-Able adjectives and the syntax of psych verbs

This paper deals with some restrictions on the formation of -able adjectives from object experiencer verbs in comparison to subject experiencer verbs, focusing on English and Greek. Building on Oltra-Massuet (2013), the paper assumes that there are two places of attachment for -able, a high one, which combines with a structure including passive Voice, and a low one, which combines with a smaller structure. While subject experiencer verbs combine with low -able in both languages, the behavior of object experiencer verbs is not uniform. The unavailability of high -able formation with object experiencer verbs is correlated with the unavailability of passivization. In English, only those object experiencer verbs that yield a well-formed passive can combine with high -able. In Greek, OE verbs do not form passives or -able adjectives. The differences between English and Greek are accounted for by appealing to differences in their Voice systems, with specific reference to passive formation.


Introduction
The class of psychological verbs has been controversially discussed in the literature, as they pose problems for theories of the argument structure-syntax interface. A first issue is the fact that they can project the experiencer theta-role both as an object or as a subject, thus creating a problem for thematic hierarchy and linking (see e.g. Belletti & Rizzi 1988;Grimshaw 1990;Pesetsky 1995;Arad 1998;Anagnostopoulou 1999;Pylkkänen 2000;Reinhart 2001;Verhoeven 2008;Landau 2010 and many others). A second problem is their aspectual behavior. In particular, they are ambiguous between eventive and stative readings, and there is no agreement as to the exact categorization of the eventive interpretation these may receive (activity, achievement or accomplishment; see e.g . Grimshaw 1990;van Voorst 1992;Tenny 1994;Martin 2006;Rozwadowska 2007;2012;Landau 2010;Marín & McNally 2011;Alexiadou & Iordăchioiaia 2014 among others).
While the paper does not directly address these issues, it shows that the formation of -able adjectives from psych-verbs depends on aspectual as well as structural properties of these verbs. Because of this, -able formation can be used as a further tool to explore the properties of psychological verbs. To illustrate this, consider (1). As (1) shows, in English certain psych verbs can build -able adjectives, while others not, see also Trips & Stein (2008). (1a) is a so-called subject experiencer verb, while (1b-c) are object experiencer verbs, see section 2 for details: (1) a. lovable b. annoyable c. *depressable/*depressible not on the basis of -sim. -Sim-affixation is disallowed with the class of object experiencer predicates, and only a sub-class thereof can combine with -tos, once additional prefixal material is added, as we will see in section 3.2: (2) a. agapitos b. *enohlisimos c. *enohlitos lovable.masc annoyable.masc annoyable.masc This behavior leads us to formulate the following questions: i) what enables -able formation with certain psych verbs in English? ii) why is it impossible to form -able adjectives in Greek with the same verb class on the basis of -sim-? iii) what explains the differences between English and Greek object experiencer verbs (OE)? I will show that -sim-affixation in Greek applies only to verbs that yield a well-formed passive. With respect to psych verbs, I will demonstrate that in Greek no psych verb of the OE class can passivize. In fact, not even the ones that have been claimed to be agentive and undergo the causative alternation can passivize. As a result, OE verbs cannot combine with -sim-. By contrast, precisely those English OE verbs that can passivize yield well-formed -able adjectives. Moreover, I will entertain the availability of a further input structure for -able formation out of OE verbs, associated with a reading of "semantic easiness". This reading is correlated with the tough-movement construction.
As detailed in Landau (2010), across languages there are certain specific syntactic properties associated with experiencers, which contribute to making the class of psychological verbs special. Landau labels these properties psych effects. A first contribution this paper makes is to establish a further psych effect, namely the behavior of psych verbs in -able affixation. While it has been occasionally noted in the literature that across languages certain OE verbs do not form -able adjectives, this paper, focusing on Greek and English data, will show that the unavailability of a particular type of -able formation correlates with the unavailability of passivization of psych verbs, echoing the passive analysis of -able, see Chapin (1967); McGinnis (2010); see also Horn (1980); Williams (1981); Kayne (1984); Fabb (1984); Roeper & van Hout (1999) and others. In particular, verbs that do not form a verbal passive do not feed -able formation. This actually holds across the board in Greek, i.e. it is not restricted to psych verbs.
The second contribution the paper makes is to explain the differences with respect to -able formation between English and Greek by showing that they relate to differences in their Voice systems. The third contribution of the paper is to link the so-called semantic "easiness reading" of -able to a structure not discussed in Oltra-Massuet (2013), but which has been proposed for Greek by Samioti (2015). As pointed out in Oltra-Massuet (2013: 27), "-able adjectives can be related to other structures", and I will argue that these other structures may be anticausative.
The paper is structured as follows: 1 In section 2, I will introduce the background for my discussion. This involves a brief summary of the literature on the main classes of psych verbs that have been recognized as well as the view on -able formation that I will adopt in this paper. In section 3, I will turn to the behavior of-able adjectives, argument alternations, tough-movement, and psych verbs in English and Greek. In section 4, I will offer my analysis, and I will conclude in section 5. Belletti & Rizzi (1988) proposed that there are three classes of psych verbs, and this classification has been adopted in subsequent research. The first class is that of subject experiencer verbs, where the experiencer appears as the subject of the clause, (3):

Classes of psych verbs
(3) John loves Mary.
The second class is that of accusative object experiencer verbs, where the experiencer appears as the accusative object of the verb, (4): (4) The message worries John.
The third class involves an object experiencer that surfaces with dative case, (5): The song appeals to John.
There is some consensus among scholars with respect to the status of class I and class III verbs: class I verbs are unambiguously stative, and behave like transitive predicates, and class III verbs are unaccusative across languages. However, Class II has been controversially discussed in the literature, and the reader is referred to Landau (2010) for a comprehensive overview. With respect to class II verbs, the following issues are important for my discussion. Firstly, Class II OE verbs may have eventive, non-agentive interpretations. Arad (1998) shows that there are three different interpretations associated with class II OE verbs, see (6): an agentive reading which includes both an agent and a change of state in the experiencer; an eventive reading implying that something unintentionally caused a change of mental state in the experiencer; a stative reading where there is no agent nor any change of mental state.
eventive, non-agentive d. John's stubbornness frightens Bill. stative Secondly, more recently, Alexiadou & Iordăchioiaia (2014) argued that several OE verbs in English have subjects which are ambiguous between a causer reading and an object of emotion. Importantly, when their subject is interpreted as object of emotion (corresponding to the subject matter/target of emotion role in Pesetsky 1995), the sentence corresponds to a stative interpretation of the psych verb, while when the subject is interpreted as a causer the sentence has an eventive causative use. For these authors, the term causer excludes agents, and refers to non-agentive causers in change of state eventive sentences. The object of emotion role is taken to be introduced very low in the structure, potentially contained within the psych root suggesting that under this stative reading class II OE verbs are unaccusative, see Pesetsky (1995). Alexiadou & Iordăchioiaia (2014) assumed that causers are introduced similar to agents in the structure, following Alexiadou & al. (2006), a claim that I will revisit and modify in this paper. Thirdly, Landau (2010) devotes a lot of attention to the question of whether OE verbs can form a passive. As he points out, two main approaches to this issue can be recognized. According to e.g. Belletti & Rizzi (1988); Grimshaw (1990) and others, OE verbs are unaccusative and thus do not form passives. According to e.g. Pesetsky (1995); Iwata (1995); Pylkkänen (2000) and others, OE verbs are transitive verbs, and thus can form a passive similar to other transitive verbs. Landau's contribution is to demonstrate that there are two types of languages when it comes to passivization of OE verbs. In languages like English only eventive (non-stative) members of class II can have a verbal passive. In languages like Italian, OE verbs have no verbal passive. For Landau, the formation of a verbal passive in English is contingent on eventivity not agentivity. Agentive OE verbs universally form passives, as they are actually change of state verbs (accomplishments), while non-agentive ones are states (depress) or achievements. He agrees thus with Tenny (1998: 595) that "verbal passives are more felicitous the more eventive the verb. A complex of factors influences the degree of eventiveness, including not only agentivity but also volitionality, punctuality, and the affectedness of change of state in the experiencer. A loose gradient can be defined from the purely stative ascription of property to the most eventive verb type […] Individual speakers vary in how strict they are with this scale in making verbal passives." Landau's typology is relevant for my discussion of -able, as it has been suggested by other authors for the non-psych domain that there is a correlation between the availability of passive formation and a particular type of -able affixation, to be detailed in 2.2. Importantly, evidence will be provided for (7): (7) Psych verbs which cannot form a passive are also not able to form an -able adjective implying an originator.
This directly relates to both Landau's partition of psych verbs, and the view that -able formation is contingent on passivization.

Building -able adjectives
As is well known, languages employ several ways to express dispositional statements, and -able adjectives in English as well as their cognates across languages figure prominently among these. That -able adjectives bring about modality is generally acknowledged in the literature, see e.g. Lyons (1977); Bauer (1983); Chierchia & McConnell-Ginet (1990); Krifka & al. (1995); Lekakou (2005); Bauer, Lieber & Plag (2015) among many others. The very rich literature on this topic, recently extensively summarized in Oltra-Massuet (2013), has made several important observations: the first one is that depending on the source verb -able applies to, a difference in modal force emerges. For instance, as Pires de Oliveira & Ngoy (2007), and Moreira (2014) note for Brazilian Portuguese, when the counterpart of English -able, namely -vel, applies to a transitive predicate the derived adjective signals possibility. However, when it applies to a stative verb, it expresses obligation or necessity (e.g. audivel 'able to be heard' vs. admirável 'deserving respect and admiration'). Moreover, similarly to modals there is a difference in entailments. As Giannakidou & Staraki (2013) point out, in (8) there is no entailment that Ariadne did solve the problem, see also Hacquard (2010).
Ariadne can solve the problem. Moreira (2014) and Oltra-Massuet (2013) argue that this also holds for possibility -able adjectives. Something can be breakable without entailing the state that it has been broken.
The obligation reading, however, that emerges in combination with stative verbs is different. As Moreira (2014) notes, if someone is admirable, there is an entailment that he/she inspires admiration, i.e. the state is manifested, see 3.1 for details. What is more interesting for the purposes of this paper, however, is the discussion on the morpho-syntactic contexts in which these two readings emerge. Specifically, there appears to exist a correlation between the two readings and the derivational history of the adjective involved in these. Importantly, the possibility reading is available only out of verbs that are transitive and can be passivized. As Chapin (1967) notes, verbs that do not form a passive do not yield -able adjectives either, (9), see also Marchand (1969); Aronoff (1976) among others. McGinnis (2010) provides more recent evidence in support of this claim, see also Roeper & van Hout (1999). Oltra-Massuet (2013) shows that this holds in Catalan and Spanish as well and similar observations have been made by Riehemann (1993) for German.
(9) a. *Her mother is resembled by Kate. b. *Her mother is resemblable by Kate.
Oltra-Massuet (2013) correlates the two readings with the two distinct derivations for -able adjectives, cf. Volpe (2005) for an alternative proposal. In particular, Oltra-Massuet (2013: 42, 148) argues that there are two heights for -able attachment and these give rise to distinct semantics. This high vs. low attachment of the affix is associated with a number of other properties summarized in From this perspective, the difference in interpretation associated with -able adjectives correlates with their morpho-syntactic make up. High -able adjectives create a generic property, according to which it is possible for some originator to achieve a resultant state. The event interpretation implies an external argument, as in e.g. translatable. This is the reading Oltra-Massuet calls potential. Low -able adjectives, by contrast, have idiosyncratic properties, and express a modality that differs from regular possibility, e.g. admirable, which Oltra-Massuet labels evaluative.
(10) a. translatable = can be translated b. admirable = no direct reference to a verbal component The differences in interpretation often correlate with differences in form, as discussed in Aronoff (1976: 128): (11) a. tolerable = moderately good b. toleratable = capable of being tolerated c. appreciable = substantial d. appreciatable = capable of being appreciated As Oltra-Massuet (213: 111-112) points out, languages such as German have a different affix to express the evaluative judgment, namely wert: Oltra-Massuet and other literature she cites observed some restrictions on by-phrases with -able adjectives. To the extent that by-phrases occur, they must be non-specific and generic. These restrictions are interesting as they are similar to those found in adjectival passives and verbal passives in some languages, as well as dispositional middles in languages such as Greek, where by-phrases are allowed (see e.g. McIntyre 2013; Gehrke 2015 among others; Lekakou 2005; for a summary of the literature and references). We can relate this to the proposal made in the literature that by-phrases when applied to states are subject to sortal restrictions, i.e. they are either responsible for continuing the state or crucial for the nature of the state.
Finally, Moreira (2014) and Oltra-Massuet (2013) note that there are aspectual restrictions that apply to high -able. According to Moreira, high -able combines with eventive predicates, and cannot combine with stative predicates. Oltra-Massuet claims that high -able combines with verbs that involve some originator who achieves a resultant state, but there is no implication that the event has taken place. This is expressed in (15): (15) high -able combines with eventive predicates that involve an originator that contributes to the achievement of a resultant state.
The corresponding structures for high and low -able are given in (16a) and (16b). While both formations are adjectival, a crucial difference between the two is that high -able includes the layer that introduces the external argument. Specifically, high and low -able differ in that high -able involves a modal that takes as an input structure the layer that introduces the implicit external argument, namely VoiceP (Kratzer 1996;. Low -able involves a modal component that simply embeds a non-verbal structure. According to Oltra-Massuet (2013: 153), in the case of low -able in (16a), first the root merges with AspP, which is a stativizer. ModP modalizes the formation, and when it merges with a stative root, the readings that are obtained may be non-potential. By contrast, in the case of high -able in (16b), the pieces involved in the formation yield a reading according to which it is possible for some arbitrary individual to perform the action denoted by the eventive predicate: (16) a. b.
Both -able formations contain a ModP component, relating to ideas concerning the treatment of modal auxiliaries in e.g. Kratzer (1991) and Hacquard (2010): there is one modal element, which can attach high and low. Note here that nothing excludes an intermediate structure excluding Voice but including vP, as has been proposed in the literature on nominalization (Alexiadou 2001), and adjectival passives (see Alexiadou et al. 2015 for discussion and references). This structure will be very similar to that of resultative participles referring to a resultant state but lacking an external argument. As will see in section in sections 3.1 and 4 this might be the correct analysis for the "semantic easiness" reading associated with -able.
In view of all this discussion, it becomes clear why -able formation is relevant for the psych domain. (15) and the structures in (16) illustrating the two heights of attachment for -able make certain predictions: assuming Landau (2010) is right that certain OE verbs do not form verbal passives, we expect high -able formation to be out with exactly these predicates. In addition, if high -able does not combine with stative predicates, we expect to find such a restriction in the psych domain as well. In other words, to the extent that -able adjectives can be formed out of psych verbs, we expect correlations between eventivity/stativity, transitivity, (un)availability of passivization and -able formation.
Let me consider these issues in detail in the next section by focusing on English and Greek. See Oltra-Massuet (2013) for some discussion of German, Bisetto (2013) for Italian, Wood & Sigurðsson (2014) for Icelandic, and Bloch-Trojnar (2017) for Polish.

-Able adjectives and psych verbs
In this section, I will discuss the formation of -able adjectives from psych verbs in English and Greek. My focus will be on class I and class II psych verbs. This comparison is necessary in order to control for the stativity and transitivity factor, since class I verbs are unambiguously stative and transitive. As we will see, in English, low -able formation is generally possible out of class I psych verbs. With respect to OE verbs, -able formation is contingent on two conditions: first, high -able is licit only if the verb can form a passive. Second, certain OE verbs can build -able adjectives receiving a reading labeled in the literature "semantic easiness" or more precisely contributing "adverbial information of easiness" (Hamada 2013: 172). This is possible only for those OE verbs that can have agentive interpretations. Low -able is possible only for transitive stative verbs of class I and not possible with stative OE verbs, which are unaccusative. I will argue that low -able affixation is prohibited in this case, because no state holder is present in the structure.
Greek is similar to English with respect to the behavior of class I, but, unlike English, it disallows high -able attachment altogether with OE verbs. Greek restrictively allows for the semantic easiness reading, if this is overtly expressed via a prefix.

English
Formation of -able adjectives with Class I verbs is generally possible, see (17): 2   (17) admirable, hateable, enjoyable, likeable Recall that this class is considered to contain stative verbs which have a transitive syntax. In Kratzer (1996), it is proposed that the subject of such verbs is the holder of the state and is also introduced in VoiceP. As briefly mentioned in the previous section, Moreira (2014) argues that the two interpretations of -able, potential vs. evaluative, correlate with stativity and eventivity. With stative verbs, the state is manifested, while with eventive ones the state is not manifested. The potential reading of -able is possible with eventive verbs, while the evaluative reading of -able is possible only with stative verbs. Consider the contrast in (18a-b), her examples (11-12). Moreira does not compare an eventive psych verb with a stative one, but her stative verbs all belong to class I psych verbs: (18) a. O vaso é quebrável, mas não está quebrado. 'The vase is breakable, but it is not broken'. b. ?? Mimi é amável, mas não desperta amor/afeição. 'Mimi is lovable, but does not inspire love/affection.' As Moreira observes, in (18a), the adjective does not entail the state. (18b) is different in that a loving state is entailed. Importantly, however, the state holds for the state holder, which corresponds to the experiencer argument. The target of emotion, following Pesetsky (1995), is evaluated by the experiencer. As Moreira (2014: 192) states, "a possibility modal reading is available for adjectives derived from eventive verbs (such as quebrável 'breakable', lavável 'washable', congelável 'freezable'). These adjectives involve objective properties and the final state may hold for the object of the verb (quebrado 'broken', lavado 'washed', congelado 'frozen'). This state is not manifested. The possibility modal reading is unavailable for adjectives derived from stative verbs (amável 'lovable', adorável 'adorable', notável 'notable'). These adjectives involve subjective properties and the state holds for the subject of the verb (an experiencer). This state is manifested." Formation of -able adjectives from OE verbs shows a non-uniform behavior. For instance, the verbs in (19) do not seem to be able to form -able adjectives: 3 Other predicates, however, such as irritate or terrify seem to be able to form -able adjectives, see (20). There is evidence to support the claim that in most cases low -able is involved. First, the form that occurs in the -able adjective must be truncated: Second, -able affixation sometimes leads to stress shifts: Thirdly, they do not seem to tolerate by-phrases, and they can be modified by very: Thus, on the basis of Oltra-Massuet's criteria, at least for these OE verbs that can form -able adjectives, low -able seems to be involved. We note a correlation between the aspectual and structural properties of OE verbs, and their ability to form -able adjectives. As Alexiadou & Iordăchioaia (2014) pointed out, building on e.g. Landau (2010); Grafmiller (2013) and references therein, there is a subgroup of these verbs that is stative only: e.g. fascinate and depress. This sub-group does not really give good -able formations. As we have just seen, low -able is available with class I stative verbs, but why can low -able not attach to stative OE verbs? The reason for this is that there is an important difference between the class I and class II stative verbs: class II stative verbs are not transitive, as shown extensively in Landau (2010), i.e. both the experiencer and the object of emotion are internal arguments. Thus, there is no state holder for which the state could hold. 4 Nevertheless, forms such as irritable are possible, which, as we have seen, morphosyntactically qualify as involving low -able. However, and unlike -able adjectives formed from class I verbs, irritable does not have the interpretation in (23b). Importantly, the state is not manifested and there seems to be no contradiction in (23c): John is irritable. = a. John is easily made angry. 5 ≠ b. John deserves/inspires irritation. c. John is irritable, but he is not irritated.
Hamada (2013) points out that in this reading -able contributes adverbial information and is paraphrasable with the PP with ease. 6 If adverbs generally modify events, this would point to the presence of an eventive layer in this structure. As discussed in Oltra-Massuet (2013: 106), this reading is close to that of X is easy to irritate, i.e. a tough-movement construction (TMC), 7 see also Val Álvaro (1981) for Spanish. This kind of interpretation led several researchers to propose that -able adjectives are derived from dispositional middles.
The reason for this is the similarity between TMCs and dispositional middles. Specifically, Massam (1992: 124-126) observed that both TMCs and middles involve stative predicates, require a modal element referring to the easiness, possibility of the event referred to, and they both ascribe a property to the subject, i.e. a certain property of the subject makes it easy to wash in (24): (24) a. This blouse is like a dream to wash. b. This blouse washes like a dream.
These properties are also recognized in Oltra-Massuet as characteristics of -able. However, Oltra-Massuet (2013: 142)  In what follows, I will compare these three structures in the domain of psychological verbs. In particular, I will examine to which extent a particular OE verb can form both a dispositional middle and a TMC, and whether there is a correlation between the aspectual class the OE verb belongs to and its behavior in middles, -able formations and TMCs, as not all aspectual verb classes yield well-formed dispositional middles. Importantly, statives do not yield well-formed dispositional middles. Note that only transitive verbs enter both TMC and dispositional middle formation, so the behavior of OE verbs in this domain can be taken as a further argument for their (in)transitivity. There is a sharp contrast between fascinate and irritate, (26a) is bad while (26b) is fine. A similar contrast is observed for (26c) vs. (26d): (26) a. * John is easy to fascinate. b. John is easy to irritate. c. * John fascinates easily. d. John irritates easily.
I noted above that fascinate is a stative verb. There is a further group of OE verbs in English, which could also be classified as states in the presence of an object of emotion argument, as discussed in Alexiadou & Iordăchioaia (2014). Evidence for this comes from the incompatibility of several of these verbs with in-adverbials, indicating that they lack a change of state reading. If we employ the easy to V paraphrase, we see that these verbs cannot be included in these paraphrases, (29): (27) a. Sue grieved over the court decision for/*in half an hour. b. Sue grieved at the court decision for/*in half an hour. c. The court decision grieved Sue for/*in half an hour.
(28) a. We puzzled over Sue's remarks for/*in an hour. b. Sue's remarks puzzled us for/*in an hour.
These verbs do not yield good dispositional middles either, (30a), and -able affixation is out, (30b): (30) a. ??Sue puzzles easily/grieves easily. b. *grievable *puzzleable If these verbs are stative, our remarks above concerning fascinate also hold. Finally, in English there is a class of predicates that are ambiguous between a change of state and a stative reading. Alexiadou & Iordăchioaia (2014: 73-74) show that this is supported by for-adverbials, which may modify the result state (RS) of the change of state reading documented in (25), or the single eventuality (SEv) like with pure states.

(31)
The Chinese dinner satisfied Bill for ten minutes. i. RS: After having the Chinese dinner, Bill was satisfied for ten minutes. ii. SEv: Having the Chinese dinner kept Bill satisfied. Both the dinner and his satisfaction lasted for ten minutes simultaneously.
These verbs give well-formed middles, as well as TMCs, (32). Verbs that belong to this group should be able to combine with high -able, if they are change of state, and in principle their -able adjectives should be ambiguous between a passive reading and an easy to V reading: (32) a. John is easily satisfiable. John is easy to satisfy. John satisfies easily. b. John is so annoyable. John is easy to annoy. John annoys easily.
In support of this, change of state OE verbs have the potential reading, i.e. the state is not manifested and as a result the statements in (33) do not involve a contradiction: (33) a. John is satisfiable, but he is not satisfied. b. John is annoyable, but he is not annoyed.
We can summarize the results of the comparison as follows. -able formation from English OE verbs is possible, if these are not stative. Two readings are available: the easiness reading, which seems to involve morpho-syntactically low -able, but excluding the manifestation of the state, suggestive of an event component. This class of verbs is also the one that can participate in a TMC. This is suggestive of the transitivity of the OE verb. A passive reading is also possible for those verbs, which, according to Landau (2010), can form a verbal passive. This latter behavior obeys the generalization in (15). As already mentioned, several researchers have pointed out the similarity between this reading of -able adjectives, dispositional middles, and TMCs. While states and achievements typically do not undergo middle formation, this restriction does not hold for TMCs. Nanni (1980) observed that the complement of a tough predicate must be volitional or intentional with respect to its subject, e.g. note the contrast in (34): The man was hard for Mary to find attractive/*sick.
Nanni also shows that stativity is not the relevant factor, and while most ungrammatical examples involve stative predicates, there are several cases where a stative predicate is acceptable in the complement of tough: The book is hard to understand.
The so-called "semantic easiness" reading is available in all these three domains. But how does it arise? It has been observed by Berman (1973) that the implicit argument of the tough adjective is construed coreference with the implicit argument of the embedded verb. Keine & Poole (2016) propose that with tough-predicates this implicit argument corresponds to the judge of the proposition. In other words, a proposition is tough according to some judge, which is co-referential with the implicit argument of the embedded verb. This judge can be overtly expressed via an experiencer PP, which specifies the judge, (36a), or be left implicit, interpreted generically, (36b): (36) a. The book is easy for John to read.
b. Hockey is fun to play.
Building on these observations, we could argue that the "semantic easiness" associated with -able plays a role similar to that of adverbs such as easily in middles and tough type adjectives in TMCs in Lekakou's (2005) account. Lekakou (2005: 161) proposes that in English middles "event modifiers are required in order for the middle-Agent to be recoverable via identification with the experiencer argument contributed by the adverb." 'Semantic easiness' performs the same job, it licenses the agent of the verb that undergoes -able formation. The modifier itself need not be overtly expressed, although it is often the preferred option. Most of the -able examples illustrated here contain such an adverb. The adverbial paraphrase of this reading when it is covert is claimed in Hamada (2013) to be with ease, which requires an agentive eventive verb (as noted in Klecha 2014). In the TMC, the implicit judge corresponds to the experiencer argument of the tough-adjective.
As irritate can in principle have an agentive subject and a transitive construal, its implicit subject is coreferential with this judge. In the case of fascinate this is not possible, as no agent is available. Thus, the former can be embedded in a TMC, while the latter cannot. Similar considerations hold for -able formation: "semantic easiness" is available only for agentive and eventive OE verbs, and thus irritable is acceptable but *fascinateable not. 9 Does this mean that irritable is analyzable in terms of (16b)? The morpho-syntactic criteria suggest that this is not the case. In fact, Oltra-Massuet (2013) explicitly argues that all non-potential readings of -able signal low -able. Thus, irritable has the structure in (16a) and as a result can also have the idiomatic reading in a bad mood. Alternatively, and to the extent that its meaning is becoming angry with ease, there is an event causing irritation, which can be modified via a PP. This reading must involve a structure between (16a) and (16b), namely one lacking Voice but including a v layer introducing the event. Importantly, however, the agent is not projected in the syntax, but is conceptually available, licensed by the easiness interpretation, as has been argued to be the case with dispositional middles, see Schäfer (2008) for details. 10 In conclusion, high -able does not combine with stative OE verbs. The combination of high -able with OE verbs is contingent on the availability of passivization. There seems to exist an intermediate -able available for agentive OE verbs that applies to those verbs that can also be embedded under a tough predicate. With this in place, let me now turn to Greek.

-Tos vs. -sim-
In Greek, there are two ways to form dispositional adjectives. The first one is discussed extensively in Samioti (2015) and it involves the affix -tos, which forms potential adjectival participles, sometimes in combination with certain prefixes, e.g. axio 'worthy'. Samioti argues in detail that ability/possibility -tos participles involve high -tos, as they can license by-phrases and manner adverbials: a. I istoria tu ine pistefti apo olus. the story his is believable.fem by all 'His story is believable by all.' b. To mathima ine efkola katanoito.
the lesson is easily understandable.neut 'The lesson is easily understandable.' The second one involves the affix -sim-: Note that (38c) does not necessarily denote that something can be eaten. In Greek, this form, especially in the plural, is used to refer to food in general. (38ac) has an interpretation similar to the one associated with high -able, as we will see immediately below. Applying Oltra-Massuet's criteria, we see that -sim-formations as in (38a) involve high -able affixation, i.e. structure (16b). First of all, -sim-adjectives license manner adverbs: To vivlio ine efkola metafrasimo. the book is easily translatable.neut 'The book is easily translatable.' They also license by-phrases as well as aspectual phrases: (40) To vivlio ine metafrasimo apo ebirus metafrastes. the book is translatable.neut by experienced translators 'The book is translatable by experienced translators.' (41) To vivlio ine metafrasimo mesa se deka meres. the book is translatable.neut with in 10 days 'The book is translatable in 10 days.' That -tos in examples of the type in (37) has a potential reading is argued for in Samioti (2015: 76). As she points out, similar to -able, the adjective does not imply that the event described has taken place.

(42)
To asteri ine orato apo ti gi, ala de to ehi di kanis akomi. the star is visible.neut from the earth but neg it has seen anyone yet 'The star is visible from the earth, but noone has seen it yet.' We can make a similar observation for -sim: To vivlio ine metafrasimo ala den to ehi metafrasi kanis akoma. the book is translatable.neut but neg it has translated anyone yet 'The book is translatable but nobody has translated it yet.' We can thus conclude that -sim-patterns with high -able. Zombolou (2004: 129) notes that in Greek formation of adjectives via -sim-is only possible with a particular verb class, which happens to be that one that forms passives as well. Specifically, Zombolou (2004: 130) observes that the following change of state verbs cannot form passives and they cannot form -able adjectives either: 11,12 (44) shizo 'tear' *shisimo 'tearable' keo 'burn' *kapsisimo 'burnable' lerono 'dirty' *lerosimo 'dirty-able' In view of this, it is necessary to offer a more systematic discussion of Greek -able adjectives and passivization, before turning to psych verbs.

The Greek Voice system
As has been discussed in detail in the literature, passive formation in Greek is restricted. The way to form a morphological passive is via non-active Voice morphology, which marks passive and reflexive verbs, as well as the anticausative member of verbs undergoing the causative alternation and dispositional middles, (45), (see Tsimpli 1989;Embick 1997;Zombolou 2004;Lekakou 2005 11 Zombolou further notes that the same class of predicates does not form -er nominals in Greek: (i) shizo 'tear' *shistis 'tearer'  suggest that only verbs based on manner roots form -er nominals. This raises the question of whether the restriction is similar for -sim-. 12 Note here that Zombolou (2004: 130) discusses a further affix used in Greek to derive adjectives of ability, namely -ik-: this affix derives adjectives that have the meaning X who V-es, i.e. eksipiretikos 'person who serves'. This affix can attach to certain Class II predicates, e.g. endiposiakos 'impressive', tromahtikos 'terrifying', enohlitikos 'annoying', but crucially no potential reading is involved in this case. As Zombolou points out, this affix can derive adjectives out of verbs that do not form a passive. With some of these verbs, there again seems to be a morpho-phonological clash that results from the combination of a particular stem with the non-active affix, e.g. (48b). I note again that none of these verbs combine with -sim-. Most of these verbs are internally caused change of state verbs which lack external arguments and in general do not form -able adjectives in other languages either, see Oltra-Massuet (2013) for Catalan.
As Zombolou (2004) further points out, passivizability is also restricted outside the domain of change-of-state verbs. For instance, the following (mono-eventive) verbs cannot easily form a passive in Greek (or not at all, for some speakers), while they can in English: haidevo 'stroke', derno 'beat', klotsao 'kick', frondizo 'take care of'. These verbs do not combine with -sim-. It could be argued that the problem with these verbs is the lack of a resultant state component, which is required by (15), nevertheless this restriction can be relaxed in English, see (49), reminiscent of the so-called job-done interpretation in the domain of adjectival participles: Wales had to be given a kickable penalty. (COCA, New York Times, 2013) Finally, there are several verbs, which, while they can combine with non-active morphology, they cannot receive a passive interpretation e.g. burn, cut and kill, see  for details. As shown in (50), agentive apo-phrases are not tolerated with such verbs, i.e. they only form anticausatives.
(50) I supa kaike me ti dinati fotia/ *apo to Jani. the soup.nom burnt.nact with the strong fire/ from the John 'The soup burned from the strong fire.' Crucially, none of these verbs combines with -sim-. The above data thus provide support for the observation that Greek -sim-formation is only allowed if a non-active form interpreted as passive is also available.

Greek psych verbs and -able
Turning now to psych verbs and dispositional adjectives derived from those, note that Class I predicates can combine with -tos but not with -sim-, (51-52), all forms bear masculine gender. 13 (51) agapi-tos zilef-tos misi-tos thavmas-tos lov-able jealous-able hate-able admire-able (52) axiolatreftos axiozileftos axiosevastos worthy-admired worthy-jealous worthy-respect 'worthy of admiration' 'worthy of jealousy' 'worthy of respect' As Samioti (2015) details, the forms in (51) have an interpretation comparable to their English counterparts, e.g. zileftos means 'worthy of jealousy', an evaluative reading. In (52), I hold that the prefix axio modifying the adjectives lexicalizes this worthy aspect of interpretation involved in the evaluative reading. 14 Thus, Greek class I predicates do not behave differently from their English counterparts.
13 There is a third archaic affix, -teos, which is no longer used productively in Greek. This, as discussed in Samioti (2015), and Haspelmath (1994), is interpreted as passive. Psych verbs do not seem able to combine with this affix either. 14 This is similar to the Hungarian facts discussed in Oltra-Massuet (2013: 111). The presence of this prefix enables, according to Samioti (2015), the licensing of a state holder, introduced by the preposition apo 'from': (i) I Maria ine agapiti se olus the Mary.nom is lovable.fem to all 'Maria is lovable to all.' OE verbs, on the other hand, cannot combine with -sim-: (53) a. *enohlisimos b. *thimosimos c. *sinhisimos annoyable.masc angerable.masc confuse-able.masc With very few exceptions, they do not combine with -tos either. The form disarestos means unpleasant or causing grief. Very often the -tos affix co-occurs with a prefix/adverb meaning 'easily' (as Despina Oikonomou pointed out to me). The examples in (54b-d) suggest that Greek transparently marks the semantic easiness reading, which, as we have seen is available in English in e.g. irritable.
As Samioti (2015) shows, these forms do not entail that the event has taken place, and they cannot refer to a specific event.
(54) a. disarestos b. everethistos c. ev-prosvlitos d. efkolo-siginitos unpleasant easily-irritable easily-assailable easily-moveable Samioti further notes that the adjectives in (54) to involve an anticausative structure and not a passive one: by-phrases are disallowed, but the by itself phrase is allowed; this test diagnoses the absence of an external argument, as discussed at length in Alexiadou et al. (2015): (55) O Janis ine everethistos apo monos tu/ *apo olus. the John.nom is easily-irritable.masc by himself/ *by all While Greek certainly has stative OE verbs, e.g. interest, which do not combine with -sim-, most Greek class II verbs are eventive causative and have a resultant state, see (56-56') from Alexiadou & Iordăchioaia (2014: 67). As these authors argued in detail, in Greek several psych verbs undergo the causative alternation, i.e. they are change of state verbs. Evidence for this is provided by the availability of a restitutive reading for the Greek counterpart of again, ksana, which detects the presence of a resultant state in the structure: Importantly, while verbs as in (59a) are change of state predicates, they do not include an agent/originator in their syntax, just a causer argument. Such arguments are modifiers of the event and are not introduced in a manner similar to external arguments. Specifically,  follow Solstad (2009) and view causers as introducing an event that can be identified with an event already introduced by the verb. From that perspective, causers are unlike agentive external arguments, which the authors take to be introduced in VoiceP. Assuming that the presence of Voice signals transitivity and thus the availability of passivization, it follows that verbs involving a causer argument do not project this argument in VoiceP, and thus cannot undergo passive formation. This was 15 See also the discussion in Landau (2010) on the third type of Hebrew passives associated with psych verbs.
Like their Greek counterparts, they do not tolerate agentive by phrases. Landau assumes that the preposition me introduces an internal argument. In principle, it could be a causer argument, see Alexiadou & Doron (2012), and in Doron (2017) for psych verbs in particular: (i) Gil hitrageš me-/ *al-yedey ha-seret. Gil was-moved of/ *by the-movie 'Gil was moved by the movie' argued for in Alexiadou (2014) for transitive internally caused change of state verbs, e.g. age, which only admit causer subjects, see section 4 for details.
There is evidence supporting the claim that causative OE verbs in Greek are not really transitive and thus cannot undergo passivization. Anagnostopoulou (1999) shows that relativization involving experiencers is identical to indirect object relativization regardless of whether the experiencer has dative or accusative case. The experiencer cannot undergo relativization unless a clitic is present inside the relative clause. While Anagnostopoulou did not take into consideration the stative vs. eventive ambiguity of these predicates, it is important to note that the clitic is required, regardless of whether the psych-construction is of the "piacere-class" (Class III), the stative "preoccupare-class" (stative Class II) or the eventive "preoccupare-class" (eventive Class II) ( The fact that a resumptive pronoun is obligatory is evidence for an approach to accusative experiencers, according to which their licensing differs from that of structural accusative objects. Specifically, Landau (2010) discusses the Greek facts and comparable facts from Hebrew as well as many other languages. In his system, such data support an analysis of experiencers as locatives, introduced by a zero preposition, regardless of whether they surface with dative or accusative morphology. Importantly, the data discussed by Anagnostopoulou suggest that change of state psych verbs of class II in Greek are not really transitive predicates. Because of that, they cannot passivize and do not combine with high -able. 16 Further evidence for the absence of an agentive external argument with psych verbs is provided by the interaction with modals. Giannakidou & Staraki (2013) show that there is a lexical split in Greek between the impersonal bori -an epistemic possibility modal form, something like might in English -and personal boro which is never epistemic, but abilitative or deontic.
(61) a. Ta pedia bori na ine sto spiti. the children might.3sg subj be.3pl to-the home Epistemic: 'As far as I know, it is possible that children are at home.' b. Ta pedia borun na pane sto spiti mona tus. the children can.3pl subj go.3pl to-the home alone them Ability: 'Children are able to go home on their own.' Deontic: 'The children are allowed to go home by themselves.' OE verbs that can have a Non-Active form related to an anticausative reading and lack a -sim-adjective cannot appear in the latter context, i.e. they can only combine with epistemic modality, (62a). This is in sharp contrast with the predicates that can form -simadjectives and passives, in (62b) The possibility of embedding a particular verb under an epistemic or ability modal correlates with the availability of an external argument. As Hackl (1998) notes, verbal passives, which arguably include an implicit external argument, are fine under an ability modal, while stative or adjectival passives, which lack such an argument are not. Thus, one gets only the epistemic reading for can with an adjectival passive, and semi-modals are ungrammatical with an adjectival passive. This is particularly clear in German that distinguishes between the two passives morphologically, see (63-64) In Greek, the difference between epistemic and ability modality is signaled by the different morphology on the modal, agreement vs. lack of agreement. The Greek data in (62) are thus consistent with the proposal that causative and eventive OE predicates lack an (implicit) external argument in VoiceP.
Turning now to the -able "semantic easiness" reading, I note that this is not possible in Greek with -sim-. The counterpart of English irritable under this reading is the form we saw in (54b), everethistos, which does not contain a passive structure. Greek lacks this particular reading with -sim-, as -sim-requires a passive input structure, i.e. it patterns with high -able. Note also that examples such as 'John is easy to please' are simply ungrammatical in Greek. In this language, a clitic is required in the embedded clause, see Tsimpli (1999) for discussion. 18 If the "semantic easiness" reading in English is available for those predicates that arguably have an agentive argument, which is not projected in the syntax, we have just seen several pieces of evidence that Greek OE verbs lack such an argument altogether.

Towards an analysis
In section 2.2, we have seen the structures proposed for high and low -able in some greater detail, taken from Oltra-Massuet (2013). These are repeated below: (16) a. b.
As discussed in section 3, nothing in this system excludes a structure containing a vP but excluding VoiceP, as has been argued for the domain of participle formation and nominalization.
In both English and Greek, subject experiencer verbs may combine with low -able, i.e. structure (16a). The psych root is embedded under modality. Thus, for those only the evaluative reading of -able is possible. In Greek, as argued for in Samioti, certain OE verbs make use of an intermediate anticausative structure, lacking Voice and containing vP, (65a). Only low -able appears, i.e. Greek -tos, see the examples (54b-d). The morphology of these adjectives in Greek supports the presence of a v layer, (65b). Specifically, forms such as everethistos 'easily irritable' contain -iz-, which is a verbalizer affix, realizing v, as argued for in Alexiadou (2009). In English, the semantic easiness reading can also be associated with this anticausative structure.
b. ev-ereth-iz-tos easily-irrit-verbalizer-affix In English, high -able adjectives derived from OE verbs involve the structure in (16b). A version of (16b) has been proposed by Samioti (2015) for potential -tos, see (66), and we can adopt it here for -sim-, with a number of refinements to be specified momentarily relating to VoiceP.
Samiotis's analysis builds on Lekakou (2005), who argued in detail that in Greek the dispositional Middle is actually built on the basis of a passive structure. More recent work on the Greek Voice system, following Alexiadou & Doron (2012), Spathas et al. (2015), and Alexiadou et al. (2015), assumes that dispositional middles, and passives in Greek all involve a particular Voice head, namely Middle Voice. This head is realized with nonactive morphology, which signals the absence of a specifier of Voice (Embick 1997). In other words, Greek Non-Active Voice is underspecified (see also Tsimpli 1989;. On this view, the Greek Voice system lacks a designated passive head, and all constructions lacking a (syntactically projected) external argument (passives, dispositional middles, a subset of anticausatives, and reflexives) are syncretic, surfacing with non-active morphology. English differs from Greek in that it has a designated passive head that takes VoiceP as its complement. English type passives thus are unambiguous. This particular view explains the restrictions observed for the Greek passive: the availability of the passive depends on properties of the roots or roots + v combinations. This is expected if the head implicated in Greek passives is close enough to the Root + v combination so that it can access it.
Recall that we established in the previous section that in Greek there is a correlation between the formation of a verbal passive and that of a -sim adjective. The verbs that do not combine with non-active Voice also do not combine with -sim. Basically, if the input to this affix is a well-formed Middle Voice structure that can be interpreted as a passive, then we expect to find the same restrictions. It has been suggested that the affix -sim-is related to non-active morphology, as speculated in Oltra-Massuet (2013), since it contains the infix -m-that authors identify as a signal of non-active morphology (Alexiadou 2001).
An important conclusion of the previous section was also that class II OE verbs are not transitive in Greek (i.e. do not include VoiceP), though they might be causative. If they are not transitive, they involve a structure as in (67) below, which is the structure proposed in Alexiadou (2014) for transitive internally caused change of state verbs, e.g. skuriazo 'rot' which also take causer arguments only and do not form passives: On the basis of (67), we expect that these psych verbs cannot form agentive passives, as also observed by Oikonomou (2011), in the light of Landau's (2010) and Anagnostopoulou's (1999) discussion. Thus, they are unable to enter any formation that would involve the presence of a Middle Voice head of the type described above, i.e. a head that is involved in the formation of agentive passives. Intransitive variants of these verbs are thus only anticausative and never passive. As a result, -able formation proceeds on the basis of an input structure as in (65a).

Conclusion
In this paper, building on Oltra-Massuet, I argued that the restrictions on -able formation in the domain of psych verbs relate to three factors, namely: i) different domains for affix attachment, ii) properties of event structure, and iii) transitivity and the availability of passivization for psych verbs. High -able formation is only possible out of transitive change of state verbs that give a well-formed passive.
I used -able formation as a further tool to explore the structure of psych verbs. I provided evidence for the view that OE psych verbs cannot form -able adjectives for the same reason that they cannot form a passive. This supports the classification of these verbs as unaccusative in their eventive causative reading as well as the observation that Greek belongs to the group of languages where psych-verbs of Class II do not passivize.
I have also discussed a second reading available with -able adjectives, that of semantic easiness. This particular reading suggests a correlation between -able formation and TMCs and seems to make use of an anticausative input structure. This particular issue awaits further investigation.
With respect to the issue of transitivity, it is clear that we need to distinguish between external arguments introduced in VoiceP and causers introduced in vP. Only the former structure feeds passivization.