Do Good or Suffer Evil: Syntactic Nominalisations in Verbal Multi-Word Expressions in the PARSEME GRC Corpus

1 Introduction

Structures such as kick the bucket, spill the beans, take a picture, and pay attention are verbal multi-word expressions, lexically speaking. Multiple lexemes form the verb phrase (Nagy et al. 2013; Savary et al. 2018). However, the three examples provided differ as regards their structural analysability, i.e. to what extent each component can be modified separately from the rest of the phrase, and semantic compositionality, i.e. to what extent the meaning of the parts can be deduced from the meaning of the whole (Ledgeway and Vincent 2022, 51).

Kick the bucket is non-analytic and non-compositional (Evert 2009). Spill the beans is non-analytic and non-compositional but the meaning of the expression is transparent, i.e. the meaning of the expression can be recovered from its parts (Sheinfux et al. 2019, 42). Take a picture is analytic and compositional if reconceptualisation of the noun is accepted as compositional, i.e. picture does not refer to the concrete object but to the process resulting in the object (Radimský 2011). Pay attention is analytic and compositional if abstraction of the verb is accepted as compositional, i.e. pay has the meaning of direct rather than hand over (Heine 2020).

Verbal multi-word expressions are ubiquitous in all text types of Greek, both classical (5th/4th c. BC) and postclassical (ca. 3rd c. BC to 7th c. AD). However, unlike in many spoken languages (and Latin!)^[1] in which dictionaries and grammar books cover verbal multi-word expressions, for (post)classical Greek they are often thought of as idioms, as collocations, or as prose phrases (see the standard dictionary Liddell-Scott-Jones). None of these conceptualisations captures all the relevant examples. Their misunderstanding can have a far reaching impact on the interpretation of the narratological context, the creation process of the work, or the Sitz im Leben of the text (Fendel 2024b, 331–336 for three worked examples across periods of time). Three desiderata for (post)classical Greek are suitably large corpora annotated for verbal multi-word expression, their representation as such in dictionaries, and their coverage in grammar books (see Fendel 2024b, 328–331 in detail). PARSEME offers a framework to achieve this.

In frameworks where compositionality is conceptualised as categorical and transparency is dismissed, only take a picture and pay attention would qualify as compositional (Mel’čuk 2023, 29 and 32). In frameworks in which additionally constraints are placed on the noun, i.e. it must be abstract and predicative, only pay attention would qualify. The PARSEME 1.3 guidelines used to annotate the PARSEME Ancient Greek corpus are such a framework.^[2]

PARSEME is a Natural-Language-Processing initiative, which is currently part of the UniDive Working group 1 Corpus annotation.^[3] Its core aim is the consistent annotation of large corpora for verbal multi-word expressions in order to train language models. Subsequent corpus releases have increased the coverage for singular languages and the number of languages and varieties represented. The latest release (PARSEME 1.3) is accessible here: https://lindat.mff.cuni.cz/repository/xmlui/handle/11372/LRT-5124. The corpus can be searched by means of GrewMATCH (https://parseme.grew.fr/?corpus=PARSEME-PL@1.3).

The universal annotation guidelines are accessible here (GRC for Ancient Greek): https://parsemefr.lis-lab.fr/parseme-st-guidelines/1.3/index.php?page=home. These are synchronically focussed, fundamentally categorical, i.e. a structure is one or the other type of verbal multi-word expression, and lexically focussed, i.e. morphosyntactic, semantic, or pragmatic aspects are not annotated.

The PARSEME 1.3 guidelines exclude two types of structures that are closely related to those annotated and may synchronically and diachronically even qualify in certain cases, i.e. cognate-object constructions, e.g. to run a race (cf. Roesch 2018 on Latin), and constructions with a syntactic nominalisation rather than a lexical noun, e.g. to do good (Meinschaefer 2016; Marini 2018). They are annotated as NotMWE (not a multi-word expression) in the PARSEME Ancient Greek corpus, i.e. they were collected but fall outside of the scope of the guidelines. Yet what is their relationship with structures that fall into the annotation scope, primarily Light-Verb Constructions (LVCs) and Verbal Idioms (VIDs)?

Light-verb constructions (LVCs) are combinations of a verb and a noun in which the noun is abstract and predicative and the verb contributes either tense, aspect, and mood marking only (LVC.full) or causative semantics additionally (LVC.cause), e.g. διάβασιν ποιέομαι diabasin poieomai ‘to make the crossing/to cross’ versus διάβασιν παρέχω diabasin parekhō ‘to grant the crossing/to let (somebody) cross’. Light-verb constructions are fully analysable, formally, and compositional, functionally.

Verbal Idioms (VIDs) are lexically, morphologically, morphosyntactically, and syntactically inflexible. This inflexibility goes hand in hand with semantic non-compositionality. This includes structures that are transparent, i.e. the meaning of the whole can be recovered from the meaning of its parts e.g. when considering reconceptualisation (Radimský 2011), metaphorical extension (Sheinfux et al. 2019), or metonymy (Kraska-Szlenk 2014) for the noun. Thus, to beat about the bush (non-transparent, non-compositional), to spill the beans (transparent, non-compositional), to take heart (transparent, compositional with metaphorical extension), and to take a picture (transparent, compositional with reconceptualisation) all fall into this category. A Greek example is δεξιὰν δίδωμι dexian didōmi ‘to give a pledge/to pledge’. From the PARSEME perspective, the metaphorically extended or reconceptualised noun would qualify as a cranberry item, i.e. an item that is limited to a specific expression or context. The form-function pairing for picture = event of taking a picture and heart = courage is unique to the VID context.^[4]

The article focusses on constructions with a syntactic nominalisation in the noun slot, e.g. πιστὰ δίδωμι pista didōmi ‘to give the trusted (i.e. pledges)’, yet provides some parallels to cognate-object constructions, e.g. ἁμάρτημα ἁρμαρτάνω hamartēma hamartanō ‘to make a mistake’, answering three research questions:

1. What is the difference between abstract, predicative, and eventive and what diagnostics can we use when annotating?

2. How do structures with a syntactic nominalisation differ from yet border onto verbal multi-word expressions (as per the PARSEME 1.3 guidelines), especially Light-Verb Constructions and Verbal Idioms?

3. Why did we integrate structures with a syntactic nominalisation (and cognate-object constructions) into the annotation scope for the PARSEME GRC corpus?

The data come primarily from the PARSEME Ancient Greek corpus (release 2024). Table 1 provides an overview.

Table 1 shows that the structures of interest account for ca. 10 % of all verbal multi-word expressions annotated in each subsample. Additionally comparative samples are drawn using the Thesaurus Linguae Graecae and the ECF Leverhulme Sketch Engine corpus of literary classical Attic historiography, oratory, and philosophical prose (Fendel and Ireland 2023). Datasets are available here: DOI 10.5287/ora-4jvx7p56m.

The texts selected for analysis were selected for three reasons. Firstly, the selection was meant to align with the text type selection of other corpora of the PARSEME family, especially the modern Greek one, in order to allow for comparison. Lysias’ speeches are courtroom speeches, which due to the Athenian democracy being a direct democracy are not specifically formalised or technical in their language because they had to be comprehensible to the average person (Willi 2003, chap. 3; Fendel 2024a). Xenophon’s report is the report of a Greek mercenary army returning from a campaign in Persia to Greece. The journey is complicated by the fact that the person who had hired them (Cyrus) was killed in battle. Plato’s dialogue is an educated exchange between Socrates and his interlocutors about the issue of a constitution and political organisation. Its dialogic character makes it possible to observe the language of turn-taking and backchannelling, i.e. the dialogic asking for clarification or confirming comprehension at points such as English got it? (van Emde Boas 2022).^[5] The data sample is not random partially due to the alignment with other PARSEME languages and partially because (post)classical Greek is a corpus language. For corpus languages, limited amounts of data are available and cannot be expanded (except by chance discoveries). The available data has undergone a preservation process that is not re-constructible in all cases and relies to a certain extent on chance (see also Hoffmann 2005, chap. 8). The last native speakers of such languages are the texts themselves (Fleischman 2000). While the present corpus selection thus provides a suitably large sample vis-à-vis available data with regard to classical Attic oratory, historiography, and philosophical prose (the traditional genre labels for the three texts selected), findings cannot be generalised to texts of different time periods, types, or from different places.

After this introductory section, Section 2 addresses research question one, Section 3 research question two, and Section 4 research question three. Section 5 summarises the results, draws conclusions, and provides an outlook for future work.

2 Abstract, Predicative, and Eventive

The PARSEME 1.3 decision tree for Light-Verb Constructions is headed by the questions: Is the noun abstract? Is the noun predicative? Abstractness is a lexical concept, predicativeness a syntactic one. The below reviews each concept but proposes eventiveness as the way forward for syntactic nominalisations in the predicative-noun slot (and cognate objects).

Whether or not a noun has an abstract meaning is often determined negatively, i.e. by ruling out that it has a concrete meaning. A concrete meaning is a meaning that has a tangible reference point in the real world, e.g. a reference to people (father),^[6] animals (horse), places (market square), physical objects (chair), buildings (palace), and the like. This still leaves many meanings in the grey area.

Furthermore, many nouns in classical Greek do not move from nomen actionis (action noun) to nomen agentis (agent noun), along the path shown in (1), irreversibly but are polysemous.

An example of such an item with concrete and abstract meanings is πρᾶγμα pragma. πρᾶγμα pragma additionally seems to have a meaning nuance that is limited to verbal multi-word-expression contexts, i.e. a cranberry form-function pairing. (2) provides an overview:

πρᾶγμα pragma is derived by means of the suffix -μα -ma from the verbal root of πράσσω prassō ‘to do’. The suffix -μα -ma ‘forms neuter effect/result nouns’ (van Emde Boas et al. 2019, 265). However, as shown, the concrete effect/result meaning ‘thing’ is not the only meaning of πρᾶγμα pragma.^[7] Only the meaning nuances ‘occurrence, matter, affair; (plural) circumstances, affairs’ and ‘trouble, annoyance’ pass the abstractness test. The cranberry form-function pairing is πράγματα pragmata (plural, often indefinite) ‘trouble, annoyrance’ in verbal multi-word expressions.

Whether or not a noun is predicative is determined in the PARSEME 1.3 guidelines by testing whether the noun has any arguments. These might be syntactic arguments, e.g. a subject, but also functional arguments that are syntactically attributes or adjuncts, e.g. an objective genitive (Croft 1991; Markantonatou and Samaridi 2017; Dalrymple et al. 2019). Thus, e.g. χάρις kharis ‘gratitude’ and ἐξέτασις exetasis ‘inspection’ are predicative as they can take a subject (in the form of a subjective genitive) and in the case of ἐξέτασις exetasis ‘inspection’ also an objective genitive, yet e.g. χειμών kheimōn ‘wintry, stormy weather’ would not qualify nor would κακά kaka ‘evil’ (a syntactic nominalisation) or ὁδός hodos ‘path’. PARSEME’s use of the term ‘predicative’ seems to conflate predicative in the syntactic sense and the semantic concept of eventiveness.

Predicative in the syntactic sense refers to elements that complement another element such that the elements form a complex predicate together. Complex predicates are monoclausal, i.e. they project into a single clause (Butt 2010). Classical Greek requires such predicative elements, i.e. nouns, adjectives, or participles, with copular verbs (εἶναι einai, γίγνομαι gignomai), as shown in (3):

The predicative element agrees in case, number, and gender with the subject of the copular verb thus showing that the copular verb and the predicative element form a complex predicate. In possessive structures consisting of a copular verb + noun in the genitive/dative ‘to have (something)’ (Benvenuto and Pompeo 2015), the possessive genitive or dative functions as the predicative element.

The nominal element in a Light-Verb Construction completes the light verb in order to form a complex predicate (see Markantonatou and Samaridi 2017; Butt 2010). It can appear in the subject or object slot of the verb or the complement slot of a preposition, e.g. Xenophon, Anabasis, 2.2.19 τοῖς Ἕλλησι φόβος ἐμπίπτει tois hellenēsi phobos empiptei ‘fear befell the Greeks’, Xenophon, Anabasis, 6.5.29 φόβον παρεῖχε phobon pareikhe ‘it caused fear’, and Plato, Leges, 635c1 ὁπόταν εἰς ἀναγκαίους ἔλθῃ πόνους καὶ φόβους καὶ λύπας hopotan eis anagkaious elthē ponous kai phobous kai lupas ‘whenever he got into the necessary labour, fear, and grief’ (Mel’čuk 1996, 59–64; Tronci 2017 on Greek).

The nominal element is the main event reference as anaphora shows. Anaphora of a Light-Verb Construction can either be by resuming the noun (Xenophon, Anabasis, 1.7.1–2 ἐξέτασιν ποιεῖται exetasin poieitai ‘he is making an inspection’ resumed by ἐξέτασιν exetasin ‘inspection’) or by resuming the Light-Verb Construction as a whole (Xenophon, Anabasis, 4.6.20 σύνθημα ἐποιήσαντο sunthēma epoiēsanto ‘to make an agreement’ resumed by συνθέμενοι sunthemenoi ‘to agree’) but never by resuming the light verb only.^[9] This shows that the predicative noun is the semantic pivot. Thus, the nominal component in a Light-Verb Construction needs to be eventive.

Eventive is a semantic concept. Grimshaw’s seminal categories of eventive nouns are shown in (4).

Complex event nouns describe an ongoing process, simple event nouns a completed event, and result nouns the outcome of an event without reference to the event.

Complex event nouns have an internal subject and object and simple event nouns an internal subject, like χάρις kharis ‘gratitude’ and ἐξέτασις exetasis ‘inspection’. When these appear in the noun slot of Light-Verb Constructions, they share the light verb’s subject component, which can take on any functional role, e.g. Volitional Undergoer or Agent (cf. Næss 2007). Thus, we test while annotating whether the semantic subject of the noun and the grammatical subject of the verb are co-referential.^[10]

All of Grimshaw’s event nouns are lexical nominalisations (Meinschaefer 2016, 395) and deverbal. However, recent work by Bel et al. (2010, 46–47) suggests that there are non-deverbal event nouns in Spanish, e.g. accidente ‘accident’ and guerra ‘war’, which differ from Grimshaw’s categories in that they do not have an internal argument. κακά ‘evil’ may resemble these. Huyghe et al. (2017, 118) suggest that there are underived event nouns in French, e.g. crime ‘crime’, grève ‘strike’, repas ‘meal’, and orage ‘storm’, the eventive meaning of which ‘can be derived from a non-eventive meaning’ (similarly Radimský 2011).^[11] χειμών ‘wintry, stormy weather’ and ὁδός ‘path’ may resemble these.

If event nouns can be non-deverbal and underived, this changes things for syntactic nominalisations (and cognate-object constructions) in the predicative-noun slot. In these, we often find non-deverbal, e.g. syntactic nominalisations of adjectives, and/or underived, e.g. nouns like meal and race, nominals. Both types of structures align with other verbal multi-word expressions in that the nominal component is (reconceptualised as) eventive (within the specific context of the structure).

According to the PARSEME 1.3 guidelines, structures with underived or non-deverbal event nouns that are lexically, morphologically, morphosyntactically, and syntactically inflexible would be classified as Verbal Idioms. Only those structures in which the nominal component passes the abstractness and predicativeness tests would qualify as Light-Verb Constructions.^[12]

Structures with a syntactic nominalisation in the predicative-noun slot, e.g. πιστὰ δίδωμι pista didōmi ‘to give pledges’, (and cognate-object constructions, e.g. ἁμάρτημα ἁρμαρτάνω hamartēma hamartanō ‘to make a mistake’), may pass the abstractness and predicativeness tests. However, syntactic nominalisations in the predicative noun slot when inflected as neuter accusatives (singular and plural) come with the issue of ambiguity in that neuter accusatives can serve an adverbial function. Thus, one could read them as modifiers of the verb, e.g. κακὰ ποιέω kaka poieō ‘to treat badly’ rather than ‘to do evil’.

Regarding research question one (What is the difference between abstract, predicative, and eventive and what are the diagnostics when annotating?), we have seen that abstract is a lexical, predicative a syntactic, and eventive a semantic concept. Abstractness is often tested negatively, i.e. by ruling out references to people, places, and the like. Predicativeness can be tested by observing co-referentiality of the internal subject argument of the noun and the syntactic subject of the verb. The above suggests including underived and non-deverbal event nominals. If these pass the abstractness and predicativeness tests, the structure that they are part of may be a Light-Verb Construction, unless lexical, morphological, morphosyntactic, and syntactic flexibility can be tested to assess whether the structure is a Verbal Idiom.

3 LVC, VID, or NotMWE

Classical Greek has a productive, in the sense of generality (Barðdal 2008, 21), set of derivational morphology in order to derive nouns from adjectives, as shown in (5) and exemplified in (6):

Additionally, nouns can be derived from adjectives by means of syntactic nominalisation, which is equally productive. Syntactic nominalisation can be applied to the positive, comparative, and superlative of an adjective unlike derivational morphology.

Syntactic nominalisation of an adjective can be morpho-syntactically marked, by the addition of a determiner phrase, (7), but does not have to be, (8).

Furthermore, vehicular contexts exist, e.g. the addition of an indefinite, (9), which can act as a pronoun or determiner, or the noun πρᾶγμα pragma ‘thing’, (10).^[13]

The optional morpho-syntactic marking of nominalisations introduces ambiguity. Furthermore, the nominalised phrase in the neuter accusative can be interpreted as an adverb, compounding this ambiguity, e.g. μέγα mega ‘greatly’ instead of morphologically derived μεγάλως megalōs ‘greatly’ (van Emde Boas et al. 2019, 366). Finally, syntactic nominalisations, especially when inflected in the neuter, can refer to physical items or to circumstances and behaviours, which creates further ambiguity. Therefore, we need ambiguity-reducing contexts.

The three subsections below discuss (i) parallels between lexical and syntactic nominalisations in verbal multi-word expressions, (ii) the stative ἐν en [noun in the dative] ἔχω ekhō ‘to have in X’ frame in which the nominalisation appears in a case other than the accusative, and (iii) instances in which the light verb par excellence, ποιέομαι poieomai ‘to do’ in the middle voice, appears with a syntactic nominalisation.

3.1 πιστά pista versus δεξιά dexia versus πίστις pistis with δίδωμι didōmi in Xenophon’s Anabasis (VID, transparent vs. VID, non-transparent vs. LVC)

In Xenophon’s Anabasis, the army leaders repeatedly broker deals with the people whose area they cross. Thus, pledges are given and received and oaths are made. The giving and receiving of pledges can be expressed by means of the Verbal Idiom δεξιὰν/δεξιὰς δίδωμι/λαμβάνω dexian/dexias didōmi/lambanō ‘to give/receive pledge(s)’ (see in detail Fendel 2025a).^[14] The phrase is lexically, morphosyntactically, and syntactically inflexible. There is limited morphological flexibility in that both the singular and plural forms of the noun appear in classical literary Attic historiography, oratory, and philosophical prose (ECF Leverhulme corpus).

The Verbal idiom co-occurs with the simplex verbs ἐπιορκέω epiorkeō and ὄμνυμι omnumi ‘to swear an oath’, which refer to the subsequent event, i.e. the oaths. The Verbal Idiom parallels in the discourse three times, (11)–(13), the phrase (τὰ) πιστὰ δίδωμι/λαμβάνω (ta) pista didōmi/lambanō ‘to give/receive pledges’. All three passages come from the PARSEME Ancient Greek corpus.^[15]

(11)

Xenophon, Anabasis, 3.2.4–5

ἀλλ᾽ ὁρᾶτε μέν, ὦ ἄνδρες, τὴν βασιλέως ἐπιορκίαν καὶ ἀσέβειαν, ὁρᾶτε δὲ τὴν Τισσαφέρνους ἀπιστίαν, ὅστις λέγων ὡς γείτων τε εἴη τῆς Ἑλλάδος καὶ περὶ πλείστου ἂν ποιήσαιτο σῶσαι ἡμᾶς, καὶ ἐπὶ τούτοις αὐτὸς ὀμόσας ἡμῖν, αὐτὸς δεξιὰς δούς, αὐτὸς ἐξαπατήσας συνέλαβε τοὺς στρατηγούς, καὶ οὐδὲ Δία ξένιον ᾐδέσθη, ἀλλὰ Κλεάρχῳ καὶ ὁμοτράπεζος γενόμενος αὐτοῖς τούτοις ἐξαπατήσας τοὺς ἄνδρας ἀπολώλεκεν. Ἀριαῖος δέ, ὃν ἡμεῖς ἠθέλομεν βασιλέα καθιστάναι, καὶ ἐδώκαμεν καὶ ἐλάβομεν πιστὰ μὴ προδώσειν ἀλλήλους, καὶ οὗτος οὔτε τοὺς θεοὺς δείσας οὔτε Κῦρον τεθνηκότα αἰδεσθείς, τιμώμενος μάλιστα ὑπὸ Κύρου ζῶντος (…)

all᾽ horate men, ō andres, tēn basileōs epiorkian kaì asebeian, horate de tēn Tissaphernous apistian, hostis legōn hōs geitōn te eiē tēs Hellados kai peri pleistou an poiēsaito sōsai hēmas, kai epi toutois autos omosas hēmin, autos dexias dous , autos exapatēsas sunelabe tous stratēgous, kai oude Dia xenion ēdesthē, alla Klearkhō kai homotrapezos genomenos autois toutois exapatēsas tous andras apolōleken. Ariaios de, hon hēmeis ēthelomen basilea kathistanai, kai edōkamen kai elabomen pista mē prodōsein allēlous, kai houtos oute tous theous deisas oute Kuron tethnēkota aidestheis, timōmenos malista hupo Kurou zōntos (…)

‘However, look at, men, the perjury and impiety of the King and look at the disloyalty of Tissaphernes, who said that he was a neighbour of Greece and was most keen to save us, and he himself swore on these things to us, and having given pledges, he deceived (us) and took the generals and was not even fear Zeus Xenios, but he sat at the table with Clearchus, and through this he deceived and killed the men. Ariaius, whom we wanted to make king, gave and received pledges not to betray each other, and this man feared neither the gods nor paid tribute to Cyrus, who was dead, although he was most honoured by Cyrus while he was alive (…).’

(12)

Xenophon, Anabasis, 1.6.6

τοῦτον γὰρ πρῶτον μὲν ὁ ἐμὸς πατὴρ ἔδωκεν ὑπήκοον εἶναι ἐμοί: ἐπεὶ δὲ ταχθείς, ὡς ἔφη αὐτός, ὑπὸ τοῦ ἐμοῦ ἀδελφοῦ οὗτος ἐπολέμησεν ἐμοὶ ἔχων τὴν ἐν Σάρδεσιν ἀκρόπολιν, καὶ ἐγὼ αὐτὸν προσπολεμῶν ἐποίησα ὥστε δόξαι τούτῳ τοῦ πρὸς ἐμὲ πολέμου παύσασθαι, καὶ δεξιὰν ἔλαβον καὶ ἔδωκα, μετὰ ταῦτα, ἔφη, Ὀρόντα, ἔστιν ὅ τι σε ἠδίκησα; ἀπεκρίνατο ὅτι οὔ. πάλιν δὲ ὁ Κῦρος ἠρώτα: οὐκοῦν ὕστερον, ὡς αὐτὸς σὺ ὁμολογεῖς, οὐδὲν ὑπ᾽ ἐμοῦ ἀδικούμενος ἀποστὰς εἰς Μυσοὺς κακῶς ἐποίεις τὴν ἐμὴν χώραν ὅ τι ἐδύνω; ἔφη Ὀρόντας. οὐκοῦν, ἔφη ὁ Κῦρος, ὁπότ᾽ αὖ ἔγνως τὴν σαυτοῦ δύναμιν, ἐλθὼν ἐπὶ τὸν τῆς Ἀρτέμιδος βωμὸν μεταμέλειν τέ σοι ἔφησθα καὶ πείσας ἐμὲ πιστὰ πάλιν ἔδωκα μοι καὶ ἔλαβες παρ᾽ ἐμοῦ;

touton gar prōton men ho emos patēr edōken hupēkoon einai emoi: epei de takhtheis, hōs ephē autos, hupo tou emou adelphou houtos epolemēsen emoi ekhōn tēn en Sardesin akropolin, kai egō auton prospolemōn epoiēsa hōste doxai toutō tou pros eme polemou pausasthai, kai dexian elabon kai edōka , meta tauta, ephē, Oronta, estin ho ti se ēdikēsa? apekrinato hoti ou. palin de ho Kuros ērōta: oukoun husteron, hōs autos su homologeis, ouden hup᾽ emou adikoumenos apostas eis Musous kakōs epoieis tēn emēn khōran ho ti edunō? ephē Orontas. oukoun, ephē ho Kuros, hopot᾽ au egnōs tēn sautou dunamin, elthōn epi ton tēs Artemidos bōmon metamelein te soi ephēstha kai peisas eme pista palin edōkas moi kai elabes par᾽ emou?

‘For my father first gave me this man to be my subordinate. When he had been ordered by my brother, as he said himself and was waging war against me occupying the citadel of Sardis, and I attacked him and achieved that it seemed best to him to stop the war against me, and I gave and received assurance. After this, Orontas, he said, is there anything that I have done wrong against you? He answered no. Cyrus asked again: Did you not later after suffering no wrong under me, as you agree yourself, desert (me) for the Mysians and treat my country as badly as you could? Orontas said yes. Did you now, said Cyrus, when you realised again your own power, go to Artemis’ altar and said that you felt repentance and convinced me and gave to me and received from me assurance again?’

(13)

Xenophon, Anabasis, 2.3.26–28

Καὶ νῦν ἔξεστιν ὑμῖν πιστὰ λαβεῖν παρ’ ἡμῶν ἦ μὴν φιλίαν παρέξειν ὑμῖν τὴν χώραν καὶ ἀδόλως ἀπάξειν εἰς τὴν Ἑλλάδα ἀγορὰν παρέχοντας· ὅπου δ’ ἂν μὴ ᾖ πρίασθαι, λαμβάνειν ὑμᾶς ἐκ τῆς χώρας ἐάσομεν τὰ ἐπιτήδεια. ὑμᾶς δὲ αὖ ἡμῖν δεήσει ὀμόσαι ἦ μὴν πορεύσεσθαι ὡς διὰ φιλίας ἀσινῶς σῖτα καὶ ποτὰ λαμβάνοντας ὁπόταν μὴ ἀγορὰν παρέχωμεν, ἐὰν δὲ παρέχωμεν ἀγοράν, ὠνουμένους ἔξειν τὰ ἐπιτήδεια. Ταῦτα ἔδοξε, καὶ ὤμοσαν καὶ δεξιὰς ἔδοσαν Τισσαφέρνης καὶ ὁ τῆς βασιλέως γυναικὸς ἀδελφὸς τοῖς τῶν Ἑλλήνων στρατηγοῖς καὶ λοχαγοῖς καὶ ἔλαβον παρὰ τῶν Ἑλλήνων.

Kai nun exestin humin pista labein par’ hēmōn ē mēn philian parekhein humin tēn khōran kai adolōs apakhein eis tēn hellada agoran parekhontas: hopou d’ an mē ē priasthai lambanein humas ek tēs khōras easomen ta epitēdeia. Humas de au hēmin deēsei omosai ē mēn poreusesthai hōs dia philias asinōs sita kai pota lambanontas hopotan mē agoran parekhōmen ean de parekhōmen agora ōnoumenous ekhein ta epitēdeia. Tauto edoxe, kai ōmosan kai dexias edosan Tissaphernēs kai ho tēs basileōs gunaikos adelphos tois tōn hellēnōn stratēgois kai lokhagois kai elabon para tōn hellēnōn.

‘And now it is possible for you to receive pledges from us to provide a friendly country to you and lead (you) to Greece unharmed while giving (you) a market. Wherever it is not possible to buy (provisions), we will let you take provisions from the country. Yet you in turn must swear to us to proceed like through a friendly (country) without harm while taking food and drink only when we do not provide a market, but if we provide a market, to get provisions by buying (them). This seemed good, and they swore (oaths) and gave pledges, Tissaphernes and the king’s wife’s brother to the generals of the Greeks and the captains, and they received (pledges) from the Greeks.’

The lexical noun δεξιά dexia ‘right hand, (in multi-word expressions:) pledge, assurance’ is replaced by the syntactic nominalisation (τὰ) πιστὰ (ta) pista ‘trustworthy, loyal things’. This makes the structure transparent.

(τὰ) πιστὰ (ta) pista ‘trustworthy, loyal things’ passes the abstractness and predicativeness tests (N-ABS and N-PRED) in the PARSEME Light-Verb-Construction decision tree shown in Figure 1 if we assume that it does not refer to ‘things’ but ‘circumstances, behaviour’. This is within the semantic scope of the nominalisation.

Figure 1:

Light-verb construction decision tree in PARSEME 1.3 (source: https://parsemefr.lis-lab.fr/parseme-st-guidelines/1.3/index.php?page=050_Cross-lingual_tests/020_Light-verb_constructions__LB_LVC_RB_).

However, contexts such as (14) show the remaining ambiguity. In (14), βαρβαρικὴν λόγχην barbarikēn logkhēn and Ἑλληνικήν (λόγχην) hellēnikēn (logkhēn) could be read as a concrete specification of the abstract event of assurance or as specifications of the concrete object ‘trustworthy/trust-inducing things’.

(14)

Xenophon, Anabasis, 4.8.6–7

οἱ δ’ ἀπεκρίναντο Ὅτι καὶ ὑμεῖς ἐπὶ τὴν ἡμετέραν χώραν ἔρχεσθε. λέγειν ἐκέλευον οἱ στρατηγοὶ ὅτι οὐ κακῶς γε ποιήσοντες, ἀλλὰ βασιλεῖ πολεμήσαντες ἀπερχόμεθα εἰς τὴν Ἑλλάδα, καὶ ἐπὶ θάλατταν βουλόμεθα ἀφικέσθαι. ἠρώτων ἐκεῖνοι εἰ δοῖεν ἂν τούτων τὰ πιστά. οἱ δ’ ἔφασαν καὶ δοῦναι καὶ λαβεῖν ἐθέλειν. Ἐντεῦθεν διδόασιν οἱ Μάκρωνες βαρβαρικὴν λόγχην τοῖς Ἕλλησιν, οἱ δὲ Ἕλληνες ἐκείνοις Ἑλληνικήν· ταῦτα γὰρ ἔφασαν πιστὰ εἶναι· θεοὺς δ’ ἐπεμαρτύραντο ἀμφότεροι.

hoi d’ apekrinanto Hoti kai humeis epi tēn hēmeteran khōran erkhesthe. legein ekeleuon hoi stratēgoi hoti ou kakōs ge poiēsontes, alla basilei polemēsantes aperkhometha eis tēn Hellada, kai epi thalattan boulometha aphikesthai. ērōtōn ekeinoi ei doien an toutōn ta pista . hoi d’ ephasan kai dounai kai labein ethelein. Enteuthen didoasin hoi Makrōnes barbarikēn logkhēn toĩs Hellēsin, hoi de Hellēnes ekeinois Hellēnikēn; tauta gar ephasan pista einai ; theous d’ epemarturanto amphoteroi.

‘They answered: Because you are coming against our country. The generals told (him) to say that: Not in order to do any harm but having been at war with the king we are returning to Greece, and we want to reach the sea. Those asked whether they would give pledges with regard to this. They said they wanted to give and receive (pledges). Then, the Macronians gave the Greeks a barbarian lance and the Greeks gave them a Greek one. They said that these are pledges. Both called the gods as witnesses.’

If the event reading applies, the syntactic nominalisation has an internal argument that is co-referential with the syntactic subject of the verb.^[16] Only active ‘to receive pledges’ passes V-LIGHT.^[17] Due to the ambiguity inherent in the syntactic nominalisation, V-REDUC is unlikely to be passed by the structure.^[18] This would mean that we would have to qualify (τὰ) πιστὰ δίδωμι (ta) pista didōmi as a Verbal Idiom. It is indeed lexically and morphologically (the plural of the adjective is compulsory) inflexible.

While a de-adjectival noun from the root πείθ- peith- (full grade)^[19] does not exist in classical times (apart from negative ἀπιστία apistia ‘disloyalty’), a deverbal event noun exists, i.e. πίστις pistis ‘assurance’, and appears in Light-Verb Constructions. Table 2 shows the support-verb-construction family around πίστις pistis ‘assurance’ in the ECF Leverhulme corpus (see Kamber 2008; Gross 1998).

Table 2:

πίστις pistis in the ECF Leverhulme Sketch Engine corpus.^a

Support verb	Singular (sg)/plural (pl)	Determiner phrase	Attributive phrase	Total
ἔχω ekhō ‘to have’	sg	–	ADJ (1)	2
ἔχω ekhō ‘to have’	pl	–	ADJ (2)	2
δίδωμι didōmi ‘to give’	sg	–	ADJ, REL	4
δίδωμι didōmi ‘to give’	pl	2	ADJ (3)	6
λαμβάνω lambanō ‘to receive’	sg	–	–	4
λαμβάνω lambanō ‘to receive’	pl	4	PP	5
ποιέομαι poieomai ‘to do’	pl	2	–	5
φέρω pherō ‘to bring’	pl	2	–	2
γίγνομαι gignomai ‘to become’	pl	2	–	2
παρέχω parekhō ‘to provide’	sg	–	–	2
VRel	sg	3	2	4
VRelb	pl	2	ADJ (1)	2
COC	pl	–	–	2

1. ^aVRel = verb of realisation, COC = cognate-object construction, ADJ = adjective, REL = relative clause, PP = prepositional phrase. ^bAristotle, Rhetoric, 3.1417b21 (ἀποδεικνύναι apodeiknunai ‘to show’) object continuity; Demosthenes, Speech, 18.164 (παραβαίνων parabainōn ‘breaching’) terminative.

Table 2 shows that the candidates with ‘to give’ and ‘to take/receive’ which parallel the Verbal Idiom and the structure with a syntactic nominalisation are indeed Light-Verb Constructions and allow for morphological and morphosyntactic variation.

It thus appears that a Verbal Idiom, a Light-Verb Construction with a deverbal event noun, and a construction with a syntactically nominalised adjective co-exist. They seem to differ semantico-pragmatically. The Verbal Idiom does not allow for resumption of the nominal part only in subsequent discourse nor does it allow for lexical, morphosyntactic, or syntactic variation which limits its usefulness in building discourse coherence (cf. Fendel and Ireland 2023), e.g. by means of pronominalisation and anaphora. The Light-Verb Construction with a deverbal event noun allows for resumption of the nominal component only as well as for morphological, morphosyntactic, and syntactic variation and thus for building discourse coherence (see also Wittenberg and Trotzke 2021).

The construction with a syntactic nominalisation is compositional and transparent. It seems to stress the stative nature of the event due to the nominalised adjective in the predicative-noun slot. Thus, the construction with the syntactic nominalisation is by no means filling a lexical gap but by drawing on a productive syntactic process allows for further nuancing of meaning while preserving discursive advantages.^[20] Under the PARSEME guidelines, it is a Verbal Idiom.^[21]

4 Discourse, Register, and Lexical Gaps

Section 3 has shown that syntactic nominalisations can appear in verbal multi-word expressions, both in Verbal Idioms and in Light-Verb Constructions in the PARSEME terms. Furthermore, structures with comparable lexical and syntactic nominalisations can co-exist yet are not equivalent from a variationist point of view. Finally, structures with syntactic nominalisation can fill gaps in the range of expression. The below revisits the structures discussed and categorised in Section 3 from the perspective of their fit into the system synchronically speaking.

When referring to the giving and receiving of pledges, Xenophon had three options to choose from: (i) the non-transparent Verbal Idiom δεξιὰν/δεξιὰς δίδωμι/λαμβάνω dexian/dexias didōmi/lambanō ‘to give/receive pledge(s)’, (ii) the transparent Verbal Idiom (τὰ) πιστὰ δίδωμι/λαμβάνω (ta) pista didōmi/lambanō ‘to give/receive pledge(s)’, and (iii) the Light-Verb Construction πίστιν/πίστεις δίδωμι/λαμβάνω pistin/pisteis didōmi/lambanō ‘to give/receive pledge(s)/assurance(s)’ (although note that PARSEME would qualify the recipient passive as NotMWE). The difference between the three is their discursive value. The non-transparent Verbal Idiom is derived from the act of shaking hands on a deal and handing over items as leverage. It can be resumed in subsequent discourse only as a whole. Conversely, the transparent Verbal Idiom with a syntactic nominalisation introduces ambiguity between physical pieces of leverage and the abstract behaviour surrounding the deal. It can be resumed in subsequent discourse by referring to the πιστά pista ‘loyal things/behaviours’ and exploiting its ambiguity (between physical item and abstract situation). Finally, the Light-Verb Construction allows for pronominalisation (e.g. by means of a relative pronoun) and lexical resumption of the noun. The noun is fully accessible and refers by itself to the event of giving pledges. It can be modified by means of attributive and determiner phrases in order to add nuancing to the picture (e.g. ἱκανάς hikanas ‘suitable’, Isaeus, Speech, 8.45) (Wittenberg and Trotzke 2021; Fendel and Ireland 2023). Thus, Xenophon’s choice of one or the other expression was likely conditioned on a context-by-context basis by subtle semantico-pragmatic nuances.

When referring to the state of being clueless and somewhat in despair as a result, Thucydides and in postclassical (Roman) times his fellow historians Lucian and Cassius Dio use the Verbal Idiom ἐν ἀπόρῳ ἔχομαι en aporō ekhomai ‘to be clueless’. The Verbal Idiom seems to be domain-specific and is not used outside of the historiographical context (except in later quotes of the former historians). The Verbal Idiom co-exists with the Light-Verb Construction ἐν ἀπορίᾳ ἔχομαι en aporia ekhomai ‘to be clueless’ which, unlike the idiom, is built around a lexical nominalisation. The Light-Verb Construction is domain-unspecific in classical times albeit limited to prose writings. The historians’ alternative seems to differ not only semantico-pragmatically as outlined above for Verbal Idioms versus Light-Verb Constructions but also in its indexical value. According to Bentein (2019, 123) ‘[l]inguistic indexes are “structures” (lexemes, affixes, diminutives, syntactic constructions, emphatic stress, etc.) that have become conventionally associated with a particular situational dimension, and that invoke that situational dimension whenever they are used’ (cf. Fendel 2024a with further references). While the Light-Verb Construction is domain-unspecific, the Verbal Idiom indexes historiographical writings and possibly something of a technical register. Conversely, in the case of the Verbal Idiom τὰ ἀπόρρητα ποιέομαι ta aporrēta poieomai ‘to keep secrets, to be secretive’ an alternative with a lexical nominalisation (i.e. ἀπόρρησις aporrēsis ‘forbidding’ and ἀπόρρημα aporrēma ‘prohibition’) does not exist before the postclassical period. Thus, the Verbal Idiom with the syntactic nominalisation seems to fill a lexical gap.

Structures with syntactic nominalisation can be Light-Verb Constructions, yet seemingly with a limitation to those syntactic nominalisations that pass the abstractness and predicativeness tests without creating obvious ambiguity between the readings of ‘things’ versus ‘circumstances/behaviours’, unlike e.g. πιστά pista which could refer to the physical items handed over as pledges or the behaviour in the situation. (τὰ) κακὰ ποιέω/ἐργάζομαι (ta) kaka poieō/ergazomai seems to be the standard phrase to express ‘to do evil (to)’. While Light-Verb Constructions referring to ‘to be evil’ (e.g. κακίαν ἔχω kakian ekhō and κακότητα ἔχω kakotēta ekhō) and ‘to become evil’ (e.g. εἰς κακότητα πίπτω eis kakotēta piptō) exist with lexical nominalisations instead of syntactic nominalisations in classical times, structures meaning ‘to do evil’ with a lexical nominalisation do not appear in classical times. The ‘do evil’ structures noticeably appear with direct objects inflected in the accusative case and functionally referring to a Patient. This is otherwise rare in verb-object-type Light-Verb Constructions in Attic non-verse texts. It seems that the doing verb has moved further towards the grammatical sphere in Boye’s (2023) terms and thus the option of a direct object becomes available. The doing verb as a productive light verb is also reflected in Isocrates’ creation of τὸ δίκαιον ποιέομαι to dikaion poieomai (vs. δικαιοσύνη dikaiosunē) ‘to do justice (to)’ instead of the more common plural of the syntactic nominalisation and active voice of the doing verb. Isocrates seems to have drawn upon a frame that was available (see also Fendel 2025c).

Overall, we notice that structures with syntactic nominalisations can fit into the system synchronically as gap fillers, indexical alternatives, or semantico-pragmatic alternatives. Not all structures are Verbal Idioms, but Light-Verb Constructions exist.

5 Summary, Conclusion, and Outlook

The PARSEME GRC guidelines distinguish between Light-Verb Constructions, such as to make a suggestion, and Verbal Idioms, such as to kick the bucket. Light-Verb Constructions need to contain a nominal element that is abstract and predicative. Verbal Idioms are lexically, morphologically, morphosyntactically, and syntactically inflexible. Syntactic nominalisations as the nominal element in Light-Verb Constructions are usually disregarded due to their structural and functional ambiguity. However, this article has considered the eventiveness of the nominal component the crucial characteristic (see research question one) and thus allowed for non-deverbal and underived event nouns, including syntactic nominalisations, as predicative nouns. It has shown that syntactic nominalisations can appear as predicative nouns in Light-Verb Constructions along with forming part of Verbal Idioms (see research question two). Synchronically, verbal multi-word expressions containing a syntactic nominalisation can be gap fillers, indexical alternatives, or semantico-pragmatic alternatives to structures with a lexical nominalisation (see research question three).

(τὰ) κακὰ ποιέω/ἐργάζομαι (ta) kaka poieō/ergazomai seems to be the standard phrase to express ‘to do evil (to)’. While Light-Verb Constructions referring to ‘to be evil’ (e.g. κακίαν ἔχω kakian ekhō and κακότητα ἔχω kakotēta ekhō) and ‘to become evil’ (e.g. εἰς κακότητα πίπτω eis kakotēta piptō) exist with lexical nominalisations in classical times, structures meaning ‘to do evil’ with a lexical nominalisation do not appear in classical times. The classical historian Thucydides and in postclassical (Roman) times his fellow historians Lucian and Cassius Dio use the Verbal Idiom ἐν ἀπόρῳ ἔχομαι en aporō ekhomai ‘to be clueless’, which seems domain-specific and thus seems to carry indexical properties. The historian Xenophon chose between (i) the non-transparent Verbal Idiom δεξιὰν/δεξιὰς δίδωμι/λαμβάνω dexian/dexias didōmi/lambanō ‘to give/receive pledge(s)’, (ii) the transparent Verbal Idiom (τὰ) πιστὰ δίδωμι/λαμβάνω (ta) pista didōmi/lambanō ‘to give/receive pledge(s)’, and (iii) the Light-Verb Construction πίστιν/πίστεις δίδωμι/λαμβάνω pistin/pisteis didōmi/lambanō ‘to give/receive pledge(s)/assurance(s)’ for semantico-pragmatic reasons. Syntactic nominalisation is fully productive, in the sense of generality, in classical and postclassical Greek and offers the option of nuancing the degree (by means of comparatives and superlatives) unlike the derivational morphology.

This study is based on the PARSEME Ancient Greek and ECF Leverhulme Sketch Engine corpora of literary classical Attic philosophical prose, historiography, and oratory with comparative (diachronic) data added where necessary. These are the only two corpora that are suitably annotated for verbal multi-word expressions in classical Greek. The sample sizes are relatively small and samples are non-random such that results do not currently allow for statistical significance testing. Noticeably, direct objects are permissible with those structures consisting of a syntactic nominalisation in the predicative-noun slot and a doing verb in the verb slot, a phenomenon otherwise only observed regularly in Ionic prose and Attic verse. Furthermore, it would be desirable to assess more widely the relationship between lexical and syntactic nominalisations. Structures with lexical nominalisations (and cognate-object constructions) are currently annotated as NotMWE in the PARSEME Ancient Greek corpus, except if they qualify as Verbal Idioms. In future, this will make possible further study on a larger sample and eventually randomisation of sampling for statistical testing.