The paper is a pilot study which argues for a constructionist and computer-based approach to the syntactic and semantic analysis of a poetic corpus in Latin. We focus on the terms felix and on its opposite infelix and perform manual annotation of their occurrences in Virgil's poems using Universal Dependencies for the syntactic analysis and FrameNet for the semantic one. Integrating the approaches of Dependency Syntax and Construction Grammar, we analyze the linguistic contexts in which the two terms occur and identify the different “constructions” (pairings of form and function) that they instantiate. Our methodology is language-independent and has the potential to aid scholars in the comparative analysis of poetic texts, allowing for the detection of hidden parallels in the style and poetics of different texts and authors.
The aim of the present study is to demonstrate the potential of a constructionist and computer-based approach to the analysis of syntax and semantics in a Latin poetic corpus. Our corpus comprises Virgil’s (70– 19 BCE) literary works, namely (in chronological order of composition) the Eclogues (Ecl.) or Bucolics, the Georgics (Georg.), and the Aeneid (Aen.). We focus on two lemmas that have been studied as key terms in Virgil’s poetics (e.g. [1]; [2]): felix ‘productive, auspicious, fortunate, lucky, happy’ and its opposite infelix ‘unproductive, unlucky, ill-fated, miserable’.1
Bellincioni [1] analyzed the meanings of the two terms in Virgil’s works and detected differences in their poetic uses. On the one hand, felix is attested in a variety of contexts, ranging from its (likely original) concrete senses ‘productive’, ‘fruitful’ to more figurative senses linked with prosperity and well-being (granted by divine will). When it qualifies humans, felix takes the religious nuance of ‘favored’ by gods and fate. Gagliardi [2] also stressed the polysemy of felix in the Virgilian corpus: the lemma may refer to fecundity, propitious benevolence, or happiness, acquiring new connotations thanks to innovative uses in Virgil’s poetics. On the other hand, according to Bellincioni [1] infelix is rarely used in the technical sense of ‘infertile’ or in the senses ‘helpless’ and ‘inauspicious’, and in the majority of cases it rather seems to be used to qualify human beings as ‘ill-fated’.
In order to identify patterns of the use of these terms in context, we combine a syntactic analysis with a semantic one. Following Osborne and Groß [4] and Osborne, Putnam and Groß [5], we integrate the approaches of Dependency Syntax and Construction Grammar. In doing so, we rely on the Universal Dependencies (UD) framework for the syntactic analysis and on the FrameNet approach for the semantic one, drawing inspiration from previous studies along these lines (e.g. [6]; [7]).
This integrated approach allows us, on the one hand, to identify the linguistic contexts in which felix and infelix occur in Virgil’s corpus and, on the other hand, to analyze correspondences between the syntactic and the semantic levels of the Virgilian passages where these two terms are employed.
0000-0002-6731-6494 (R. Ginevra); 0000-0001-5100-5539 (F.Iurescia);) © 2024 Copyright for this paper by its authors. Use permitted under
Creative Commons License Attribution 4.0 International (CC BY 4.0). 1 We rely on translations provided by [3].
By combining syntactic and semantic analyses, we explore the potential of an approach that integrates Universal Dependencies with FrameNet. In doing so, we aim at demonstrating that ours is a viable methodology to retrieve the contexts in which the two terms occur in Virgil’s corpus and to study the correspondences between their syntactic and semantic uses.
The term “Construction Grammar” encompasses a series of approaches to grammar, which share the premise that all levels of grammatical analysis involve so-called “constructions”, i.e. “learned pairings of form and function”, including “morphemes or words, idioms, partially lexically filled and fully general phrasal patterns” ([8], p. 5). Within this framework, no rigid division between lexicon and syntax is assumed: constructions are rather arranged along the lexiconsyntax continuum, varying in their degree of internal complexity and schematicity.2 The different instances of constructions (i.e. their tokens in a type-token distinction) are called “constructs”.
Construction Grammar is in turn the formal counterpart of Frame Semantics, originally developed by Fillmore [10], which posits that word meanings are understood through the “semantic frames” they evoke. A semantic frame may be defined as “any system of concepts related in such a way that to understand any of them you have to understand the whole structure in which it fits” ([10], p. 111). The presence in a text of words evoking specific frames reveals different ways in which the speaker conceptualizes the situation.
In order to identify the constructions instantiated by felix and infelix within Virgil’s corpus, the relevant occurrences were analyzed within the framework of Dependency Syntax. This choice aligns with Osborne and Groß’s [4] claim that Dependency Syntax is more compatible with Construction Grammar’s theoretical assumptions and practical goals, compared to Phrase Structure (or Constituency) Syntax.3
Osborne, Putnam and Groß ([5], p. 354) introduced the concept of “catena” to refer to “a word or a combination of words that is continuous with respect to dominance”, and proposed to regard it as the fundamental unit of syntax. As argued by Osborne and Groß [4], most constructions discussed within the framework of Construction Grammar can be analyzed as catenae, i.e. as chains of words linked together by dependencies.
Given the high compatibility of Dependency Syntax with Construction Grammar, we adopt the UD framework [12] to perform the syntactic annotation of sentences in Virgil’s corpus which included occurrences of felix and infelix. The annotation served as a basis for the identification of catenae and of the corresponding constructions.
Our corpus of Virgil’s texts originates from the Opera Latina corpus [13] developed by the LASLA research centre in Liège.4 The Opera Latina corpus is enhanced with sentence-splitting, tokenization, lemmatization, PoS-tagging and the annotation of morphological features according to a format developed by the LASLA team. The texts in the corpus were converted from the LASLA format into the CoNLL-U format, and into the UD formalism [14].5 This textual resource is included among the linguistic resources for Latin that are made interoperable through their linking to the LiLa Knowledge Base.6 The interlinking of the Opera Latina corpus in the LiLa Knowledge Base allowed us to build upon the existent annotation in order to add a further layer. Thanks to the LiLa Interactive Search Platform (LISP), one of the online services designed to query the Knowledge Base [15],7 we were able to retrieve all occurrences of felix and infelix in Virgil’s works: 90 tokens distributed across 89 sentences (see Table 1 in the Appendix).
The sentences were collected into a separate CoNLLU file that was then enriched with syntactic annotation, manually performed according to UD guidelines.8 2 A single expression may instantiate both less complex and phonologically specific constructions (e.g. morphemes, words) and more complex and schematic constructions (e.g. syntactic constructions, such as the transitive one), as long as they may all be analyzed as pairings of form and meaning ([9], p. 7). 3 Constituency Syntax “views the links between the units of sentence structure as indirect” and “mediated by additional groupings that are present as additional nodes in the syntactic structures” ([11], p. 33), in contrast with construction-based approaches, where “no underlying syntactic nor semantic forms are posited” ([8], p. 7). 4 Laboratoire d’Analyse Statistique des Langues Anciennes. (https://www.lasla.uliege.be/cms/c_8508894/fr/lasla). 5 This conversion process was managed by the CIRCSE research center of the Università Cattolica del Sacro Cuore in Milan (https://centridiricerca.unicatt.it/circse/en.html). 6 https://lila-erc.eu/. 7 https://lila-erc.eu/LiLaLisp/. https://github.com/CIRCSE/LiLa_LISP. 8 This annotation will be released as expansion of the UD_LatinCIRCSE treebank (https://github.com/UniversalDependencies/UD_Latin-CIRCSE).
In order to detect the main catenae involving felix and infelix (see Section 2.2), we exploited TüNDRA, a web application for querying treebanks that allows users to upload their own CoNLL-U files.9
Table 2 and Table 3 in the Appendix provide an overview of the tokens’ distributions according to their dependency relation10 (deprel) to their heads. Tokens sharing the same deprel were then systematically analyzed to identify recurrent catenae with varying degrees of extension and abstraction. The analysis took into account the relations between each token of felix or infelix and both the upper and the lower nodes of the trees, starting from the deprel of the token to its head.
In what follows, the identified catenae are conventionally represented using square brackets (as per [11], pp. 60–61), which indicate the degree of dependency between words:
DEPREL1 [DEPREL2 [DEPREL3]]
According to this notation system, dependents are
enclosed in more brackets than their head, thus 0
brackets for the root, 1 for its dependents, 2 for their own
dependents, and so forth, as in the following example:11
(
The instances of the recurrent catenae were then analyzed with respect to their semantic structure. Due to the lack of a resource specifically developed for Latin, the semantic analysis was based on FrameNet,12 a lexical database of English grounded on Frame Semantics. Within this resource, each frame (e.g. APPLY_HEAT) describes a type of event, relationship or entity, along with the participants involved in it, referred to as “frame elements” (e.g. COOK, HEATING_INSTRUMENT, and FOOD), while the words that evoke a given frame are called “lexical units” (e.g. cook, grill, and roast). For the semantic analysis, an expert manually assigned Latin 9 https://weblicht.sfs.uni-tuebingen.de/Tundra/about. 10 UD provides a list of syntactic relations available at https://universaldependencies.org/u/dep/index.html. 11 English translations of Virgil texts are taken from [16]. 12 https://framenet.icsi.berkeley.edu. 13 Georg. I, 345. lemmas to the same frames as their corresponding English translations.
For each instance of a recurrent catena in the corpus, we identified the semantic frames evoked by the tokens that occur with the same deprel within the catena. In what follows, the correspondences between the syntactic and semantic levels of analysis are illustrated by enhancing the notation of the catenae (as per [6], p. 132 and passim) in order to represent them as constructions, i.e. as form-meaning pairings, where frames are represented by superscripts preceding the lexical units that evoke them:
FRAME.ADEPREL1 [FRAME.BDEPREL2 [FRAME.CDEPREL3]]
For instance, the semantic analysis of (
COMMUNICATION_MANNERROOTcano
The most recurrent constructions in which felix and infelix occur allow for the identification of different usages of these two terms in Virgilian poetics. As shown in Table 4 in the Appendix, both felix and infelix often occur as adjectival modifiers (amod) of a noun, but significant differences exist in their respective uses.
Felix is attested only once as amod of a subject (nsubj).13 In 5 out of 17 attestations as amod,14 felix rather occurs as amod of an oblique nominal (obl), i.e. of a non-core argument or adjunct of the verb, in a construction that may denote various entities (winds, tree branches, marriage, death, auspices) and thus evoke various semantic frames: [WHEATER | PLANTS | FORMING_RELATIONSHIPS | DEATH |
In contrast, infelix predominantly occurs as amod of a nsubj, i.e. in 22 out of 40 instances. In 12 occurrences the nsubj refers to human characters,15 but it may also denote other entities.16 This use can be represented by the construction: 14 Aen. III, 118-120; Aen. VII, 598-599; Aen. XI, 29-33; Aen. XII, 819-825; Georg. II, 78-82. 15 Aen. XI, 85; Aen. X, 730; Aen. II, 456; Aen. XI, 563; Aen. IV, 68; Aen. I, 749; Aen. IV, 450; Aen. XII, 870; Aen. I, 712; Aen. III, 50; Aen. VI, 618; Aen. XII, 641. 16 Ecl. V, 37; Georg. I, 154; Georg. II, 314; Aen. XII, 941; Aen. II, 772; Aen. VI, 521; Aen. XII, 608; Georg. III, 37; Georg. III, 498; Georg. II, 198. [PEOPLE | PLANTS | ARTIFACT | ENTITY | ANIMALS |
All in all, infelix is significantly more frequent than felix in our corpus (see Table 1). The distribution of the lemmas in terms of their most frequent dependency relations shows that felix tends to modify adjuncts, while infelix tends to modify subjects (see Table 4).17 Infelix even occurs with the nsubj deprel in 5 occurrences,18 whereas felix never does so.
In what follows we provide two case studies of particularly interesting constructions in which felix and infelix occur.
When infelix and felix occur as amod of a vocative noun or as vocative themselves, they instantiate constructions with different functions, which point to different meanings for the two terms.
As for infelix, 4 occurrences attest the following catena:
[Xverb19 [VOCATIVEinfelix |20 VOCATIVE
[AMODinfelix]] [OBJ] [NSUBJ | OBL [DET]]]
(
All these passages feature a rhetorical interrogative
that conveys emotional turmoil (due either to despair or
frenzy) experienced by the character addressed with the
vocative. In (
[ MENTAL_PROPERTYNSUBJ[DET]]]
As for felix, it occurs as amod of a vocative in two
passages:
(
Both passages attest a verb (dicite and ite, evoking the frames STATEMENT and MOTION, respectively) in the 2pl of the imperative present. The command is first addressed to a larger group (PEOPLE and AGGREGATE), evoked by a vocative (animae and pecus) and described as felix. Then, it is addressed to a specific entity within that group (PEOPLE_BY_VOCATION and ANIMALS), also evoked by a vocative (vates and capellae): [STATEMENT | MOTIONXverb.2pl.imp.21 [PEOPLE | AGGREGATEVOCATIVE [AMODfelix] [PEOPLE_BY_VOCATIONCONJvocative] | [MOTIONCONJverb.2pl.imp. [ANIMALSVOCATIVE]]]]
This construction is in turn a subtype of a more
general construction that also underlies the only
instance of felix as vocative (
VOCATIVEfelix]]
(
As shown by these examples, different constructions are instantiated by felix and infelix when they occur as attributes of a vocative or as vocative themselves. Each construction has a specific function: • the construction with infelix is employed to address the vocative’s referent in a rhetorical interrogative that emphasizes the pathos of the discourse; • the construction with felix is employed to qualify the addressee of a command expressed in the imperative present.
Infelix is used as epithet of Dido, queen of Carthage, in 8 occurrences within the Aeneid.22 In two of these, it instantiates the same complex catena:
ROOTverb.3sg.pres. [NSUBJPhoenissa | Dido [AMODinfelix]
[ACL [OBL | OBL:AGENT ]]] [CONJverb.3sg.pres.]
(
These two examples also attest common semantic
features: they introduce the character of Dido,
conveying the idea of her predestination to a fate of
death and destruction. The passages correspond to
critical points in the plot: in (
DESTROYINGOBL | DESTINYOBL:AGENT]]] [ EMOTION_HEAT | EXPERIENCER_FOCUSED_EMOTION CONJ]
In both (
Moreover, the coordinated verb (conj) in both
instances relates to Dido’s emotional state, which is
different in the two examples: in (
The initial and the final moments of Dido’s story are thus expressed by means of the same catena, evoking her impending ruin. This construction seems to encapsulate the whole thematic arc of Dido’s role in the Aeneid, which is framed both at its inception and at its conclusion by a linguistic structure that highlights the inevitability of her fate. 5. Conclusion and future work With the present study, we show the potential of a constructionist and computer-based approach in the analysis of a poetic corpus in Latin. By integrating syntactic information based on UD with semantic annotation grounded on FrameNet, we were able to identify recurrent constructions involving two key lemmas of Virgilian poetics, felix and infelix. This enabled us to uncover differences and parallels in the uses of these two terms within Virgil’s language.
The present work is a pilot study which may pave the way for future research. Our approach is languageindependent, and may thus be applied to different corpora across various languages and historical periods, for instance to explore similarities in the poetics of various authors within different traditions. Our investigation relied on manual annotation for both the syntactic and semantic analyses due to the lack or poor performance of automatic annotation systems for Latin poetry at the time of writing. The feasibility and effectiveness of such systems can vary significantly across different languages, depending on the resources available. Future improvements in automatic annotation for Latin may allow us to scale up this approach to perform analyses of even larger corpora.
Virgil’s poems played a crucial role in shaping later poetic traditions for centuries: an interesting application of our integrated approach may thus be to investigate whether the same constructions attested in Virgil’s poems also occur in the works of later poets who are known to have been influenced by him, both in Latin (e.g. Valerius Flaccus’s Argonautica, Silius Italicus’s Punica, Publius Papinius Statius’s Thebaid), as well as in other languages, such as Italian (e.g. Dante Alighieri’s Commedia).
Table 2 and Table 3 provide an overview of the tokens’ distributions according to their (deprel) to their heads (“query:edge” in the table) listed in decreasing order:
AMOD [NSUBJ [AMOD]] [OBL [AMOD]] OTHER