Spanish participios activos are adjectival antipassives

2.1 Argument structure: internal vs. external arguments of source verbs

All adjectival participles have a single argument that is either a head noun in an attributive construction (2a) or an argument in a predicative construction as in (2b).

An uncontroversial fact about adjectival participles such as (2) is that they are derived from either passive or unaccusative verbal sources and that the argument of the derived adjective (=puerta in (2)) is interpreted as the internal argument of the passive or unaccusative verbal source (see Borer 1984; Bruening 2014; Levin and Rappaport 1986; McIntyre 2013; Meltzer-Asscher 2011; among others for detailed discussion of this generalization and the different ways that have been proposed to capture it). The main difference that sets participios activos apart from these canonical adjectival participles is that the argument of the adjective in contexts such as (2a) and (2b) does not correspond to the internal argument of a verbal source but rather the external argument. In (3), niño (=child) is the agent of the source verb comer (=eat) rather than the theme.

If all adjectival participles were derived solely from passive and unaccusative verbal sources, examples like (3) would be unexpected. One potential way to deal with participios activos without having to weaken the passive/unaccusative generalization is to treat cases like (3) as verbs. That is, a possible analysis of (3a) would be a reduced relative clause of an active perfective sentence like that of (4).

There are number of problems with this idea. First, reduced relative clauses are only possible if the copular verb, ser or estar, is elided but not if the perfective auxiliary haber (= have) is elided. Second, verbal participles in active sentences are non-agreeing while adjectival participles agree in number and gender with their argument. Since agreement obtains in participios activos as in (5), this supports treating them as adjectives and not verbs.

Third, a reduced relative clause would not be able to appear as the complement of a copular verb as in (3b). These observations provide arguments against treating participios activos as reduced forms of active verbs.

Additional arguments that participios activos are indeed adjectives and not reduced active perfective verbs comes from the distribution of the quantifiers mucho (=a lot) and muy (=very). Observe, for instance, the difference between the active perfective verb ha viajado (=has traveled) and the participio activo, viajada (=well-traveled) in (6). The former only admits the verbal/nominal quantifier mucho (=a lot) whereas the latter only admits the adjectival quantifier muy (=very).

In sum, even though participios activos have exceptional argument structure properties, they have the distributional characteristics of canonical adjectival participles and cannot be treated as reduced forms of active perfective verbs.

2.2 Implicit arguments of the verbal source

Much work on adjectival participles has come to the conclusion that part of the process of deriving an adjectival participle from a verbal source involves existentially binding the covert event variable of the source verb and creating a resultant state, and, in some cases, existentially binding the implicit argument variable that corresponds to the agent argument of a passive (see Bruening 2014; Koontz-Garboden 2010; Kratzer 2000; McIntyre 2013; Meltzer-Asscher 2011 for different ways of implementing this). The examples in (7) are derived from passive transitive verbs and describe resultant states of the painting and tying events, respectively. In addition, the implicit agent argument of the passive source is argued to be existentially bound and this is what licenses the presence of by-phrases such as in (7a) as well instrumental PPs that must be under agentive control as in (7b) (see Bosque 1999; Bruening 2014; Gehrke and Marco 2014; Meltzer-Asscher 2011; McIntyre 2013 for different accounts of such modifiers).

The examples in (8), on the other hand, are derived from the unaccusative sources sentar-se (=sit down) and congelar-se (=freeze), respectively, and thus do not admit, in these contexts, agent-oriented modifiers. These examples involve existential binding of an event variable only and not an implicit agent argument variable.

Participios activos also have implicit arguments of a verbal source, but there are important differences in how these argument variables are bound in the conversion of verb to adjective. I will show in the next sub-sections that participios activos differ in two ways from canonical adjectival participles. First, instead of having implicit agent arguments that are bound by an operator in the conversion from verb to adjective, they have implicit theme arguments (see Borgonovo 1999; Di Tullio 2008; Felíu 2008; Piera and Varela 1999; Varela 2002, 2003, 2008). This characteristic motivates treating participios activos as adjectival antipassives (AAs), rather than adjectival passives. Second, this theme argument variable as well as the event variable of the verb are unselectively bound not just by existential operators, but also by habitual and generic operators in some cases. The type of operator that binds the theme and event argument variables of the verbal source divides AAs into three different aspectual classes that receive the descriptive labels resultatives, habitual perfectives and tend-to properties. Each of these will be treated in turn.

2.3 Modification

An additional area in which canonical adjectival participles can be compared with AAs concerns modification. Many diagnostics of the internal structure of adjectival participles are based on the types of agent and event-oriented modifiers that are licensed by adjectival participles in addition to how the dimension of degree quantifiers is determined. Let us consider how such modification facts apply to AAs.

As discussed at the beginning of the previous section, part of the motivation of deriving canonical adjectival participles from passive source verbs is that they can be modified by agent-oriented modifiers such as by-phrases (cf. example 7 above). PP modifiers that identify the implicit theme argument of AAs are also possible as in (23).

However, as is the case with by-phrases in adjectival participles derived from passives (see references at the beginning of Section 2.2), these PP modifiers are not productive as shown in (24).

The type of modification in (23) marks a similarity between AAs and canonical adjectival participles, but they differ with respect to event-oriented modifiers. Many adjectival participles derived from passive and unaccusative verbs admit adverbial modifiers that target some aspect of the event that is described by the source verb as in (25) (see Bruening 2014; Koontz-Garboden 2010 and Meltzer-Asscher 2011 for more examples from English).

All types of AAs, on the other hand, seem to systematically reject event-oriented modifiers as shown in (26).

However, some AAs are compounds that contain a bound adverbial modifier with an event-oriented interpretation as in (27).

In (27) bien is a bound manner modifier that is required to form this particular tend-to property AA, which means that the lawyer ‘tends to speak well.’ Compounds such as (27) are rare in Spanish but more common in English canonical adjectival participles and AAs (cf. Bowers 2010; McIntyre 2013).

Finally, what AAs do readily admit are degree quantifiers that have event-oriented interpretations. This characteristic of some degree terms has been noted by Bosque (1999) and Kennedy and McNally (2005) for canonical adjectival participles like (29).

In (29a) much does not appear to be interpreted as fixing a degree of ‘admired-ness’ but rather seems to be sensitive to the verbal argument and event structure that they inherit. For instance, it could mean that ‘a lot of people admire the statesman’ or that ‘he has been admired for a long time.’ Likewise, in (29b) muy is not intensifying a property of ‘visited-ness’ but rather indicates that the museum is visited often or by many people. Kennedy and McNally (2005) explain this variety of readings in the following terms:

Any of the various aspects of verb meaning that support measurement (temporal extent, number of occurrences, number of participants, intensity, etc.) can be used to fix the dimensional parameter of the derived adjective’s scale. We may assume that any particular adjectival form (needed, admired, etc.) is compatible with several dimensions, one of which must be settled upon in a context of utterance. (Kennedy and McNally 2005: 364)

In AAs, degree morphology is sensitive to a variety of dimensions, some of which are clearly related to the verbal structure from which they are derived and others that are more appropriately understood in terms of the derived state or property that is formed. Consider the examples in (30).

The dimension of the scale in (30a) may be the amount of alcohol, which is derived from the verbal structure, or the state of drunkenness that results from the drinking event. In (30b), the dimension of the scale can only be understood in terms of verbal structure from which the adjective is derived. It may refer to the number of places to which María has traveled or the amount of times that María has traveled to a particular area, or both. However, there is no sense in which the degree modifier may be understood as fixing a degree on a scale of the property of “traveled-ness.” Finally, the dimension of the scale in (30c) is one of intensity. It does not appear to refer to the number of times the comedian entertains nor the amount of people that he tends to entertain.

In this section, it has been shown that, like canonical adjectival participles, AAs also admit different types of modification that provide indirect evidence of their derivational history. Like canonical adjectival participles, they admit PP-modifiers that identify the implicit verbal arguments and give rise to event-oriented interpretations of degree quantifiers. On the other hand, AAs are more limited in their ability to take adverbial event-oriented modifiers than canonical adjectival participles.

2.4 Productivity and semantic drift

Some important facts that will not have escaped the reader are that AAs are semi-productive within certain semantic fields such as verbs of consumption, learning, traveling and object-experiencer psychological verbs and also exhibit a strong tendency toward ‘semantic drift’ ^[5] with respect to their source verbs. The exceptional argument structure properties of AAs, coupled with their lack of productivity and tendencies to yield unpredictable meanings, could be used as grounds to treat them as atomic/stored lexical entries, as opposed to canonical adjectival participles, which are more productive and show more regularity in terms of meaning. On the other hand, gaps in productivity of canonical adjectival participles have been noted for both passive and unaccusative source verbs (Bruening 2014; Kratzer 2000; Meltzer-Asscher 2011; McIntyre 2013; among others) and ‘semantic drift’ or ‘lexicalization’ of canonical adjectival participles has been documented in many languages (Bosque 1999; Bruening 2014; Di Tullio 2008; Horvath and Siloni 2008; Meltzer-Asscher 2011). So, in principle, it should be possible to account for lack of productivity and ‘semantic drift’ in AAs in the same way that they are accounted for with canonical adjectival participles. I will briefly discuss two issues related to lack of productivity and ‘semantic drift’ in AAs and show how these are not problems unique to AAs, but to adjectival participles in general. This being so, this does not constitute a reason to treat AAs differently in either a lexicalist or syntactic theory of adjectival participle formation.

Let us start with lack of productivity, using resultative AAs as a case in point. Recall from Section 2.1 that a generalization regarding this class of AAs is that only verbs capable of taking unspecified objects in episodic contexts can have resultative AAs. However, this generalization, as is, over-generates. For example, it does not explain why verbs outside of the consumption class that have missing objects in episodic contexts do not give rise to resultative AAs. For instance, there are no resultative AAs for creation verbs such as pintar (paint) and tejer (sew) and performance object creation verbs such as bailar (dance) and cantar (sing), which is shown in (31).

Thus, there must be an additional restriction imposed, perhaps related to idiosyncratic semantic features of groups of lexical items or certain diachronic aspects of lexical items that are not obviously accessible to modern speakers, in order to account for which resultative AAs are actually acceptable and which ones are not. This is reminiscent of constraints on unaccusative inputs to adjectival participle formation described in McIntyre (2013). Not all unaccucative verbs give rise to adjectival participles as shown in (32).

This means that internal argumenthood is a necessary but not sufficient condition for giving rise to an adjectival participle and there must an additional condition that filters out the unacceptable forms in (32). In sum, the lack of productivity in AAs is not unique in the realm of adjectival participles.

Let us now consider the issue of ‘semantic drift’ that is present in many AAs. For resultative and habitual perfective AAs, there are few, if any, cases of semantic drift that go beyond metaphorical extension of the state or property derived from the verbal source. For instance, fumado (lit. smoked, ‘stoned/high’) might be extended to domains that include states that do not involve smoking something but apply to behavior reminiscent of someone who has (= ‘crazy’, ‘out of it’). Likewise, viajado (= traveled) is a property that requires that the traveler actually have some in depth knowledge of the places she has been to. On the other hand, there are many cases of ‘semantic drift’ in tend-to properties. Two clear examples are agarrado (‘stingy’, derived from the verb meaning ‘to grasp’) and negado (‘useless’, derived from the verb meaning ‘to deny). If agarrado were to maintain the meaning of its source verb, we expect the generic characterizing statement in (33) to have a meaning akin to ‘not spend money’, but this is not the case.

Instead, agarrado has a meaning as an adjectival participle that is not directly derived from the verb agarrar (=grasp). However, as mentioned above, there is nothing about this particular phenomenon that singles out AAs with respect to other adjectival participles. As Bosque (1999: 283) notes, the adjectival participle autorizado (lit. ‘authorized’=‘expert’) has a special meaning when it combines with the head noun opinión (=opinion) that is not related to the verb autorizar (=to authorize).

To conclude this section, we have seen that AAs are not fully productive and can exhibit significant semantic drift with respect to the verbal source. However, these characteristics can also be observed in canonical adjectival participles, and thus should ideally receive a similar or identical explanation.

2.5 Summary

The main objective of this descriptive section has been to demonstrate that participios activos are adjectival antipassives (AAs) and that they share many properties with canonical adjectival participles but differ in two key respects: they have ‘external’ verbal arguments and the event and argument variables in their source verbs may be bound by one of three flavors of an aspectual operator (existential, habitual or generic). Approximately 100 participios activos have been described in the literature cited in this section. This, in addition to the fact that many speakers interpret and use novel participios activos that are not registered in any dictionary (cf. Felíu 2008) indicate that they constitute more than a small list of exceptions to the general rule regarding adjectival participles and should be integrated into general theories of the phenomenon. Table 1 summarizes the properties of participios activos with respect to canonical adjectival participles.

3.1 Lexicalist analyses

There are two lexicalist analyses of AAs in the literature. Bresnan (1995) first noted that English has adjectival participles with agent arguments as in (35).

She uses these examples, in addition to other limits on adjectival participle formation, to motivate an analysis of adjectival participles in terms of semantics rather than syntactic structure. Specifically, the argument of an adjectival participle is not necessarily the direct internal argument of a verbal source, but rather is contingent on whether the argument has undergone a change that yields a salient result state. While this analysis could be applied to the resultative AAs and habitual perfective AAs described in Sections 2.1 and 2.2, it is unclear how it could be extended to habitual perfective and tend-to property AAs, which do not involve result states. For instance, the interpretation of the tend-to property AA aburrido (lit. bored,=“boring”) does not describe the result of having bored other people.

Varela (2002, 2003, 2008) claims that AAs are derived by merging an adjectivizing head with a verbal structure in a lexical component of the grammar. In these works, Varela argues that AAs are one possible result of merging a stativizing head with a verb’s lexical conceptual structure (LCS, see Levin and Rappaport Hovav 1995). In Varela (2008), she claims that the stativizing head, which is realized by participial morphology in most cases, is [-eventive] and can combine with at least three different LCS structures as shown in (36).

The examples in (36a) and (36b) correspond to passive and unaccusative bases of canonical adjectival participles. The structure in (36c) is meant to derive various aspects of AAs. The proposal is that AAs originate as verbs that are decomposed into the predicate do and two arguments, one corresponding to an experiencer and one a null theme. The predicate do describes an undelimited manner of acting. Its undelimitedness is the source of the label [-tel] while its manner component is represented as a modifier of the whole structure. AAs work as follows: the derived adjective describes a state that is attained by the experiencer in the LCS of (36c) through doing the action described by the verb many times or with a certain intensity. Since the verb is not inherently delimited, part of the conversion process from verb to adjective is determining some bound on the number of times/intensity with which the action must be performed. According to this analysis, this is the function of the manner predicate in the LCS in (36c). For AAs such as entendido (= learned) that do not require modifiers, the amount of understanding is determined contextually and manner has no modifier. However, in other cases such as bienhablado (= well-spoken), the manner modifier is necessary to establish some boundary on the event in the conversion from verb to adjective.

Varela’s (2002, 2003, 2008) approach attempts to account for all AAs as if they were habitual perfectives, which seems plausible for entendido (lit. understood,=learned) but not for bienhablado (= well-spoken) or for resultatives such as comido (lit. eaten,=full)). Note that bienhablado and its English counterpart well-spoken, do not describe someone who has come to attain the state of speaking well through many acts of speaking well in the past. It simply describes someone who tends to speak well. Treating it otherwise is somewhat confusing and it leads to the questionable move of linking bien (= well), a manner modifier, with some delimitative capacity in the conversion of the verb to an adjective, which it does not appear to have.

While these two lexicalist analyses of AAs can account for some of them, they do not provide adequate coverage of the full range of AAs that were presented in Section 2.2. The main problems stem from not separating distinct classes of AAs from each other.

3.2 Syntactic analyses

There are two syntactic analyses of adjectival participles that have addressed the English examples first noted in Bresnan (1995): Embick (2004) and Bruening (2014). Embick (2004) claims that examples like a confessed killer are examples of adjectival participles that are not derived from a verbal source, which are called ‘stative participles’ in the proposed system. In order to understand the status of stative participles, it is necessary to go into some details regarding how participles are formed in this theory. They are created by merging an Asp head with a verbal structure or directly with an acategorial root, which yields three possible types as in (37).

The differences between these three types of participles are accounted for by claiming participial morphology is the spell-out of different aspectual heads that have different selectional properties and meanings. In eventive passives, Asp selects an agentive verb that projects no specifier. The agentive verb contains an agent predicate that may be modified with a by-phrase, which accounts for why eventive passives admit by-phrases whereas resultative and stative participles do not. Resultative participles contain a distinct aspectual head, Asp_R, which selects a different type of verb, v_[fient]. This verb describes an event of change and introduces the argument that undergoes the change in its specifier. Asp_R stativizes the event described by v_[fient]. This is meant to account for the observation that resultative participles do not admit by-phrases but do admit event-oriented modifiers such as carefully (37b). Finally, stative participles contain a third type of aspectual head, Asp_S. This aspectual head is root-selecting and creates an adjective from a root. Unlike eventive passives and resultative participles, stative participles do not entail an event.

Embick (2004) suggests that cases like (35) are instances of stative participles that involve Asp_S merging directly with a root. He adopts an idea first proposed by Alec Marantz (see Marantz 2013 and references therein) that the first categorizing head that merges with a root defines a domain of special form and meaning. Since Asp_S merges directly with a root, idiosyncrasies in form and meaning are to be expected. One such idiosyncrasy in meaning is the fact that the adjectival participles in (35) have agent arguments. While this account could potentially be applied to the examples in English cited by Bresnan (1995) and explain why event-oriented modifiers are not admitted in these cases, many difficulties arise if we attempt to apply it to the Spanish data presented in Section 2. From the formal perspective, all AAs have verbal theme vowels (which are glossed as tv in the lists presented in Section 2) and some derivational verbal morphology. Within the Distributed Morphology (DM) framework employed by Embick (2004), it has been argued that verbal theme vowels and derivational verbal morphology are vocabulary items that spell out the presence of a v head (see Embick and Noyer 2007 and especially Oltra-Massuet 1999, 2014 and references therein). If AAs are treated as stative participles, the consistent presence of a verbal theme vowel and/or derivational verbal morphology is unexpected. From the interpretative perspective, it was shown that, even though there are limitations on event-oriented modifiers in AAs, there exist systematic relations between the events described by active transitive and unergative verbal sources and derived AAs that would be left mainly to chance if these were underived adjectives. For these reasons, I rule out treating AAs as stative participles.

Another syntactic approach to AAs based on English data is outlined in Bruening (2014). Unlike Embick (2004), Bruening (2014) argues that all adjectives with participial morphology are derived from a verbal source which may be either a passive transitive or an unaccusative. The main claim in Bruening (2014) is that all examples of apparent active adjectival participles in English are really instances of passives. They appear to be active participles because the internal argument that is externalized during the adjectivization process is either co-referential with an agent (= reflexive) or is the argument of a covert passive causative verb. For instance, in (35a), well-prepared students are “students_i who have been prepared ____i well (by themselves)”, in (35b) (un)declared majors are “students_i who have (not) been made ___i to declare a major” and in (35c) a confessed killer is “someone_i who has confessed to being ___i a killer.” Once the meanings of these derived adjectives are properly understood, it is possible to maintain the strong generalization that all adjectival participles are derived from verbal structures in which an internal argument is ‘externalized.’ Bruening (2014) presents a plausible argument that nearly all of the English examples that appear to be active adjectival participles could be derived from the types of passive structures mentioned above, but the Spanish data presented in Section 2, when considered in its entirely, is less amenable to this suggested analysis.

Let us first explore the possibility that covert passive causatives are behind what we observe for AAs in Spanish. If on the right track for Spanish, it would mean that resultative AAs like estar comido (= lit. be eaten,=“be full”) would be derived from a structure meaning something like “someone_i who is in a state of having been made ____i to eat (something)” (see Di Tullio 2008 for a suggestion along these lines). However, there is no entailment that the argument of comido (=eaten) or bebido (= drunk) in addition to newer forms such as fumado (=smoked), cenado (=dined), almorzado (=lunched) or desayunado (=having eaten breakfast), is an undergoer that has been ‘fed’ or ‘made to consume’, even though this is possible in some cases. If these were all derived from covert passive causatives, we would expect a strict entailment pattern in which an implicit external argument distinct from “consumer” would always be interpreted or somehow identifiable. I am aware of no such entailment pattern for resultative AAs generally. In addition, it would be very difficult to extend this idea to habitual perfective AAs and tend-to property AAs. Habitual perfective AAs such as muy leído (= well-read) do not involve the presence of an implicit external causer that “makes someone read many things.” Likewise, tend-to property AAs are clearly not derived from covert passive causatives that yield a resultant state. When confiado (= trusting) is used, it is does not entail that the argument of the adjective “has been made to overly confide in people.” In conclusion, while the possibility of analyzing some cases of AAs as coming from covert passive causative structures, this is not a promising way of accounting for the general phenomenon described in Section 2.

Alternatively, all instances of AAs may be derived from reflexive interpretations of passive transitives as in the case of well-prepared mentioned above. The main problem with this idea is that, as discussed at length in section 2, the direct internal argument of the source verb is always implicit in AAs and there are systematic connections between unergative verbal constructions in certain tense/aspects and AAs. If AAs were derived from reflexive interpretations of passives, we would expect the internal argument position of the verb to be linked to the external argument position of the adjective, but this is not the case in any of the classes of AAs we have seen: comido (=someone who has eaten, not someone who has been eaten (by himself)), leído (=someone who has read, not someone who has been read (by himself)) and apañado (=someone who fixes things, not someone who has been fixed (by himself)).

3.3 Summary

In this section, lexicalist and syntactic analyses of AAs in Spanish and related types of examples in English have been discussed in terms of how they might account for the data presented in Section 2. While each of the accounts is able to explain some of the examples in section, there is no account that can capture the entire range of AAs as discussed in Section 2. As shown here, when considered in its entirety, the data in Section 2 represent a problem for current theories of adjectival participle formation regardless of whether or not modular distinctions are appealed to. Since the main objective of this paper is simply to outline an analysis that can capture these basic properties, I will not take a stance on this issue in the next section and will propose an analysis of AAs by modifying the syntactic framework of Embick (2004) and Bruening (2014).

4.1 Types of AAs

In order to accurately account for these three types of AAs, I propose that Asp in (38) comes in three ‘flavors’: Asp_R, Asp_H-P and Asp_G. Note that this is completely in line with both Bruening (2014) and Embick (2004), which are both crucially based on the idea that Asp/Adj heads have different selectional and semantic properties. What is proposed by Bruening (2014) is particularly relevant for the task at hand. His Adj head binds the event argument of a verb and the external argument variable when it selects Voice_tr but it only binds the event argument when it selects Voice_intr since there is no external argument variable in adjectival participles derived from an unaccusative verb. In both cases, the head Adj creates a resultant state by binding one or more unsaturated variables that are part of the meaning of the verbal base. I claim that the Asp head responsible for creating AAs does the same kind of work. Like Bruening’s (2014) Adj that selects Voice_tr, it unselectively binds an event and argument variable. However, unlike Bruening’s Adj, Asp does not always create a state or property by existentially binding all of the unsaturated variables in the verbal construction it selects. The labels on the different types of Asp correspond to the operators that bind the unsaturated variables in the verbal structure that Asp selects. These are described in (40).

Proposing different flavors of operators yields a very simple explanation for why three types of AAs exist and it is in line with many approaches about the semantics of the conversion process from verb to adjective. As mentioned in Section 2, a leading idea in the research on adjectival participles is that the event variable of the verb (and its external argument variable if it is passive) is bound by an existential operator. The idea that this type of variable binding takes place in the creation of adjectives from verbal bases has also been proposed by Oltra-Massuet (2014) for –ble adjectives, which involve existential binding and modalization of the event argument of the verb. Thus, different types of adjectives result from different ways of binding the event variable of the verb in the process of conversion from verb to adjective. In addition, it has been proposed independently that different languages may link operators to different types of morphology (see Lekakou 2005 for an account of modals that is based on different ways in which generic operators are linked, or not, to tense morphology in Greek/French versus Germanic). Therefore, it is not entirely unexpected that a language may have different ‘flavors’ of an operator whose main function is to unselectively bind variables in the conversion process from adjective to verb and that some variation between languages (like Spanish and English) will be observed in this respect.

Another positive consequence of (40) is that the generalizations outlined in Section 2.2 regarding the types of verbal constructions that provide the “inputs” to the different classes of AAs can also be easily accounted for. The proposal is not that the AAs are actually derived from tensed verbs but rather that the way in which the interpretative component of the grammar (the Encyclopedia in DM) assigns an interpretation to a given root with no object in combination with a type of aspectual/tense operator is the same across different types of constructions (verbs and adjectives in this case). For instance, the interpretation of comer (= eat) in the absence of a theme in past tense is “eat some quantity of food.” The same interpretation is assigned to this root when it appears in a resultative AA as in (41b). What matters here is that both constructions share a VoiceP and the same type of operator that binds the event and internal argument variables introduced below VoiceP.

I argue that this is what enables us to account for the generalization identified in Armstrong (2011) and discussed in Section 2.2: if there is no interpretation that can be assigned to missing object in an episodic contexts as a verb then there will also be no interpretation assigned to it as a resultative AA. This is what is shown in (42).

To close this section, I propose that a natural way of accounting for copula selection in AAs is by linking copula choice with the type of Asp head. While all canonical adjectival participles are treated as resultatives and combine with estar, the stage-level copula verb form adjectives, this is not the case with AAs. Asp_R gives rise to a stage-level adjectives with a resultative interpretation that combines with estar while Asp_H-P and Asp_G both give rise to individual-level properties that combine with ser as shown in (43) (see Camacho 2012 and Fábregas 2012 for recent overviews of the ser/estar distinction and SL/IL distinction in Spanish). As far as I can tell, this has not been addressed with much attention in the literature. The properties of the three aspectual operators proposed here yields a straightforward account of the difference.

4.2 Modification

I will remain agnostic regarding where PP and adverbial modifiers merge with respect to AspP (externally or internally) since these issues are still to be worked out in any theory of adjectival participle formation (see Gehrke and Marco 2014; Meltzer-Asscher 2011 and references therein for many important observations and proposals regarding by-phrases and adverbial modification of adjectival passives). The modifiers in bold could be treated as modifiers of the state/property created by the adjectivizing head in (44) or as adjuncts of a verbal projection below the adjectivizing head.

Since such modifiers are subject to semantic restrictions based on particular roots, the details of which have not be fully worked out, whatever ends up being the best way of accounting for their presence in (44) would also extend naturally to examples like (45).

The main point here is not to provide an explicit and detailed theory of how event-sensitive modification of adjectival participles works, but rather to point out that the same kind of modification exists in both canonical adjectival participles and AAs. Thus, regardless of how these questions are worked out in a lexicalist or syntactic theory of adjectival participles, the solution should be equally applicable to AAs. An important question that deserves further attention is why such modification is much more limited in AAs than in canonical adjectival participles (even though it is not fully productive in this latter class either).

Unlike PP and adverbial modifiers, degree modification is very productive and even obligatory for some speakers with some AAs. The most interesting fact about degree modification as mentioned in Section 2.3 is that that in many AAs, the dimension of the scale that provides the degree argument of the adjective is determined in terms of some aspect of the source verb. As discussed in Kennedy and McNally (2005), this would be quite unexpected if some derivational process has not occurred. The example in (46) (repeated from (30) above) shows that degree terms may target different dimensions associated with verbal source or the derived state/property).

I will follow Bosque and Masullo (1998) in assuming that degree quantifiers may merge externally to a category they modify but be linked to a meaning component that is part of the derivational history of that category. In the case of (46a), bastante is a degree quantifier the merges in the specifier of Deg, which is part of the functional domain of the derived adjective. The dimension of the scale provided by the adjective could be the amount of alcohol drunk (a component inherited from the source verb) or the state of being drunk.

Finally, there are cases such as (46b) in which the absence of a degree term makes the AA somewhat unnatural. I propose that, especially in the case of habitual perfective AAs, the property described by the derived adjective is only felicitous if it holds to a high standard on the scale provided by the AA. So, the adjectives leído (= read) or viajado (= traveled) are only interpreted as holding to high degrees and thus require a degree modifier muy to be used felicitously.

4.3 Productivity and semantic drift

As mentioned in Section 2.4, an important difference between AAs on the one hand and canonical adjectival participles on the other is that the former are not productive outside small groups of verb roots that are related to the same semantic fields (consumption, learning, traveling, object-experiencer verbs, etc). This has been used to relegate them to the status of unstructured, ‘lexicalized’ adjectives. However, we have seen in the previous sections that such an analysis misses a great number of properties that AAs share with other canonical adjectival participles. The question is how we can reconcile these two apparently contradictory facets of AAs.

While future diachronic and experimental studies are needed to confirm the remarks that follow, I would like to propose as a starting point that AAs are marked with respect to canonical adjectival participles and they most likely originated as special adjectival uses of particular participles that fit into a certain semantic categories in which a resultant state or property could be easily applied to a “doer” rather than an “undergoer” of the verbal source. Consumption verbs provide potentially salient resultant states that hold of the consumer, thus these are apt roots for giving rise to the interpretations that the grammar assigns to resultative AAs. If a special use spreads by analogy to other related verbs and then appears in ones that are not semantically related, as is the case with resultative AAs, we might posit that there is a degree productivity gained by the structure underlying the AA and that speakers do use a similar mechanism in forming and interpreting these adjectival participles as they use with other ones. The same idea could also apply to habitual perfectives and tend-to properties, but in these cases, it is the assignment of some attained state of status (through repetition of some action) that is culturally relevant or the possession of some property (in terms of a type of action that is performed) that is culturally relevant.

While this idea is mere speculation at present, it is worth mentioning that such semantic restrictions appear to be operative in canonical adjectival participles as well. For instance, verbs that do not entail changes of state, do not, at least in the absence of a very specific context, give rise to adjectival participles (Bosque 1999; Bresnan 1995; Kratzer 2000; McIntyre 2013; among others)

Taking these observations into consideration, I claim that lack of productivity of AAs is due to a semantic/conceptual restriction on how the output of the syntactic structure that is built on top of the root is interpreted by the Encyclopedia (in DM) and not because the syntactic process that builds AAs in the first place is impossible or non-existent. To conclude this part, the example in (48), which is repeated from (31) above, is unacceptable because there is no concept in the list accessed by the Encyclopedia that can be appropriately linked to the resultative AA that is built on top of the root cant- (sing) that means “a person who has sung a lot.”

Finally, it was also noted in Section 2.4 that AAs exhibit semantic drift with respect to their source verbs. We saw above that this is by no means a unique characteristic of AAs with respect to other adjectival participles. Since this is so, I will follow Bruening (2014) in treating semantic drift as simply a form of idiomatic interpretation. For Bruening (2014), all adjectival participles, regardless of whether they have special interpretations with respect to verbal sources or are derived from “missing” verbal inputs, are still deverbal adjectives derived from a passive or unaccusative source. Applying this idea to cases such as agarrado (= stingy), negado (= useless) and others, the conclusion is that these AAs have the same syntactic structure as any other tend-to property AA. The main difference is that the entire structure defined by the AspP constituent has a special interpretation.

4.4 Summary

In this section an analysis of AAs has been presented that accounts in a straightforward way for the basic properties presented in Section 2. This marks a major improvement with respect to previous analyses that, for the most part, relegate AAs to a small group of idiosyncratic adjectives with no internal structure that are only formally related to other, more productive adjectival participles. The analysis outlined here extends the syntactic approaches of Bruening (2014) and Embick (2004) to languages in which “past/passive” morphology can actually be used to derive an adjective from an active transitive or unergative verbal source. Motivation for this move has been presented throughout this paper and it was shown here that all of the properties of AAs follow from this and other general principles that are used in DM to explain the properties of canonical adjectival participles.