1. Introduction

Ontological approaches to morpho-semantics in Modern Greek derivation⋆

Nikos Vasilogamvrakis

Michalis Sfakakis

and Maria Koliopoulou

1 0 , 2 Department of Archives, Library Science and Museology, Ionian University , Corfu , Greece 1 National and Kapodistrian University of Athens / Department of German Language and Literature , Athens , Greece

2025

8 9

In the present article, we explore the affixational morpho-semantics of Modern Greek (MG) and how these can be expressed in the MMoOn ontology extension schema for MG. First, we set our approach by building on Lieber's lexical semantics model and Frame semantics. Accordingly, we adopt a flexible data-driven approach which attributes meaning to morphs horizontally, simplifies the way of this attribution and accounts for cases of morpho-semantic under-specification. Then, to corroborate our approach, we refer to some of the major issues in morpho-semantics such as affix polysemy, affix rivalry and morpho-semantic discrepancies. Afterwards, we try to represent all these cases in our ontological extended schema and provide some indicative SPARQL queries to prove the usability of our modelling.

eol>Linguistic Linked Data Modern Greek derivation Morpho-semantics Ontological morphology

1. Introduction

Language morphology studies two major aspects of words: structure and meaning. By structure we refer to the decomposition of morphological units practiced in concatenative processes [ 9, 10 ] while word meaning consists of the individual semantic properties of morphs1, ranging from pure grammatical, derivational or more extra-linguistic lexical meanings. These types of meaning are not static but are infiltrated within more complex structures during derivational processes. Namely, bits of morphological meaning are transferred onto new derivational entities (derived stems or words) comprising a process known as percolation [ 7 ]. Meaning, however, is not something that can stand alone but it is always expressed morphologically by form. And when we refer to form, we bear in mind its core representative unit, the morph, either a stem or an affix. What we are particularly interested is the morpho-semantics of affixes since it is this type of morph that plays a key role in complex derivational structures. More interestingly, form-meaning relation in affixes is not always one-to-one, but it is very frequent a specific form to have n meanings and a specific meaning to be expressed by more than one form. An instance of form may also happen to be identical with another instance of form, with both having completely different meanings. Other times form becomes inconsistent with the meaning of a structure, which, for example, we encounter in bracketing paradoxes or in conversion cases.

On the other hand, in ontological terms or NLP language representational models, morphosemantics of affixes has not been dealt with sufficiently so far, since most of the models tend to describe their morpho-syntactic aspect (Ontolex, OLiA, Unimorph, Universal Dependencies etc.).

In this article, we deal especially with aspects of morpho-semantics of affixes and how these are processed ontologically within word structures. More specifically, in section 2, we focus on a couple of morpho-semantics approaches, describing related weaknesses and strengths. Then, we explain our approach and represent MG lexical examples ontologically. In section 3, we construct adequate SPARQL queries in the ontology to test our representational analysis in 2, and in section 4, we briefly summarize and conclude on the topic.

2. Morpho-semantics framework and Ontological representation

2.1.

Morpho-semantics approach

Regarding morphological semantics in general, we will assume that there are three types of meaning: grammatical, derivational and lexical. The first pertains to the inflectional and syntactic function of units, including their POS category (i.e. inflectional affixes), the second to the abstract conceptual part dependent on the bases it is attached to (derivational affixes) and the third solely to extralinguistic concepts (stems) [ 7 ]. We will define our approach for morpho-semantics building on Lieber’s Conceptual Semantics (CS) framework [ 3, 4 ], which is in principle decompositional and consistent with the lexicalist approach. Lieber, defines a semantic skeleton of cross-categorial features that pertains to syntax accompanied by a body of encyclopedic meanings. Accordingly, she assumes that like simple lexemes, affixes have some independent meaning, beyond the effect of the argument structure of their base and that these bits of meaning tend to be unitary or closely related among them. To balance the semantic character of features cross-categorially she proposes basically two broad featural atoms, i.e. material and dynamic2, to which she attaches binary (+/) or privative behavior. Figure 1 depicts Lieber’s approach to affixational semantics in English and potentially in all languages. However, despite its flexibility to connect cross-categorially most of English suffixes by exploiting the semantic potential of those two features, her model has three core disadvantages:  It is quite abstract in a way that it cannot be extended to more specific and autonomous affixational micro-meanings (e.g. evaluation, depreciation, profession, origin, doctrine or proponent to say a few), allowing the overgeneration of meanings [ 2 ].  Despite its resourcefulness, the difference between binary (+/-) and privative semantic be havior is very indefinite. Additionally, one could oppose that should a feature be necessary to be absent, the presence of its opposite or the attribution of a more accurate feature would be sufficient. In other words, what would be the difference, if any, between +literate and -popular pertaining to style? Or if it would be necessary to define semantics more ac curately (e.g. +colloquial or +demotic)3 why wouldn’t that be sufficient?

There is a lack of justification for cases of under-specification, where a feature is latent because there is no morph to carry its semantics or - otherwise put - there is a zero morph. For example, in Modern Greek (MG) there is no morph (inflectional suffix) in the noun μητέρα ‘mother’ to carry the Case and Number values (nominative, singular) or there is no derivational suffix present to change the POS category in the ‘supposed’ conversion of οδηγώ ‘to drive’ <> οδηγός ‘driver’ and vice versa. 2 She also defines two other features, location and quantity. 3 These may well be accompanied by additional features (e.g. +vulgar), if necessary, to determine style more accurately. Copyright © 2025 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). Inherently decompositional is also the Frame Semantics model [ 2, 5 ], which we cite here for analyzing affix polysemy. In Figure 2, we see how the three semantic instances of the suffix -ment (event, stimulus, result state) in English can be identified with each of the base argument. Each semantic attribute is connected to its value (-ment1, -ment2, -ment3) in the same way an ontology would connect two instances via an object property (e.g. has meaning). It is interesting here that all three different instances of -ment can be simultaneously found in the same lexical instance (e.g. confoundment), each of which is combined with the appropriate semantic argument of the verbal base (e.g. change-of-state-verbs, psych causation verbs etc.).

2.1. Our approach to morpho-semantics

To deal with CS framework lack of specificity we propose to:  Extend feature semantics to all possible micro-meanings, either abstract or specific. We also assign these horizontally and not hierarchically.  Disassociate +/- signs with featural semantics and assign (+) to features specified by the presence of a morph and (-) to features not specified by that presence. However, since in most cases a morph is present as the carrier of a semantic feature, it would be more eco nomical to define just the feature f with no (+) signage at all (e.g. f and not +f). We should also note that in cases of morph absence, we generally assume that a zero-morph resides for carrying the latent or underspecified feature. This can be either derivational or inflectional, carrying normally the features of their present counterparts with the addition of a (-) sign. Furthermore, to represent morpho-semantics ontologically:  For affixational polysemy we define a morpho-semantic pattern X1_1, where the first number denotes the morphemic form and the second number divided by a hyphen _ the morphemic meaning.  Thus, for multi-form morphemic instances (morph homonyms) we would have X1_1, X2_1 etc. and for multi-meaning X1_1, X1_2 etc.

For representing our model, we use the MMoOn4 model, as we have done before. However, as we unfold the Meaning class of the model we have to make the following observations:  The model attributes the same name to a class and an individual (e.g. Diminution hastype Diminution etc.) in the Derivational Meaning class, adopting OWL 2 punning technique. For now, we will not stick with that till we evaluate its functionality. However, the dividing of Derivational Meaning sub-categories is overall questionable as some derivational features can refer to more than one derivational class. For example, the Diminutive derivational meaning is used here only as a characteristic of Noun Expansion, which is not true in MG (see Table 1, μικρόςADJ ‘small’ > μικρούτσικοςADJ ‘tiny’ for adjectival diminution) while both Diminution and Augmentation would rather refer to a more abstract Evaluation category as subclasses.  Categorial change representation (e.g. Adjectivization) would be more functional if it would incorporate both the initial and final category. As we have elsewhere observed [ 8 ], the categorial change can be incorporated in the Meaning class just once in the model, decongesting the Morphological relationship class. Thus, we proceed to form X-X and X-Y instances for Expansion and Word-class Changing Derivation respectively, representing the initial and the final category of the source and target word in one instance. Then, via the object property owl:sameAs we map each of the categorial combinations to the respective mmoon conceptualization, e.g. ell_schema:N-Adj owl:sameAs mmoon:Adjectivization etc. (Figure 3).

Grammatical category

adjective noun

Feature5 MG Morph Input-Output word

diminutive -ούτσικ-1_1 (-ος) etc.

-ούλη-1_1 (-ς) etc.

μικρός > μικρούτσικος etc.

μικρός > μικρούλης1 etc. -ούλη-2_1(-ς) μικρούλης1 > μικρούλης2 4 The ttl file of the MMoOn core model can be reached here: https://github.com/MMoOn-Project/MMoOn/blob/master/core.ttl.

 There are cases where closely related or identical meanings are repeated between types of derivation e.g. Derivative Negation in the Expansion class and Privative Adjectivizer in the Semantic Derivation class. Since affixes express abstract chunks of meaning that tend to be common between grammatical categories, it would be more economical and consistent to define a feature just once and use it for every type of derivation.

For example, in Table 1, we can view the analysis of affix rivalry [ 2 ] of the two suffixes ούληand -ούτσικ- attached to the same base μικρ-. In this case, we have to note that the two suffixes:  Attach to the same base μικρ- (Adj)  Belong to the same grammatical category, -ούλη- (Adj) and -ούτσικ- (Adj)  Express the same derivational semantics, i.e. diminutive

Then, in Figure 4, we represent this analysis ontologically. It has to be noted, here, that the instances μικρούλης1, μικρούλη-1 and -ούλη-1 should be different from μικρούλης2, μικρούλη-2 and -ούλη-2, because μικρούλης2Ν has been derived from the conversion or the zero affix derivation of μικρούλης1ADJ (μικρούλης1 ‘pretty small or young’ > μικρούλης2 ‘the pretty small or young’).

Is attached to nouns Denotes a feature closely related to the base. Therefore, its derivational meaning complements the argument and the lexical semantics of the base Has an adjectival grammatical meaning It expresses more than one meaning -ιν-1_1, -ιν-1_2, -ιν-1_3, -ιν-1_4, adhering to our pattern X1_Μeaning1/Μeaning2/Μeaning3/Meaning4.

Copyright © 2025 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). αλήθεια > αληθινός Ζάκυνθος > Ζακυνθινός κοντά > κοντινός νύχτα > νυχτερινός χαλκός > χάλκινος In its ontological representation (Figure 5), we exploit the ontology class Representation, where all different semantic instances of -ιν- (-ιν-1_1, -ιν-1_2, -ιν-1_3, -ιν-1_4) end up to one common morphemic Representation instance Rep_-ιν-.

Finally, in Table 3, we analyze the homonymous but morphologically different morph μετα-. In the word μετα-τρέπω the morph μετα- is a prefix that comes from the grammaticalization of the preposition μετά [ 1, 6 ], while in the word μετα-αποικιακός the morph μετα- constitutes a confix that has been inserted in MG as a translation of the foreign prefix post- (post-colonial = μετααποικιακός).

Grammatical category Feature5 MG Morph Input-Output word

τρέπω > μετα-τρέπω post αποικιακός > μετααποικιακός Both morphs have the same representations, but they are completely different morphs. That is why, in Figure 6, we have created two different Representation instances, Rep_μετα-1_1 and Rep_μετα-2_1. 2.2.

Form-meaning mismatches

Usually, the morphological structure of a derived word corresponds to the semantic structure of that word. For example, the word χορευτής is morphologically structured according to its meaning: [[χορευ]VτήςN] = someone that dances

There are times, though, where this morphological principle does not apply, forming a bracketing paradox e.g. the words αχαρακτήριστος ‘someone that cannot be defined’ or αφύλακτος ‘someone that cannot be guarded’ are structured according to the semantic structure in (1): Copyright © 2025 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

(1) *[α[χαρακτήρισ]VτοςN] (2) α[[χαρακτήρισV]τοςN] which is basically ungrammatical because the prefix α- is attached only to nominal bases (Nouns, Adjectives) as depicted in (2):

One solution in (1) is that the possible words i.e. *χαρακτηριστός and *φυλακτός would be included in the MG lexicon [ 4 ] so that the prefix α- could be attached to them. The latter, however, would be not absolutely right because it would not depict the derivational process pragmatically and a further derivational rule would also have been necessary. What we propose instead is to create a morphemic discontinuous pattern of α-_-τ- that would attach to verbs like that (see Figure 7). This pattern would be rather preferable because it would explain similar productive cases with the addition of a double derivational rule (prefixation and suffixation): φυλάσσω > αφύλακτος, χαρακτηρίζω > αχαρακτήριστος, προσδιορίζω > απροσδιόριστος, λυγίζω > αλύγιστος etc. In Figure 7, we can view the representation of χαρακτηρίζω > α-χαρακτήρισ-τος as a double Prefixation and Suffixation derivational rule coming from the α-_-τ- pattern introduced as a morph discontinuity entity. This enables the implementation of two concomitant rules and bridges the form-meaning inconsistency. Then, because α-_-τ- is a complex entity, it should further consist of two morphemic sub-entities, i.e. the prefix α- and suffix -τ. It has also to be noted that this morph discontinuity takes the grammatical meaning of the suffix -τ-, adhering to the right head rule elaborated in [ 7 ].

What is more, there is often a special type of derivation, i.e. conversion, which is also called zeromarked cases affixation because of the absence of a derivational morph. In these cases, a change in meaning does not correspond to a change in form. Either there are times, where both directions between input and output word are possible e.g. οδηγός ‘driver’ <> οδηγώ ‘to drive’ [ 8 ] or there is just one direction possible (e.g. Κρητικός ‘cretan’ > Κρητικός ‘inhabitant of Crete’). Rally [ 6 ] argues that the former are related to a conversion pleonastic rule while the latter can be explained by a zero Copyright © 2025 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). derivational suffix. Lieber [ 4 ] claims that “conversion is in fact not a form of derivation at all but rather is a form of lexical relisting of items” [5, p. 2116]. However, we think that this view lacks sufficient argumentation in terms of the change of the category. By contrast, and according to Ralli’s view [ 6 ], we prefer to model two grammatically different bases οδηγ-1 (nominal) and οδηγ-2 (verbal) as start-offs towards the one word or the other. Then, we just have two derivational options: either οδηγός > οδηγώ or οδηγός > οδηγώ but not both simultaneously (see case (3) below and Figures 8, 9). To this problem, we think that the presence of a zero derivational suffix would provide sufficient justification for the change of the category in all the below cases (3), (4), (5) and (6), because it would be consistent with percolation, a process attested in MG [ 7 ]. We also divide this special type of suffix by grammatical category and gender: zDerSuf1=noun/masculine, zDerSuf2=noun/feminine, zDerSuf3=noun/neutral, zDerSuf4=verb). In this case, we define in place of conversion a suffixation rule:

From Verb to Noun and vice versa (3) οδηγ1N > οδηγόςN > [[οδηγ1N]zDerSuf4V]ώ

οδηγ2V > οδηγώV > [[οδηγ2V]zDerSuf1N]ός From Adjective to Noun (4) [ΚρητικADJ]ός > [[ΚρητικN]zDerSuf1N]ός (5) [νομικ ADJ]ός > [[νομικN]zDerSuf1N]ός From Noun to Noun (6) [κεφάλαN] > [[κεφάλαN]zDerSuf1N]ς [γυαλάκιαN] > [[γυαλάκιαN]zDerSuf1N]ς Copyright © 2025 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

3. Usability through SPARQL

To see the ontological applicability and usage of the previous morpho-semantic representations, in this section, we construct SPARQL queries for different morpho-semantic cases. For executing the queries, the respective SPARQL syntax is given below as well as the necessary prefixes referring to the used ontologies. These queries should be executed into the last version of Modern Greek inventory: https://github.com/nvasilogamvrakis/mmoon_project/blob/main/inventory/finals/ ell_inventory1.owl. This file also imports the last version of ell_schema: https://github.com/nvasilogamvrakis/mmoon_project/blob/main/ell_schema/finals/ell_schema1.owl as well as the MMoOn core ontology: https://github.com/nvasilogamvrakis/mmoon_project/blob/main/mmoon/mmoon_core.owl. PREFIX : <http://mmoon.org/ell/inventory/omg/> PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> PREFIX owl: <http://www.w3.org/2002/07/owl#> PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#> PREFIX xsd: <http://www.w3.org/2001/XMLSchema#> PREFIX mmoon: <http://mmoon.org/core/> PREFIX ell_schema: <http://mmoon.org/ell/schema/omg/> 3.1.

Affix rivalry and polysemy

SELECT ?a ?b

WHERE { ?a rdf:type ell_schema:DerSuf . ?a mmoon:derivationalMeaning ?b .} ORDER BY ?b 3.2.

Morph homonymy

SELECT ?a ?b ?c ?d

WHERE { ?a mmoon:hasRepresentation ?b .

?b mmoon:morphemicRepresentation ?c .

?b mmoon:phoneticRepresentation ?d .} ORDER BY ?a 3.3.

Morph discontinuity entity

SELECT ?a ?b ?c

WHERE { :χαρακτηρίζω ell_schema:ell_generates ?a . ?a mmoon:consistsOfMorph ?b .

OPTIONAL { ?b mmoon:consistsOfMorph ?c .} } 3.4.

Zero affix derivation

SELECT ?a ?b ?c

WHERE { ?a rdf:type ell_schema:zDerSuf . ?a mmoon:belongsTo ?b .

?b mmoon:belongsTo ?c . } ORDER BY ?a

4. Conclusion and Future research

In the previous analysis, we have dealt with aspects of morpho-semantics of affix atoms in MG derivation. We have placed our approach within Lieber’s lexical semantics (LS) framework for its lexicalist character and for exploiting its presence-absence semantics in a way that is more accurate and functional. We also paralleled Frame Semantics approach to our modelling of different semantic connotation and have shown cases where a horizontal representation of featural micro-meanings can prove practical and consistent. We dealt specifically with issues of affix polysemy, affix rivalry or multi-form semantics and cases of conversion or zero affix derivation. In special cases such as bracketing paradoxes, we proposed a new morpho-ontological entity for bridging the morphosemantic gap. All these issues were modeled in our extended MG schema of the MMoOn core ontology by adequate lexical examples and specific representational solutions were proposed. Finally, indicative SPARQL queries have been executed to test the linguistic usability of the examples in the ontology. In the research to come we aim to insert a minimum of language data into the ontology with carefully selected lexical instances so that specific SPARQL queries can be executed to analyze data morphologically according to related linguistic quests.

Declaration on Generative AI

The author(s) have not employed any Generative AI tools.

[1] Giannoulopoulou , G. : Morphosemantic comparison between affixes and confixes in Modern Greek and Italian (in Greek) . Aristotle University of Thessaloniki (AUTH). School of Italian Language and Literature ( 1999 ).

[2] Kotowsky , S. , Plag , Ingo eds: The Semantics of Derivational Morphology: Introduction . In: : Theory, Methods, Evidence. De Gruyter ( 2023 ). https://doi.org/10.1515/9783111074917.

[3] Lieber , R. : Morphology and

Lexical

Semantics . Cambridge University Press, Cambridge ( 2004 ). https://doi.org/10.1017/CBO9780511486296.

[4] Lieber , R. : Semantics of derivational morphology . In: Heusinger, K. von et al. (eds.) Semantics: An International Handbook of Natural Language Meaning . pp. 2098 -2119 De Gruyter Mouton ( 2012 ). https://doi.org/10.1515/9783110253382.

[5] Plag , I. et al.: A frame-semantic approach to polysemy in affixation . In: The lexeme in descriptive and theoretical morphology ,. pp. 467 - 486 Language Science Press, Berlin ( 2018 ).

[6] Ralli , A. : Morfologia (in Greek) . Patakis , Athens ( 2005 ).

[7] Vasilogamvrakis , N. et al.: Ontological Approaches to Morphological Semantics in Modern Greek Derivation . In: Sfakakis, M. et al. (eds.) Metadata and Semantic Research . pp. 337 - 348 Springer Nature Switzerland, Cham ( 2025 ). https://doi.org/10.1007/978-3- 031 -81974-2_ 28 .

[8] Vasilogamvrakis , N. et al.: Testing the Word-Based Model in the Ontological Analysis of Modern Greek Derivational Morphology . In: Chiusano, S. et al. (eds.) New Trends in Database and Information Systems . pp. 572 -584 Springer International Publishing, Cham ( 2022 ). https://doi.org/10.1007/978-3- 031 -15743-1_ 52 .

[9] Vasilogamvrakis , N. , Sfakakis , M.: A morpheme-based paradigm for the ontological analysis of Modern Greek derivational morphology . In: MTSR 2021 . Communications in Computer and Information Science . , Online ( 2021 ).

[10] Vasilogamvrakis , N. , Sfakakis , M. : Ontological Modeling of Lists for Modern Greek Derivation . In: Garoufallou, E. and Sartori , F . (eds.) Metadata and Semantic Research . pp. 181 - 192 Springer Nature Switzerland, Cham ( 2024 ). https://doi.org/10.1007/978-3- 031 -65990-4_ 16 .