Semantic and syntactic demarcations of Classical Greek object cases: An object(ive) study

In Classical Greek, many verbs take direct objects marked with genitive (GEN) or dative (DAT), rather than accusative (ACC) case. Traditional grammars (Smyth 1956, Boas et al. 2019) fail to offer principled descriptions or accounts of the distribution of ACC, GEN, DAT object case for transitive verbs. This paper analyzes a corpus involving case-assigning transitive verbs, and examines Luraghi’s 2010 Transitivity Hierarchy in this context. We find that, while her ranking of verbs’ transitivity is correct, the features used to determine the hierarchy are not. Our study demonstrates a highly significant correlation between a verb’s level of transitivity (as indicated by the case marking on its object) and the Proto-role Properties of Change of State and subject Volitionality (Dowty 1991).

typically given, namely that genitive-marked objects are partitive (Hansen & Quinn 1992, Boas et al. 2019, which is an analysis that also ignores the systematic assignment of dative-marked direct objects. Section 1 reviews prior work on this topic, primarily that of Luraghi (2003Luraghi ( , 2010, and discusses some of the shortcomings in that literature. Section 2 discusses the usefulness of Dowty's (1991) conception of Proto-Roles, particularly in terms of the goals of the present study. Section 3 then presents the data domain and methodology that figure in our study, and section 4 describes some preliminary (pre-statistical) observations that come to the fore. In section 5, we present the statistical results of our examination of the distribution of ACC, GEN, and DAT case-marking on objects -first under the lens of Luraghi's featural analysis of these, and then using a revised set of features that yields superior results. Finally, Section 6 provides a discussion of the above, along with some conclusions and directions for future research.

Previous treatments of Classical
Greek case. Outside of Classical Greek pedagogical grammars, there have not been many linguistic treatments of case assignment or argument structure in CG, particularly in reference to the atypical behavior of dative and genitive direct objects. A largely general description of the distribution of ACC, DAT, and GEN case is provided in Luraghi 2003, and a somewhat more specific description of their distribution as casemarking on direct objects, equating this with semantic "levels of transitivity" in Luraghi 2010. According to Luraghi (2010), ACC objects (whose semantic role is characterized as "theme") are "wholly affected patients, which undergo a change of state" (p. 64) and which canonically occur with the most highly transitive verbs. It is claimed that ACC case is nearly always assigned in such highly transitive contexts, and that GEN case is exceptionally assigned to objects having a partitive interpretation. In Luraghi's terms, GEN objects that are complements of highly transitive verbs are "partitive genitive" and seen to be less affected (although still "somewhat" affected) than are ACC objects. This analysis mirrors the explanation put forth by most pedagogical grammars, claiming that GEN objects are partitives and are relics of GEN indirect objecthood (Hansen & Quinn 1992, Boas et al. 2019. Verbs that are subcategorized for DAT objects are, according to Luraghi, even less transitive than those that are subcategorized for ACC or GEN objects. The proposed Transitivity Hierarchy, intended to account for the mapping of verb meaning to case, distinguishes three levels of transitivity: High transitivity (HT), medium transitivity (MT), and low transitivity (LT). These levels are distinguished on basis of (i) whether the verb's object undergoes a Change of State (COS), and (ii) whether the object is affected. While Luraghi provides a worthy point of entry into the topic, the analysis is weakened by certain methodological issues which the present research aims to correct. First, Luraghi (2003Luraghi ( /2010 presents an analysis of Ancient Greek (AG), a corpus spanning some 1200 years (encompassing the period from 900 BCE to 500 CE). Accordingly, the analysis utilizes data from all of Homeric Greek (~800 BCE), Classical Greek (~500-300 BCE), and Koine Greek (~300 BCE-400 CE), thereby trying to capture generalizations about varieties of Greek that are separated by many centuries, if not a millennium. Utilizing a range of data that is diachronically distant as is Old English and Modern English runs the risk of conflating meaning-case correlations that may not be consistent across time. By restricting our study to consider only Classical Greek texts, we carefully examine a corpus spanning only 2 centuries (~500-300 BCE), which is likely to be systematically more consistent.
Secondly, the data brought to bear in this analysis are limited to some 20 examples of casemarking verbs, suggesting (if not representative of a much larger data collection) that the observations and analysis are rather tentative and preliminary. In our study here, we present a description of case marking that considers nearly 200 sentences, representing examples of some 25 verbs.
Finally, where Luraghi 2010 only considers two semantic properties of direct objects in framing the Transitivity Hierarchy (COS and Affectedness), the present study expands the range of semantic properties considered analysis by assessing the relevance of all 10 Subject and Object Proto-role properties proposed in Dowty 1991 to the matter of transitivity.
3. Argument structure and thematic roles. Thematic roles (θ-roles) have been treated as useful features by which to describe and organize the relationships between the semantic and syntactic properties of a verb's arguments (i.e. its subject and objects). Traditionally, these include roles such as Agent, Patient, Goal, Source, Experiencer, and Theme, among others. θroles are assumed to be assigned to argument positions and provide semantic interpretation of the relation between a verb and its arguments, as well as to determine the argument structure projected and cases assigned by any given verb. The use of a fixed inventory of θ-roles as an explanatory device has come under some criticism in the past (Jackendoff 1987, Dowty 1991.
Dowty 1991, in particular, takes issue with their existence as unanalyzable units of analysis, proposing instead a group of semantic properties that serve as 'tendency-makers' which map a verb's arguments into subject or object grammatical slots based on the collective weight of those properties, as exhibited by any given argument.
The semantic properties that Dowty proposes are divided into two major classes: i) Proto-agent properties and (ii) Proto-patient properties, wherein arguments having Protoagent properties tend to be mapped into subject positions and arguments having Proto-patient properties tend to be mapped into object positions. The individual Proto-roles are grouped as follows under the two headings shown below in The actual mapping of arguments to grammatical functions, utilizing these Proto-roles, is actually handled by an Argument Selection Principle (Dowty 1991: 576), by the application of which "the argument for which the predicate entails the greatest number of Proto-Agent properties will be lexicalized as the subject of the predicate; the argument having the greatest number of Proto-Patient entailments will be lexicalized as the direct object". We would contend that Dowty's more nuanced approach to the mapping of arguments to subject and object functions is useful in shedding light on the case marking observed on the objects of CG verbs, which have so far resisted traditional analysis.

Data and methodology.
While previous considerations of CG case have attempted to provide answers that generalize across all of Ancient Greek, the present study concerns itself only with Classical Greek (c. 500-300 BC). By virtue of this constraint, we are able to examine a data domain that is internally consistent and avoiding possibly confounding semantic changes that might arise over a longer time span. Our study also considers only works of prose, excluding poetic and theatrical works, so as to avoid the inherent anomalies that arise in works stylistically constrained by meter (which often involve, for this reason, violations of canonical grammatical form). For example, in poetry (but not prose) it is not uncommon for an author to choose a nontraditional DO case if that case, for being syllabically longer/shorter or containing a long/short vowel, better fit the metric flow of the construction. Prose (e.g., historical accounts and rhetorical works) is not so grammatically fluid and therefore offers more reliable indicators of DO casemarking parameters of CG. In order to avoid potential confounds inherent in semantically and grammatically anomalous verb phrases, the data examined was limited to sentences containing monotransitive predicates in active constructions with a single direct object complement. Accordingly, psychological predicates were disregarded, as the subject-object relationship of verbal complements in these constructions tends to be more fluid, as in example (1). Filip (1996: 3) asserts, apropos of this concern, "linking in the domain of psychological predicates is problematic if we assume that there is a direct and uniform association between thematic (or lexical semantic) arguments and syntactic arguments".
Propositional attitude verbs were also disregarded, as such predicates take (typically clausal) propositions rather than direct object NPs as complements, and object case is not clearly assigned by the propositional attitude verb to its complement, as in example (2). (2) Tom believes [Sam fears snakes].
4.1. DATA COLLECTION. Data was collected using Tufts University's Perseus Online Digital Library, a corpus of 13,507,448 Ancient Greek words in 2,908 Ancient Greek works. Each verb identified as selecting an ACC, GEN, or DAT-marked direct object was searched for in all of its possible forms (i.e. in all six principal parts and inflections). Instances of sentences containing target verbs that fit our data constraints were located in their original CG and then translated into English. Thirty-one verbs were initially identified, six of which were excluded for a lack of tokens, so that 195 instances across twenty-five verbs were collected.

DATA ANALYSIS.
Under the present analysis, data was analyzed both descriptively and statistically. Collected tokens were coded according to all of Dowty's Proto-Role Properties, as well as according to Luraghi's Transitivity Hierarchy, and again for the Revised Hierarchy that we will posit in section 6.2. The present study investigates which factors affect the assignment of the three different Object Case-markers, ACC, GEN, and DAT. Accordingly, the present data was coded such that the dependent variable, Case-marking, was coded as either ACC, GEN, or DAT. While the class of verbs which alternate between ACC/GEN have been previously mentioned, it is the task of the present study to investigate the first three, more traditional, verb-classes.
The data was examined utilizing Dowty's (1991) inventory of Proto-Role Properties, in order to determine whether Luraghi's (2010) transitivity hierarchy would be attested by the data or need to be revised. By invoking Dowty's inventory of subject and object semantic properties, we aimed to test the predictive capacity of these with respect to correlations between the semantic nature of verbs and their assignment of case to objects.
Each verb-token was coded as +/-for Dowty's Agent Proto-Role features as indicated by their subjects: undergoing a change of state, being volitional, movement, independent existence, and sentience to the event. Each verb-token was also coded as +/-for each of Dowty's Patient Proto-Role features as indicated by their direct objects: undergoing a change of state, incremental theme, being causally affected, movement, and dependent existence. Utilizing the two direct object semantic features (COS and Affectedness) which determine a verb's level of transitivity according to Luraghi's analysis, the transitivity of each monotransitive construction was calculated, such that (+COS, +Affected) instances were rated high transitivity, (−COS, +Affected) instances were rated medium transitivity, and (−COS, −Affected) rated low. The results of this analysis are provided in section 5.1 below. Then, after this initial analysis, the entire corpus of constructions was examined to independently assess the distribution of each according to the full range of object and subject semantic proto-role properties. From this examination, it was found that the subject/object feature COS and the subject feature Volitionality were both significant in regard to the assignment of object case. The transitivity of each monotransitive construction was then recalculated, such that (+COS, +Volitional) instances were rated high transitivity, (−COS, + Volitional) instances were rated medium transitivity, and (−COS, −Volitional) rated low, as described below in section 6.2.

Preliminary observations.
Here our preliminary observations note descriptive trends in the data which we noted before running any statistical analysis. In our data, ACC case-marking verbs typically denote events that are telic and which affect the object. Such verbs focus on events wherein there is a clear distinction between the before and after (highly telic, and a clear affect). The subjects and object of these verbs consistently align with Dowty's Proto-roles, such that their subjects consistently have Proto-Agent Properties, and their objects consistently have Proto-Patient Properties. This observation is consistent with descriptions given in Luraghi 2003 and2010, and are consistent with our typical understanding of "theme" and the canonical direct object semantic role, as in example (3) illustrating the ACC case-marking verb, apekteinan 'kill'.
(3) tous ortug-as kai tous alektruon-as … apektein-an the quails-ACC.M.PL and the roosters-ACC.M.PL kill.AOR.IND.ACT-3PL 'They killed the quails and the roosters.' (Aeschines, Against Timarchus 1.59) GEN 2 case-marking verbs typically denote states or activities (i.e., are atelic). Their effect on the object is not always clear or important to the action described. Often the effect on the subject is more important and focused than the effect on the GEN object. The arguments of GEN case-marking verbs are much less consistent than ACC case-marking verbs with respect to their having canonical subject and object Proto-role Properties. An example of a GEN case-marking verb, amelein 'neglect', is shown in (4). DAT case-marking verbs typically denote events that are directed towards the object, although the effect on the object is often not clearly realized. The arguments of DAT casemarking verbs are, like GEN case-marking verbs, less consistent than ACC case-marking verbs with respect to their having canonical subject and object Proto-role Properties. An example of a DAT case-marking verb, apeilēsas 'threaten', is shown in (5). 6.1. EXAMINING OBJECT PROPERTIES. Counts of the data indicate that Dowty's Patient Proto-roles of Change of State (COS) and Affectedness correlate with differences in the case-marking of objects. In contrast, the Patient Proto-roles of Independent Existence, Movement, and Incremental Theme played no distinguishable role in classifying verbs that variously mark their objects with ACC, GEN, or DAT case.
Considering the Patient Proto-roles of objects alone, it would appear that COS and Affectedness correlate with ACC/GEN/DAT case marking choices, and indeed Luraghi's analysis, focusing as it does on these properties of the object to the exclusion of subject properties, builds its transitivity hierarchy utilizing these two features. The rows in Table 3 are arranged according to Luraghi's Hierarchy, with ACC being highest transitivity, GEN being next, and DAT being last. The columns in Table 3 are also arranged according to Luraghi's Hierarchy, (+COS, +Affected) Table 3. Results utilizing Luraghi's Transitivity Hierarchy (COS and Affectedness) As Table 4 below shows, objects of ACC case-assigning verbs uniformly undergo COS, whereas objects of GEN and DAT case-assigning verbs do not. Objects of ACC and DAT case-assigning verbs are uniformly causally affected, but objects of GEN case-assigning verbs are not consistently so.
COS (undergo change of state) Causally Affected ACC-marking verb YES YES GEN-marking verb NO SOMETIMES DAT-marking verb NO YES Table 4. Corellation of Patient Proto-roles and case Reexamining Luraghi's transitivity hierarchy, we find that her predictions are not borne out by the observed data, in that objects of GEN case-assigning verbs are not consistently Affected (contra Luraghi). Furthermore, as we will demonstrate later, Affectedness turns out not to be a usefully clear or consistent property to account for the mapping from meaning to case.
Our descriptive assessment of this data is confirmed through statistical analysis. A Chisquare analysis reveals a highly significant interaction between the features of COS and Affectedness and the assignment of case ([X2(4, 141)=121.82, p<.0001]). However, while these two properties do indeed form the basis of Luraghi's Transitivity Hierarchy, the Chi-Square residuals (Table 5) contradict Luraghi's claim that GEN objects are more transitive than DAT objects. Thus, while the interaction is highly significant, the results show DAT case-assigning verbs to be more transitive than GEN case-assigning verbs, contrary to the ranking proposed in Luraghi 2010. Thus, either her ranking is wrong, or the properties used to determine that ranking are so. The balloon plot of the residuals (Figure 1)  Here below in Table 6, once again, the rows are arranged according to Luraghi's Hierarchy, with ACC being highest transitivity, GEN being next, and DAT being last. However, the columns are arranged according to the two Agent/Patient and Agent properties (COS and Volitional) for which we found a strong correlation. Here, (+COS, +Volitional) is most transitive, (−COS, +Volitional) is next, and (−COS, −Volitional) is last. In Table 6, we see that the largest number of ACC-case verb constructions are in the first column (n=29), the largest number of GEN-case verb constructions are in the second column (n=47), and the number of DAT-case verb construction is split across columns two (n=29) and three (n=24).   Our observations reported above are confirmed through statistical analysis. A Chi-square analysis reveals a highly significant interaction between the features of COS and Volitionality and the assignment of case ([X2(4, 141)=123.7.82, p<.0001]). In Table 8, we can see from the Chi-Square residuals that these two properties do confirm the distribution of case-marking in accordance with Luraghi's original hierarchy. This interaction shows ACC case-assigning verbs to be the most transitive, with GEN case-assigning verbs being next most transtive, and DAT case-assigning verbs being least transitive, just as proposed in Luraghi 2010. Our results confirm that Luraghi's Transitivity Hierarchy is essentially correct (with ACC > GEN > DAT), but that the relevant properties are in fact Volitionality of Subjects and COS, rather than Affectedness of Objects and COS. The balloon plot of the residuals (Figure 2)   We therefore conclude that the latter solution is the correct. There are two additional reasons for favoring the solution provided here. First, there is a class of verbs which can case-mark their objects with either ACC or GEN. If such verbs are intermediate between High Transitivity (HT) and Medium Transitivity (MT), as we have shown, then this optionality makes sense. However, if ACC verbs are HT and GEN verbs are Low Transitivity (LT), as would be the case if their levels of transitivity are determined by object Affectedness, then one would not expect this category to arise, since the ACC/GEN alternation would occur across non-adjacent categories. Second, we believe that Volitionality is a property which is more ontologically-salient than Affectedness, in that the volitional actions of a sentient subject are more readily observable than whether or not an object is affected. Compare examples (4) and (5), wherein the GEN object of amelein 'neglect' would be [+affected] and the DAT object of apeilēsas 'threaten' would not be so, according to Luraghi. It is not clear to us, nor to anyone with whom we've consulted, that the object of 'neglect' is necessarily affected (or more affected than the object of 'threaten'). For this reason, we conclude that the indeterminacy of Affectedness, relative to Volitionality, makes it a poor candidate for predicting transitivity.

High
This analysis, therefore, confirms Luraghi's ranking of ACC>GEN>DAT case-marking verbs within her Transitivity Hierarchy, but demonstrates COS and Volitionality (rather than COS and Affectedness) to be the determining factors. ACC case-assigning verbs are [+COS] and others are not, and among the two others, GEN case-assigning verbs are always [+volitional] and DAT case-assigning verbs are not (always).
In Classicists' discussion of Greek case, such an analysis fundamentally changes the conversation around how GEN case-assigning verbs are described. In nearly every authoritative grammar of Classical Greek, including the most recent (Boas et al. 2019), GEN case on direct objects is described as "partitive", and while there certainly are instances of partitive GEN case in the data, not all GEN objects are so. Where ACC and GEN case-assigning verbs are distinguished by the parameter of [±COS], we submit that both the case assigned by both categories of verbs is "pure" and "lexically-assigned", reflecting the core meaning of the verb.