Ditransitive ‘teach’ and the status of the Theme “argument”(?)


 This paper offers new insights into the much-debated topic of double accusatives, taking διδάσκειν as a case study. By focussing on the different syntactic and semantic properties of the two accusatives in expressions such as διδάσκω σε σωφροσύνην ‘I teach you moderation’, it is shown that the mere reference to distinct semantic roles (Recipient vs Theme) does not provide a satisfactory account of some crucial properties of these constructions. As emerges from textual evidence, the so-called “Theme” may alternate with an infinitival complement (e.g. διδάσκω σε σωφρονεῖν ‘I teach you (how) to exert moderation’). Both the infinitival and the nominal complements are bound to the object of διδάσκω through a relationship which may be defined by the notion of control. This finding reveals the predicative function performed here by the “Theme”, thus supporting a multi-predicative approach to the double accusative construction of διδάσκειν. This hypothesis has relevance to the analysis of other double accusatives of Ancient Greek, and opens a new path for the analysis of comparable data offered by modern languages.

ferent alignment patterns that these constructions display cross-linguistically. According to a common definition in typological literature, a "ditransitive construction is defined […] as a construction consisting of a (ditransitive) verb, an agent argument (A), a recipient-like argument (R), and a theme argument (T)" (Malchukov, Haspelmath & Comrie 2010: 2). The way in which these semantic arguments are encoded across languages is usually described in terms of variable alignment patterns, defined by comparing the encoding of R[ecipient] and T [heme] with the encoding of the monotransitive Patient (P). In brief, we have indirective alignment when T is treated like P, and differently from R; secundative alignment when R is treated like P, and differently from T, and neutral alignment when both T and R are treated like P (Malchukov, Haspelmath & Comrie 2010: 2;Haspelmath 2008Haspelmath , 2015Malchukov 2017). In the double accusative construction, where both arguments take the accusative case, i.e. the same case as the monotransitive P, we have thus an instance of neutral alignment.2 According to Malchukov (2017: 183), identical case marking for R and T is favoured by their semantic asymmetry (in terms of animacy / referentiality), which rules out ambiguity in the interpretation.
It is also worth mentioning that the double accusative construction has been repeatedly claimed to be somehow related to semantic causativity (Humbert 1960: 259;Jacquinod 1989: 185;Luraghi & Zanchi 2018); this idea recalls some recent models of lexical decomposition (cf. Dowty 1979; Van Valin 2001, who identify an abstract operator cause in the semantic of verbs such as 'teach' = cause [become know]).
The achievements of typological and theoretical linguistics have undoubtedly had a valuable impact on the analysis of such a traditional topic as double accusative verbs in Ancient Greek. However, as we believe, traditional (mainly text-oriented) analyses may also provide a stimulus to further inquiry, which would in turn produce interesting theoretical implications.

3
Διδάσκειν as a case study: some data Verbs denoting cognitive transfer typically occur in ditransitive constructions in a variety of languages, as repeatedly observed in the literature. Among them, 2 The double accusative construction of διδάσκειν-and of Ancient Greek double accusative verbs in general-does not admit alternative alignment patterns. In other words, it does not present anything comparable to the so-called dative shift which is very common, e.g., in English. For a survey, see Haspelmath (2015); against the assumption of dative shift in Modern Greek, see Efstathopoulou (2007). one should mention verbs for 'teaching': within the Indo-European domain, the double accusative construction occurs not only in Greek (διδάσκειν) but also in Latin (docere) and German (lehren); for recent accounts, see Napoli (2016) and Lee-Schoenfeld & Diewald (2017) respectively, with further references.3 In our analysis, we selected Ancient Greek διδάσκειν 'teach' as a case study in order to reflect on some issues related to the asymmetry between the two accusatives. A more focused approach is expected to yield interesting results.
The double accusative construction of διδάσκειν is widely attested in all sorts of Greek texts, at different epochs:4 (2) τὸν dem:acc ἡ art:nom The data collected in our corpus are consistent with the traditional assumptions concerning passivization,5 even though it must be stressed that evidence is scanty.6 The examples of passives with both T and R expressed (which provide contrastive evidence for the expected passivization of R over T) are very few. The Homeric passage frequently quoted in the literature is inconclusive: it is an accusative and infinitive clause, with both T and R in the accusative case and verb in the infinitive (not carrying subject agreement feature): this does not allow us to determine with certainty which of the two is the clause subject; for which, see (5) More reliable evidence comes from passages like (6) and (7), in which the passive subject is undoubtedly R: Here again we have an Accusative and infinitive clause; but, unlike what we saw in (5), only T is expressed (ῥητορικήν 'rhetoric'). So ῥητορικήν might be taken as the passive subject (in the accusative case, according to the usual Accusative and infinitive pattern). However, this is not the only possible reading: alternatively, one might suppose the infinitival clause to have a null subject, coreferential with the preceding ἄλλους (the object of διδάσκοντες): 'and teaching these things to others (ἄλλους), they believe that those (= the others) have been taught rhetoric completely by them' (as we may observe, the Loeb edition opts for an active traslation). The quantitative data on passivization extracted from our corpus are summarized in Table 1.9 7 The passive agent is here unexpressed (as is often the case in Ancient Greek texts) and the middle verb form διδάξομαι might in principle be assigned a reflexive reading ('I shall teach myself'). However, this hypothesis is to be excluded in the specific context: the situation described clearly involves a mutual exchange between the two participants, Menelaos and Neoptolemos, each one exposing his reasons and listening to the reasons of his interlocutor. 8 In all these cases the passive aorist stem διδαχθη-occurs (excluding a reflexive reading). 9 The question mark refers to the passage in (9), for the reasons given above. We class as a As we may see, the overwhelming majority of passive structures undoubtedly involves the promotion of R to subject. Moreover, R-promotion is also attested under co-occurrence of T, whereas the reverse does not hold. This fact is probably significant: the acceptability of T as the passive subject might actually be related to the absence of a better candidate for subjecthood, namely R. This would be in line with what has been observed in other languages, including Modern Greek: here "passivization of the theme across a goal [= Recipient: mb] in the double accusative construction is impossible" (Anagnostopoulou 2001;cf. also Efstathopoulou 2007).10

4
Beyond the state of the art: the Theme "argument" and infinitival complements The data we have been discussing so far fit into the general pattern outlined both by traditional historical grammars and by recent research. In what follows, we shall try to go a step beyond this, and suggest a new approach. Let us take into account a different construction that is widely attested since the earliest documents up to later times. Here, διδάσκειν takes an infinitival complement clause, besides a R in the accusative, as in (10)-(12): non liquet example (5) above and two passages in which a reflexive reading is not excluded (Sol. fr. 18 and Ar. Nub. 127; in both διδάσκω, in the present, has mediopassive morphology, which allows both a passive and a reflexive reading). 10 More generally, in the literature on ditransitives it has been observed that "R-passivization is generally preferred over T-passivization" (Malchukov, Haspelmath & Comrie 2010: 30 We can say that T is a distributional alternative to an infinitival complement. This can immediately be illustrated by comparing cases showing lexical identity between the predicate of the complement clause and the nominal complement T. For example, the passage in (10) above can be compared with (13) The pair (10)-(13) is noteworthy because of the close thematic affinity, an affinity which has been repeatedly observed in the literature; see Arrighetti (1992); González (2013); Stern-Gillet (2014); Semenzato (2019). In both, Hesiod describes the poetical training he received from the Muses, which inspired the composition of his poems, respectively the Works and days and the Theogony.
In epic poetry, a further nice example is offered by (14), which should be compared with (12)  As we may see, as a complement to [διδάσκω σε] 'I teach you' , the accusative ἔργα 'works' (T) offers an alternative to the infinitival clause ἔργα ἐργάζεσθαι 'to do works' (with a figura etymologica).
Finally, let us observe the passage in (15), to be compared with (12)  Cf. Stern-Gillet (2014). 12 The figura etymologica ἀοιδὴν ἀείδειν is attested, e.g., in Aes. 268.2, 10; E. Tr. 514; X. An. 4.3.27.3. Needless to say, when one works with corpus languages, any inference on possible (but not attested) forms and on their interpretation is highly hypothetical. Nevertheless, any linguistic analysis inevitably runs the risk of surpassing the limits of documentary evidence in order to create an experimental domain and formulate hypotheses on general patterns. The data "constructed" here are based on a careful investigation of the documentary evidence (which, in the case of Ancient Greek, is actually rich) and are-in our view-the result of reasonable hypotheses.
Journal of Greek Linguistics 20 (2020)  ἀνὰ for πᾶσαν each:acc ἡμέραν day:acc σωφρόνως temperately διάγοντας live:ptcp.acc.pl 'They teach the boys moderation also; and it greatly conduces to their learning to exert moderation that they see their elders also living temperately day by day' (X. Cyr. 1.2.8.1 ss.) As a complement to [διδάσκω σε] 'I teach you' , the accusative σωφροσύνην 'moderation' offers an alternative to σωφρονεῖν 'to exert moderation' . This emerges not only from the comparison with (12) but also from the wider context of (15): the effect of teaching the boys moderation (σωφροσύνην) is described as their learning to exert moderation (μανθάνειν σωφρονεῖν). Given the close semantic and syntactic affinity between μανθάνειν and διδάσκειν, the expressions διδάσκειν σωφροσύνην '(to) teach moderation' and μανθάνειν σωφρονεῖν '(to) learn to exert moderation' constitute a very close pair, with the noun and the infinitive alternating in a sort of stylistic variatio.
In the examples selected above (to which others might be added) we see that there is lexical identity between the nominal complement (i.e. the theme T) and the clausal complement. Similar cases immediately suggest a comparison between the two types, thus inviting us to further inquire into a far broader phenomenon, which goes far beyond the instances of lexically related pairs in paradigmatic alternation (in absentia) such as διδάσκω σε σωφροσύνην~σωφρονεῖν.
Evidence at the syntagmatic level (in praesentia) shows that the two complement types may cooccur within a compact textual unit and may even be coordinated (cf. (16) and (17) respectively).
If we go back to the passage in (15), which describes the education of Persian boys, and we look at a wider portion of this text, we may observe three occurrences of διδάσκειν (here conventionally identified as a, b and c): καὶ and ποτοῦ drink:gen 'They teach the boys moderation also […]. And they teach them likewise to obey the officers […]. And besides, they teach them selfrestraint in eating and drinking' (X. Cyr. 1.2.8.1 ss.) In a sequence of strictly parallel sentences introduced by διδάσκουσι δέ 'and (they) teach' (one may also notice the gradient in the sequence referring to R, i.e. to the person taught: a full NP in a; an anaphoric pronoun in b; zero in c), nominal and clausal complements alternate: σωφροσύνην 'moderation' , ἐγκράτειαν γαστρὸς καὶ ποτοῦ 'self-restraint in eating and drinking' and, inserted between them, πείθεσθαι τοῖς ἄρχουσι 'to obey the officers' in b.
Very interestingly, Xenophon's passage is quoted in Stobaeus' Anthology with a variation: the accusative NP ἐγκράτειαν γαστρὸς καὶ ποτοῦ 'self-restraint in eating and drinking' , is replaced by the infinitival complement ἐγκρατεῖς εἶναι καὶ γαστρὸς καὶ ποτοῦ 'to be self-restrained in eating and drinking' . One would hardly expect to find a better proof of the possible equivalence between these constructions.
Furthermore, in a stylistically dense passage of Sophocles' Antigone, in which the Chorus praises the extraordinary progresses of human beings, accusative Themes and clausal complements occur side by side: we find three accusative NPs (φθέγμα 'speech' , ἀνεμόεν φρόνημα 'wind-swift thought' , ἀστυνόμους ὀργάς 'temper that rules cities') depending on ἐδιδάξατο ('[mankind] taught himself' , hence 'learned') and-after a break induced by the insertion of the verb form ἐδιδάξατο-an infinitival clause (φεύγειν 'to escape' etc.): These facts, which can hardly be irrelevant for our understanding of the nature of T, are not usually observed in the literature, nor are they easily accounted for in traditional approaches.
Traditionally, the double accusative construction is dealt with in chapters on nominal cases, and the infinitival construction in chapters on infinitives, where no mention is made of the construction with the noun, and vice versa. This attitude clearly reflects the role which long-lasting categorial distinctions (nouns, verbs etc.) continue to play in grammatical and linguistic analysis.
Completely denying the merits of a purely categorial approach would not be fair, but nevertheless we cannot ignore its limits: such an approach is not apt to describe correlations which go beyond categorial boundaries, like the ones we have been establishing here.
In the specific case under analysis, we believe that a comparison with the infinitival construction opens new paths for considering the double accusative in a new perspective.
Of course, the fact that nouns may alternate with infinitives is not surprising, in principle. The nominal nature of infinitives is widely acknowledged, and is confirmed not only by synchronic alternations (like the ones we have been discussing here) but also by diachronic developments (such as the increase of productivity of verbal nouns to compensate the decline of the infinitive in the history of Greek; for which, see Joseph 1983: 44-46).
What we wish to stress here is that the alternation between noun complements and infinitival complements is not possible in all contexts. Rather, it is found in specific contexts, mostly related to specific verb classes (on this cf. Cristofaro 2008); verbs for 'teaching' (at least judging from the behaviour of Ancient Greek διδάσκειν)13 belong here.
Obviously, the point here is not simply the fact that [διδάσκω σε] 'I teach you' may take nominal as well as clausal complements. The point is that we are dealing with cases in which "the meaning of the sentence [scil. with NP complements/mb] is similar to that obtained with infinitival complements" (Cristofaro 2008: 578). For example, διδάσκω σε σωφροσύνην 'I teach you moderation' is essentially equivalent to διδάσκω σε σωφρονεῖν 'I teach you to exert moderation' . Or, as we have seen above, διδάσκω σε ἐγκράτειαν γαστρὸς καὶ ποτοῦ 'I teach you self-restraint in eating and drinking' , can be replaced by διδάσκω σε ἐγκρατῆ εἶναι καὶ γαστρὸς καὶ ποτοῦ 'I teach you to be self-restrained in eating and drinking' without any significant change in the interpretation. These facts 13 As a matter of fact, the phenomenon highlighted here on the basis of Ancient Greek data is cross-linguistically widespread (as immediately emerges from possible paraphrases in other languages). We believe that this finding, and the analysis suggested in the next paragraph, can be fruitfully tested on a much wider scale.
are revealing of the affinity between the two complement types, thus raising the issue of a possible syntactic correlate of their semantic affinity.14 5

Control into clauses and control into nominals
The infinitival construction under analysis is syntactically more transparent than that with the double accusative. It is commonly ascribed to the so-called "control" type: this means that the construction imposes obligatory co-reference between an argument of the matrix clause-here, specifically, the direct object-and the (null) subject of the complement clause.15 Semantics can easily account for this property: the sense of διδάσκειν which is relevant here16 involves a semantic component "ability", which-as argued by Jackendoff & Culicover (2003: 538)-identifies a class of control predicates: Another basic semantic predicate that selects an actional argument is 'be able'-a relation between an entity and an action. One cannot have an ability with respect to someone else's performance of an action; that is, 14 This semantic affinity may be stated in terms of identity of the semantic roles assigned by the nominal and the verbal predicate respectively. This does not imply a complete semantic identity. A thorough investigation of the different properties of nominal and verbal predicates (radically differing in the inflectional categories involved) lies beyond the scope of this paper. For important remarks on data from Greek and Latin, see Porzig 1942 andRosén 1981. In principle, a further alternative complement to [διδάσκω σε] 'I teach you' would be a finite clause, e.g. 'I teach you how you should/could exert moderation' . This pattern-which does not seem to occur in Classical Greek, at least to our knowledgeshows up in Modern Greek, where, as is well known, infinitives have totally disappeared (see also n. 16). 15 As is well known, the notion of "control" represents a highly debated issue in linguistic theory, which we need not raise here. For the aim of this paper, the informal definition given above will suffice. On control infinitives in Ancient Greek, see Joseph (1983Joseph ( , 1992Joseph ( , 2002; Sevdali (2013) Διδάσκειν may denote transmission of an ability ('teach how to' , as in the cases we are dealing with here) or transmission of knowledge ('teach that'). The two meanings are clearly distinguished, being associated with different complement types: infinitive on the one hand, finite clauses introduced by a complementizer (e.g. ὡς, as in T. 3.71.2.3) on the other. The two types never formally merge; the contrast survives into Modern Greek under renewed formal expressions, opposing two different types of finite clause, respectively introduced by the modal particle νά ('teach how to') and by the conjunction ότι ('teach that'); cf. Holton, et al. (2012: 547). On the diachronic development leading to the loss of the infinitive and its replacement by finite clauses, cf. Joseph (1983).
the person with the ability must be bound to the actor position in the action.
[…] This predicate is a component of the adjective able, the noun ability, and the root modal can. It is also a component of one sense of learn to VP, roughly 'come to be able to VP' and teach NP to VP, roughly 'cause to come to be able to VP' . Thus learn is a subject control verb and teach is an object control verb.
In Ancient Greek, διδάσκειν + inf. denotes the transmission of an ability, and the person taught coincides with the one who has to perform that ability.17 It is the kind of infinitival clause which is replaced by a finite να-clause in Modern Greek (cf. n. 16): here, the co-reference between the subject of the dependent clause and the R(ecipient) argument of διδάσκειν is explicit through verbal agreement. What about the corresponding double accusative construction? Here too διδάσκειν refers to the transmission of the ability to perform a given action. And here too the teaching is directed to the person (R) who shall perform that action (T).
As we have seen, in the infinitival construction the co-reference between the null subject of the infinitive and the object of διδάσκειν is commonly captured by the notion of control. It would be interesting to see whether the notion of control-which is usually adopted for verbal complements-could be extended into non-verbal (i.e. nominal) complements as well.
A preliminary issue concerns the ability of nouns to perform predicative function, which includes the ability to license arguments. Of course, it is commonly assumed that nouns can project arguments within the internal syntax of their NP, e.g. in the form of a dependent genitive or of a possessive adjective, as in Socrates' moderation or your moderation.18 Besides this, there is strong evidence supporting the idea that, in appropriate contexts, nouns may project their arguments onto clausal syntax. This is the case, for example, with predicative possession constructions, like Socrates has moderation.19 Here, Socrates, the clause subject, is licensed by the noun predicate moderation (and have is a 17 We have here an instance of the infinitive variously labelled as 'prospective' or 'dynamic' (as opposed to 'declarative') or 'virtual' (as opposed to 'factual'): it refers to a state of affairs which exists potentially (δυνάμει), and whose realization depends on a prior event described by the matrix verb (hence it cannot express an independent time reference). The presence of some "modal" nuance is confirmed by the fact that this infinitival clause consistently takes the negation μή. See Jannaris ( supplementary, auxiliary predicate).20 This syntactic condition implies restrictions on the combinatory possibilities of the noun itself, e.g. on its ability of taking a subjective genitive or a possessive adjective (which would conflict with the clause subject), as in Socrates has Triphon's moderation. As a matter of fact, an expression such as Socrates has Triphon's moderation is not ruled out. The relevant point is that this expression entails a peculiar interpretative effect: the moderation would be both Socrates' and Triphon's (thus producing readings such as 'Socrates has the same moderation that Triphon also has' or 'which is typical of Triphon' , or similar). This effect is the semantic correlate of a syntactic condition, namely, the cooccurrence of two distinct subjects of moderation, at two distinct syntactic levels. It is evident that this does not happen in cases where the clause arguments are licensed by a verbal predicate, as in Socrates praises / criticizes Triphon's moderation. In the framework adopted here (see n. 20), all nouns (not only verbal ones) potentially perform a predicative function and may, in the appropriate context, license an argument with the syntactic function subject, commonly associated with the semantic role Possessor, as in Achilles has a spear, or Achille's spear.21 Turning now to the type διδάσκω σε σωφροσύνην, we observe that here σωφροσύνη displays the kind of combinatory restriction mentioned above: it does not admit a subjective genitive or a possessive adjective. A hypothetical expression such as διδάσκω σε τὴν ἐμὴν σωφροσύνην 'I teach you my moderation' could possibly mean 'I teach you to have (after my training) the same moderation that I have' . As a matter of fact, cases like this do not occur in our corpus: the noun filling the T slot is never combined with a subjective genitive or with a possessive adjective. This restriction does not depend on the nature of the noun itself (cf. n. 18); it depends on the specific context, and therefore one should further investigate the function of T.
The real nature of T, in our construction, is not satisfactorily captured by labels such as 'inanimate' . Of course, it is not wrong to claim that T does not 20 More precisely, here moderation has both a predicative and an argumental function; as a predicate, it licenses a subject, which finally surfaces as the clause subject. On this type of constructions in Ancient Greek, see Benedetti & Bruno (2012). We follow here some basic assumptions on the relationship between nouns and predicativity developed by Rosen (1987); Mirto (1990); La Fauci & Loporcaro (1997);Pieroni (2015). In the framework of La Fauci (2000), argument and predicate are not absolute notions, but functional features, [± argument], [± predicate]; the corresponding positive values may relate to different elements of the clause or also combine in one and the same element. Nouns, in particular, can perform jointly predicative and argumental functions in the clause. 21 This clarification is necessary in view of some issues which shall be dealt with in Section 6.
Journal of Greek Linguistics 20 (2020) 153-178 refer to animate entities, but-on closer inspection-this is not the relevant feature. In the context under investigation, in fact, T does not denote an entity at all, but rather a predicative process. This fact is more immediately evident with some lexemes than with others; in any case, it can be made clearer through a verbal paraphrase. This holds even for a noun like φάρμακα 'healing remedies' , 'medicines' , which may have a concrete referent (in principle, a first-order entity in the sense of Lyons 1977, ii: 438ff.), as in (18), which describes a scene in which Paeëon spreads healing herbs on Ares' wound: But in the passage (5) above, in context with διδάσκειν, the noun φάρμακα does not denote a concrete entity: what Chiron teaches Achilles is how to deal with drugs (on Chiron instructing Achilles on medicine, see e.g. Craik 2010).22 This is, in our view, the relevant property of T in the construction with διδάσκειν: independently of its possible use in other contexts, the nominal filling the T-slot here describes a predicative process with an unexpressed subject; this unexpressed subject, in turn, is bound to R, the object of διδάσκειν. So διδάσκω σε σωφροσύνην / σοφίαν / ἀοιδήν / ἔργα / ἐγκράτειαν γαστρὸς / φάρμακα etc. can be paraphrased 'I teach you (your) moderation-to-be / (your) wisdom-to-be / (your) song-to-be / (your) works-to-be / (your) medical practices-to-be' (where the meaning component "to-be" is imposed by διδάσκω, as we have seen above).
The assumption that R and T are merely two distinct arguments licensed by the verb διδάσκειν, on the same syntactic level, does not easily account for their mutual relationship.
Its understanding requires a shift of the focus from categorial notions such as noun and verb to functional notions such as argument and predicate (where the latter is not exclusive of verbs, see n. 20). This allows the construction under investigation to decompose into two distinct predicative levels, centered, respectively, on a verbal predicate (διδάσκειν) and on a nominal one (T), with 22 This is not an unusual phenomenon. Any noun can, in the appropriate context, describe a process, as clearly stated by Pieroni (2015: 357ff.): "a nominal which may refer to an entity is not per se a nominal which denotes a first-order entity: the ontological reference to an entity is, obviously, nothing more than a possible circumstance". R as their shared argument. In other words, we have a verbal predicate licensing a subject (A), an object (R) plus a complement containing a noun predicate (T), which binds its (null) subject to R (we may recall that the condition of a necessarily null subject, coreferent with R, is a shared feature of both T and the infinitival complement).23 Phenomena of this kind can be described by reference to the notion of control into nominals, as developed in e.g. Jackendoff & Culicover (2003) and Alba Salas (2006). These studies have focused on cases in which a verb's subject controls the (null) subject of an event noun in complement position; in the simple formalism adopted by these authors, this is represented as in (19): The type we are dealing with here offers an instance of object control, which could be represented as in (20): Conclusions and further issues The identification of the predicative nature of T finds interesting echoes in some remarks emerging in traditional historical handbooks. Kühner & Gehrt's (1898) grammar notes that διδάσκω τὴν μουσικήν σε may be paraphrased as 'ich musiklehre dich' ('I teach you music'), thus intuitively capturing the predicative nature of both διδάσκω and μουσικήν and the status of σε as their shared argument.26 23 We may observe that also in Cambridge Grammar of Classical Greek the two accusatives are distinguished as "direct object" and "complement" respectively; see. van Emde Boas, et al. (2019: 363). 24 Example from Jackendoff & Culicover (2003: 553). Co-reference is notated by subscripts: the postposed subscript represents the controller and the preposed subscript represents the null controlled subject of the complement. 25 The object of διδάσκειν may occasionally be left unexpressed, thus receiving a generic reading. As observed above (n. 13), our hypothesis arose within the investigation of Ancient Greek data, but it can be fruitfully tested on a much wider scale, as comparable data (mutatis mutandis) may be observed in many other languages. 26 We are aware that the paraphrasis with musiklehren may evoke the idea of some sort of noun incorporation. This would be misleading as our proposal does not involve any sort of incorporation. It can nevertheless be added that, in the exemplum fictum proposed by Kühner & Gehrt (1898), διδάσκω τὴν μουσικήν could be emended into διδάσκω μουσικήν: as The presence of double predication, one overtly manifested in the verb form διδάσκειν and one, less patent, in nominal form (T), is more evident in cases in which the noun predicate has a verbal alternative, as in διδάσκω σε σωφροσύνην σωφρονεῖν (hence the comparison with the infinitival construction offered our starting point). But our analysis is independent of the lexical nature of the noun filling the T-slot: as we saw in the discussion on φάρμακα above, the context of διδάσκειν activates the potential predicative nature of any noun.
The focus on a single verb, διδάσκειν, has allowed to deeply investigate some crucial aspects which have not emerged in previous, more extensive, studies. Our hypothesis accounts for some empirical properties which are not easily explained under the traditional analysis, such as the scarce propensity of T to surface as subject (repeatedly highlighted in the literature, as we have seen above, Sections 1 and 3).27 Moreover, it accounts for restrictions-which were not noticed in the previous literature-within the internal syntax of T (concerning the combination with subjective genitives and possessive adjectives). These data are an inescapable starting point for any possible analysis of the double accusative construction of διδάσκειν. Hence, the hypothesis presented here has a validity for the specific case examined. Furthermore, it may provide refreshing insights into a wider domain, and invite to further pursue the investigation of the peculiar status of (a class of) Ancient Greek double accusatives.
Beyond the specific case of διδάσκειν, a major innovative contribution of our analysis is that it invites to focus not only on the relationship between the verbal predicate on the one hand and the two accusatives on the other, but also-or even primarily-on the relationship between the two accusatives. As we saw above, previous studies have stressed the asymmetry between them. Conversely, the present analysis, based on both empirical evidence and theoretical considerations, has gone a step beyond, reveling their mutual semantic and syntactic correlation (which is, in fact, a facet of this asymmetry). This has been accounted for within an approach which, based on the functional notions argument / predicate, highlights the predicative potentiality of nouns.
This idea offers a promising challenge for further research. It may offer a clue for understanding what lies behind phenomena that have repeatedly been observed in the literature on double accusatives. For example, in many cases a peculiar relationship exists between two accusatives, which is usually emerges also from the examples quoted in this paper, T usually occurs without a definite determiner (and this confirms its predicative nature). 27 On the general incompatibility between predicativity and subjecthood, see La Fauci & Loporcaro (1997). described in terms of possession.28 This is illustrated e.g. by verbs of 'depriving' , such as ἀφαιρεῖσθαι: this takes two accusatives, one denoting the person who is deprived and the other one denoting the possession which is taken away (cf. Jacquinod 1989: 215ff.).29 In a structure such as ἀφαιρέομαι σε ἵππους 'I deprive you acc horses acc = I deprive you of your horses' (here artificially created on the basis of Hom. Il. 8. 107) a possessive relationship (hence a predicative one; see Section 5) is implied between σε and ἵππους (remarkably, we observe, in this context, a non-occasional, but necessary possessive relationship).30 Furthermore, a double accusative construction which imposes a (non-occasional but necessary) possessive relationship between the two accusativesusually ascribed to the type of inalienable possession-is the famous σχῆμα καθ᾽ ὅλον καὶ μέρος "whole and part schema", a transitive construction in which, besides an accusative Patient licensed by the verbal predicate, a further accusative occurs denoting the part of the body on which the action is localized, as βάλλω σε ὦμον 'I hit you acc shoulder acc = I hit your shoulder; I hit you on the shoulder' .31 In the "whole and part schema", as usually observed in the literature,32 only one of the accusatives, namely that denoting the Patient, may alternate with the passive subject, thus determining a similar asymmetry to that observed in the case of διδάσκειν. The relationship between the two accusatives is not determined by the verbal predicate,33 but rather arises in the combination of the two nominals, as if a compressed possessive predication were implied in the construction.34 This kind of possessive relationship-a 28 We may recall that the hypothesis of a possessive relationship involved in double object constructions repeatedly shows up in recent theories; see, among others, Harley (2002) and Efstathopoulou (2007), who applies this notion to Modern Greek double object constructions. 29 As far as I can see from the rich collection of examples in Jacquinod (1989), the person deprived is always the possessor of the thing which is taken away. Interestingly, an alternative to the double accusative is offered, in some authors, by a construction in which the noun denoting the person appears in the genitive, thus confirming the possessive interpretation; see Jacquinod (1989: 222 and passim). 30 The inquiry can be extended to other verbs, though a unitarian description of the composite domain of double accusatives is not a priori to be expected. On the composite nature of this class, see Jacquinod (1995), Luraghi & Zanchi (2018 Though it has to be compatible with it, in the sense that the verbal predicate must describe an action which may affect the Patient in one of its parts. This property identifies a class of verbal predicates; see Mirto (1998). 34 Actually, as argued by Mirto & Rosen (1994) and Mirto (1998), we have here a specific "meronymic" one (cf. Mirto & Rosen 1994)35-involves, once again, a predicateargument relationship between the "meronym" (i.e. the name of the part) and the "holonym" (i.e. the name of the whole). What does the construction of διδάσκειν have in common with these seemingly heterogeneous types? For the sake of simplicity, in order to highlight the relevant issue, we can take up again and extend the kind of notation applied above (cf. (20) The notation in (21) represents the fact that σε 'you' , the object licensed by the verbal predicate (hence the candidate to promotion to clause subject) corefers with the (null) subject of the nominal predicate (variously interpreted as the performer of an action, as the possessor of an object, or as the holonym with respect to its meronym). This analysis is confirmed by empirical data. In fact, the same restriction observed above uniformly holds in all cases: in this context, σωφροσύνην, ἵππους etc. cannot have an independently expressed subject (e.g. in the form of a subjective genitive or a possessive adjective); in fact, their (null) subject is syntactically bound to σε, the clausal object. Despite non-negligible differences, the constructions in (21) appear to reflect a more general common compositional pattern: the relevant property is the presence of predicative relationship between the two accusatives, with different interpretative effects depending on the semantics of the nominal predicate and on the interaction with the semantics of the verbal predicate.
These findings may have considerable research fallouts. They point to a possible reduction of the gap between double accusatives related to some specific verbs (such as διδάσκειν) and double accusatives related to specific construction types (such as "whole and part schema"), thus opening new paths for a more general understanding of Ancient Greek morphosyntax.36 instance of predicative relationship, a meronymic one; in several languages it presents specific features within the general domain of inalienable possession. 35 We refer to this study also for an unitarian account embracing both the "whole and part schema" and the "accusative of respect" (which are merely diathetic variants of the same pattern). 36 A major issue concerns the kind of process assigning the accusative case to the noun predicate. In principle, it is not the same process assigning the accusative to the other noun, i.e. the object argument licensed by the verbal predicate (see σε in example 21, which would be classed as a "structural" case). All we can say for the time being is that when a predica-