the unergative transitive alternants like (29b) involve “indirect causation”, as opposed to the “direct causation” interpretation in transitive inchoatives (Brousseau & Ritter 1991). Levin & Rappaport Hovav (1995: 110) refer to transitives like (29b) as the “causative pair”, while the regular transitive alternant in (31a) is “causative”.

Despite wide-ranging interest in transitive unergatives, few formal implementations of the object’s syntax and semantics have been developed. Previous studies concentrate on lexical accounts of the interpretation of the unselected object. This is because the objects of other unergatives in (apparently) the same frame are not interpreted as responsible for events named by the verb (32a) (at least, not without massive coercion); similarly, in resultatives like (32b), Mary is the singer; John cannot be a singer. In consequence it is assumed the interpretation of the object in (29b) is specified in relation to the root or verb, and cannot be determined structurally.

In an early (non-resultative) lexical implementation, Brousseau & Ritter (1991) propose that a lexical representation of the transitive variant derives from that of the intransitive by the addition of a CAUSE predicate to its Lexical Conceptual Structure (33). The two representations have the same pronunciation. In this approach, the predicate DO is included in the semantic representation of verbs which select an argument that is directly responsible for the action. In both representations of jump in (33), the predicate DO has an argument. English consequently has (a limited class of) causatives of unergatives. Assuming that Brousseau & Ritter (1991) are not adopting extraordinary rules of mapping between lexical and syntactic structure, in contemporary terms, the lions are presumably Agent external arguments in both (29a) and (29b); and there is no internal argument in either example.

Ramchand (2008) offers a different assessment of the data in an influential syntactic, resultative analysis. In her first-phase framework, in Michael ran Karena to the coconut tree, the object Karena holds a “composite” role: Karena is doing the running and so holds a role in relation to a ProcessP (Ramchand 2008: 118); Karena also holds a “Resultee” role in relation to a Resultative projection, corresponding to the PP (34a). Discussion of the “special interpretation” of the object is left brief, but the object of the unergative occurs in the same organization as that of a resultative built on a verb like pound (34b) (Ramchand 2008: 127). I argue against a “pound analysis” in Section 4.4.2.

Finally, in a framework closest to that adopted here, Folli & Harley (2006) develop a root-based treatment. They begin by arguing for a unified resultative analysis of (2a) and (2b). Folli & Harley (2006) then point out that a single structure approach presents a challenge to frameworks that adopt a syntactico-centric approach to argument interpretation: if waltz John around the room and sing her throat hoarse have the same underlying syntax, how is it that two different argument structures (different argument interpretations) are generated? Folli & Harley (2006: 142–51) argue it is necessary to invoke formal lexical semantic distinctions between roots to get the contrast, and propose an articulated feature calculus (+/–Agent, +/–Path). Different classes of root bear different organizations of features, so that $\sqrt{\text{SING}}$ is in a different class to $\sqrt{\text{WALTZ}}$ . When these roots adjoin to the resultative syntax, the sets of features associated with $\sqrt{\text{SING}}$ and $\sqrt{\text{WALTZ}}$ activate or generate different semantic outputs of the structures, so that the two verb phrases have different argument interpretations: the object in (2b) is interpreted as an agent because of the formal lexical semantic features borne by $\sqrt{\text{WALTZ}}$ ([+Agent, +Path]); the object of (2a) does not because $\sqrt{\text{SING}}$ does not bear the correct features that trigger this interpretation. The specification of these arguments’ interpretation by root class is established in the Encyclopedia (Folli & Harley 2006: 142–51).⁹

An agent object, and a lexical treatment, present a number of formal challenges that have yet to receive explanation. First, what is the nature the participant role being assigned to the object? As noted in the introduction, an agent internal argument would violate standard mapping principles (thematic theories hold that internal arguments are formally patients or themes, never agents; (e.g. Marantz 1984; Levin & Rappaport Hovav 1995; Reinhart 2003)). The presence of apparently “two agents” in examples like (2b) (one held by the subject, one by the object) also violates Thematic Uniqueness (Bresnan 1982; Carlson 1984). Previous analyses have not addressed what it means to say that the objects in these verb phrases are “unselected”, if the interpretation is determined by the main verb. This question is particularly puzzling in resultative analyses, where an “unselected object” is usually thought to only have a syntactic and semantic dependency with the secondary predicate. Finally, and particularly interesting, a lexical analysis commits to the possibility that a unique phrasal syntactic structure can systematically produce multiple argument structure mappings in accordance with a root class specified rules of interpretation; and that a meaning generated by a root rule can generate an interpretation that violates standard, formal principles of syntax-semantics correspondence (i.e., an agent object).

The argumentation of this section starts from the conclusion of Section 3, that waltz John around the room and sing her throat hoarse do not have the same structure. This means that the interpretation of the object is not to be explained within the confines of a resultative syntax. With this premise in hand, this section reviews the status of MMT objects, addressing the following three questions:

1. Where is the “unselected object” introduced in the verb phrase?

2. What is the interpretation of the “unselected object”? Is it an Agent?

3. How is the interpretation of the “unselected object” determined? With respect to what linguistic object is it interpreted?

I propose a new syntactic analysis of manner of motion verbal structures that makes use of the framework set out in Section 2. I argue that the internal argument of transitive motion verb structures is a (particular type of) patient/theme, and is not an agent. The object is interpreted in accordance with regular rules of syntax-semantics mapping. Its interpretation is not specified by the root.

4.2 The syntactic position of the object

This subsection presents evidence that the internal argument of MMTs is syntactically introduced within the verb phrase, further confirming the conclusion in Section 3, that the PP adjunct does not syntactically or semantically license the direct object in MMTs.

4.3 The interpretation of the object

4.4 The structure of objects in motion verb phrases

The remaining question is how the in motion objects receive their interpretation. Making use of the framework set out in Section 2, I propose a structural analysis of these verb phrases, and argue that the in motion object is interpreted in relation to this verbal structure.

4.4.1 A structural analysis

I have already shown that the object of waltz in Mary waltzed John around the room is syntactically an object of the verb. The structure of this verb phrase is consequently transitive (at least informally), where “transitive” just refers to the syntactic organization of the arguments (still glossing over considerations such as case) in which the verb has a nominal complement, and the specifier of Voice is filled (see Section 2).

One possibility is that in the waltz clause, a root $\sqrt{\text{WALTZ}}$ occurs in the same simple transitive structure as that typically assumed for manner transitives like hit, sweep, or pound.

(61)	Simple verbal structure in transitive syntax (To be rejected)

A particular facet of the framework adopted in this paper is that more than one verbal substructure can occur in a transitive syntax, however. This means that the verbal substructure of a transitive could be syntactically different to that of the simple structure of hit or sweep, either in the syntactic organization of the verb, or in the sets of syntactic features involved. As such, the root $\sqrt{\text{WALTZ}}$ could, in principle, modify either the simple structure in (61), or a different, more complex verbal structure.

There are a number of ways in which a verbal structure might be complex, relative to (61). I propose the waltz verbal substructure might differ from (61) in that its verbal structure might include an affix. This possibility is illustrated in (62). The verbalized root combines with an affix, forming a complex head. I assume the affix is a prefix, but the affix is not pronounced in English. I discuss the structure in (62) in more detail in Section 4.4.4.

(62)	Affixed verbal structure in transitive syntax (Final structure)

In both (61) and (62) the verb (e.g. waltz) pronounces a complex syntactic structure in which a root combines with v; but the latter also has the affix. In addition, in both structures the verbal complex names an activity. The choice between the two verbal substructures (the simple or complex structure) that could underlie waltz is subtle. Analysis is further challenging because the waltz type structures do not appear to be attested outside of English (see Section 5.2),¹² and there is therefore no clear comparative cross-linguistic evidence to draw on.

However, the two syntactic configurations in (61) and (62) make different predictions with respect to the properties of their internal arguments. If $\sqrt{\text{WALTZ}}$ occurs in (61), waltz might be expected to relate to its object in the same manner as pound and its object; if it occurs in a structure like (62), the object of waltz should have different properties to objects of verbs like pound. Once the direct object is considered, I argue that there are a number of reasons to favor the complex structure for MMTs.

4.4.4 The complex verbal structure

In light of these considerations, I propose the θ_motion Role is interpreted in relation to a complex verbal structure, as in (67).

(67)	Final structure: Mary waltzed John around the room.

The syntax of (67) is transitive in the sense that the specifier of Voice is filled with an argument; and there is a second argument, John, that is the complement of the verb. Also shown in (67) is an (optional) PP, which adjoins to vP.

The verbal substructure is complex. I suggest it includes an affix that denotes Move, or Motion. The affix does not seem to have an aspectual effect on the event named by the verb, in the sense that (67) is interpreted as an activity. I propose that the object is interpreted in relation to the complex motion event given by the affixed verb, in accordance with its verb internal structural position. The θ_motion role of the object is established via the interpretation of the complex functional structure.

If the interpretation of the object is determined in relation to the complex structure, the interpretation of the object is limited to this syntactic context. The proposed verbal substructure may underlie other expressions, besides the waltz John around the room and jump the lions through the hoop type. Verb phrases like paddle the canoe are one example. Others are possible.

The overall effect of the framework, as illustrated by this case study, is a fine-grained approach to notions of transitivity, argument interpretation, selection, and argument-hood. According to (62), the θ_motion object holds a participant relation to “the verb”, where the verb is defined as the affixed categorized root [[ $\sqrt{\text{WALTZ}}$ v ] v ]. The internal argument holds a participant relation to the complex verbal structure, but is “unselected” in the sense that it does not hold a participant relation to the categorized root [ $\sqrt{\text{WALTZ}}$ v ]).

A point raised by a reviewer is root coercion, observing that (68) cannot mean that the internal argument the passengers are independently in motion in relation to an activity event sleep, even with world knowledge that it is commonly the case that people fall asleep on planes or trains.

(68)	*The pilot slept the passengers smoothly.

Such examples seem to indicate that the structure I have proposed is only felicitous if the lexical/conceptual semantics of the root can describe manner of motion; as $\sqrt{\text{SLEEP}}$ , as well as $\sqrt{\text{CRY}}$ , $\sqrt{\text{SING}}$ , and $\sqrt{\text{LAUGH}}$ , express only manner, they are not possible (69).

(69)	a.	*Mary cried/sang/laughed John (for hours).
	b.	*The comedian laughed the audience hoarse/to hysteria.
	c.	*The comedian laughed the audience around the room.

The formal means by which root distribution is to be implemented is a matter of current debate, not specific to the data examine here; see Embick & Marantz (2008) and comments in Section 2 for relevant discussion.

Overt evidence for a complex structure would be ideal. However, as already noted, the waltz-type “transitivized manner of motion unergative” has been reported to be unattested outside of English, and so comparative study may not be possible. Somewhat indirect motivation for a complex structure is that, from a comparative perspective, motion verbs (and transitives of unergatives) commonly involve a morphosyntactically complex verbal structure, not simple structures (i.a. Talmy 1985), including prefixes (e.g. Slavic), verbal “classifiers” (including in voiced causatives) (a well known feature of Athapaskan languages),¹⁶ and light verbs (for example, Mandarin and Korean make use of deictic light verbs). I leave to future study whether cross-linguistic comparisons that relate specifically to transitivity can be identified. I hope to have laid out how such investigations might proceed.

4.5 Interim summary

This section has shown that MMTs are transitive, with the “unselected” object of waltz a syntactic direct object. This object is semantically interpreted with a type of theme/patient role, which I have called a motion role. I implemented a syntactic analysis of MMTs that derives this relationship. Making use of a framework in which descriptive notions such as “unergative”, “transitive”, and “verb” are teased apart into their component syntactic pieces, I argued that when manner of motion roots like $\sqrt{\text{WALTZ}}$ appear in transitive argument structure syntax, the root modifies a complex verbal structure; and it is in relation to this complex structure that the object is interpreted.

Close study of MMTs thus shows they do not provide evidence that there are special compositional rules, specific to certain verb classes, that can determine argument interpretation that violate general rules of argument interpretation. Instead, the interpretation of MMT objects seems to be specific to certain syntactic contexts, and can be analyzed as a structural effect.

5.1 Objects in motion and resultative formation

The pair in (70) involve an adjectival resultative secondary predicate and a manner of motion main verb waltz. The string is grammatical with a “fake reflexive” object (70a), but not the full nominal object (70b).

The fake reflexive has been much discussed in previous literature on resultative formation, and further examples are given in (71), drawn from Levin & Rappaport Hovav (1995); Wechsler (1997); Rappaport Hovav & Levin (2001).

The examples in (72) replicates the contrast from (70), showing that the fake reflexive cannot be replaced with a nominal argument.

The contrast in (70a) and (70b) is a problem if both are resultatives, and nothing more is said. The analysis developed here raises the possibility that the difference follows from a structural difference between the pairs. I assume that the fake reflexive is possible in resultative structure as a dummy marker that is the holder of the result state in a resultative.

In support of structural correlations, a reflexive is not licensed in the absence of a secondary predicate, as shown by the application of traditional constituency tests (73), and ungrammaticality with those modifiers identified in Section 4 (74).

These judgements are replicated if the AP is substituted by a Path PP, as in (75) and (76).

Interestingly fake reflexives have regularly been reported as odd with Path and Directional PPs in the previous literature (77) (from Rappaport Hovav & Levin 2001: 782). The analysis developed here suggests a novel account of this, as an effect of the fact that the fake reflexive is introduced in resultative contexts; while (77) (on the account here) would require a motion verb structure to get the in motion object interpretation.

Of course, if the fake reflexive is a reflex of resultative structure, we do not expect it to occur in transitive motion syntax. Exactly as expected on this reasoning, it is difficult to coerce an interpretation where a reflexive is in motion. To the extent that an argument is interpreted as in motion in (70a) and (71), it is the subject of the manner verb that is in motion. This recalls a point in the text around (65a), where it was observed that resultative formation seems to be generally available on manner of motion verbs where the object does not hold a θ_motion Role. Similarly, the objects in (78) cannot be interpreted as in-motion, but are interpreted as result-state holders.

In sum, the availability of the result-holder interpretation of the object in (70a) and (71), but not an in-motion object interpretation, correlates with the syntax of these sentences. The ungrammaticality of the objects in-motion in (70b) and (72) similarly show that objects with the in-motion interpretation are not possible in resultative syntax/semantics. The generalization across both contexts may be that the resultative cannot be built on the complex verbal structure I have proposed. There are a number of ways analysis might be cached out, which I leave to future study.

5.2 Motion and resultative formation, cross-linguistically

A pattern that is currently unexplained concerns the distribution of manner of motion verbs in resultatives cross-linguistically. Resultatives can be built on “unergative” or “non-specified object” verbs in languages like German and Dutch (79)–(80).

As these examples illustrate, German and Dutch exhibit a greater flexibility than English in the lexical semantics of the verbs that can be the primary predicate of resultative secondary predication (for summary of patterns in English, see Wechsler 2005; Beavers 2012). Yet translations of the “causatives of unergatives” of the waltz John type discussed in this paper are not possible in these languages. While some native speakers seem to accept (81) and (82), many outright reject the data (particularly with the inanimate object) in a manner that is not observed for resultatives of the kind in (79) and (80). The judgements seem to vary particularly across different Dutch speakers, indicated by % in (82); Dutch speakers accepted (82b). (Many thanks to Boris Haselbach and Ava Creemers for help with the examples and discussion).

The cross-linguistic data suggest that the restricted distribution of particular “unergatives” in resultative syntax observed for English and cross-linguistically requires further investigation, on the approach here, both of the set of complex verbal structures generated, and the interaction of roots with those structures in different languages.