This paper presents the manual annotation of ellipsis phenomena in a Latin treebank syntactically annotated following Universal Dependencies framework. Building on ongoing research in the field, it provides an overview of syntactic constructions that pose particular challenges for the annotation and reconstruction of ellipsis. By providing Latin examples, the work contributes to cross-linguistic comparisons and the broader understanding of ellipsis across languages. As a preliminary contribution, the paper ofers insights into aspects that remain underspecified in current UD guidelines, suggesting directions for future refinement of annotation standards.
This work describes manual annotation of ellipsis in treebanks, which are syntactically annotated texts. The source data is a portion of UD_Latin-CIRCSE Treebank, a treebank manually annotated following Universal Dependencies (UD) framework.1At the time of writing, the treebank consists of three tragedies authored by Seneca (1st CE) – Hercules Furens, Agamemnon, Oedipus – and a treatise authored by Tacitus (1st-2nd CE) – Germania. Tacitus’ Germania, being a prose work, was chosen for the present study to circumvent the stylistic challenges associated with tragic texts, particularly those arising from their diverse metrical structures. We relied on gold data as in the 15th UD release.2Nevertheless, the annotation did present several challenges and provided valuable insights to support ongoing research in the field of ellipsis in UD. Section 2 describes the state of the art in this ifeld. Section 3 illustrates some examples of ellipsis and outlines the challenges encountered during the annotation. Section 4 concludes the paper and outlines potential avenues for future research.
Before delving into the core of the present work, this section provides a general introduction to UD (subsection 2.1), ellipsis (subsection 2.2), and the annotation of ellipsis in UD (subsection 2.3).
Universal Dependencies is a framework for
morphosyntactic annotation of diferent human languages [
Being a phenomenon that operates at the intersection unnatural syntactic structure, the orphan dependency
reof diferent linguistic domains [ 4, p. 341], ellipsis has been lation is used instead. However, this dependency relation
the subject of numerous studies from various perspec- inevitably obscures the underlying syntactic structure,
tives and theoretical frameworks. Concerning syntactic thereby entailing a loss of syntactic information.
analysis, a substantial body of research has addressed the It is therefore evident that UD does not directly
adtopic within a constituency-based approach ([
In this sentence, predicate ellipsis results in the promotion of the accusative reges (“kings”) to the root position, leaving the dependent ex nobilitate (“on the ground of birth”) orphaned.
This structure is illustrated in the basic dependency tree, where reges governs both nobilitate as orphan and sumunt (“they take”) as a conjunct (conj).
Basic dependencies fall short of adequately representing implicit structures such as ellipsis. Instead, to thoroughly annotate elliptical constructions, a more suitable strategy within the UD framework is ofered by enhanced
As mentioned in section 2.2, dependency-based treebanks are generally not inclined to introduce empty nodes into the syntactic structure. Currently, the most widely adopted and state-of-the-art framework for dependency treebanks is UD (see section 2.1), which handles ellipsis diferently depending on the level of annotation—basic or enhanced. 7https://universaldependencies.org/u/overview/specific-syntax. html#ellipsis 5In the literature, the term antecedent is typically used in a broad 8https://universaldependencies.org/u/overview/specific-syntax. sense, not necessarily referring to an element that precedes the html#ellipsis-in-nominals ellipsis. In fact, the element supplying the missing content may 9For a proposal on explicitly marking ellipsis with a dedicated also follow the ellipsis, in which case the term postcedent would subtype, see https://unidive.lisn.upsaclay.fr/lib/exe/fetch.php? be more accurate. However, to remain consistent with established media=meetings:general_meetings:3rd_unidive_general_meeting: usage, we use the term antecedent in both cases. 23_how_to_ellipsis_a_proposal_.pdf 6Refer to Mikulová [16] for further discussion of ellipsis in the PDT. 10Latin translations are drawn by [19].
Given the current state of enhanced dependency annotation and the importance of gold-standard annotated data, we adopt the approach proposed by [20], which ofers a consistent annotation scheme explicitly designed to address predicate ellipsis. The syntactic annotation of ellipsis in Tacitus’ Germania was performed manually by a team of four annotators equipped with a background knowledge of Latin language and literature and expertise on syntactic annotation in the UD framework.13
More specifically, in this work we focus on predicate ellipsis, presenting cases beyond this scope when they are relevant to illustrate annotation-related issues and decisions. We extend the enhanced annotation to cover primarily cases in which the basic annotation layer displays the orphan dependency relation, as well as, where applicable, cases in which predicate ellipsis is conveyed without being explicitly marked by means of orphan.
There are 71 tokens out of 5,674 annotated with the deprel orphan in the basic layer.14 They are distributed across 49 sentences out of a total of 299 sentences.15 No
The main instance of predicate ellipsis found in the Germania is gapping. Gapping is a type of ellipsis in which a repeated verb is omitted from a coordinated clause, leaving behind only the contrasting elements [22]. Example 1a illustrates the enhanced annotation of the sentence presented in example 1:16
Example 1a – Enhanced Annotation reges ex nobilitate (sumunt) duces ex uirtute sumunt
conj reges ex nobilitate (sumunt) duces ex virtute sumunt king.acc.m.pl from nobility.abl.f.sg (take.prs.3pl) general.acc.m.pl from courage.abl.f.sg take.prs.3pl
obl root orphancase obj case obl dependencies relations. At the current state of the art, en- specific pattern emerges regarding the distribution of hanced dependencies for ellipsis involve the reconstruc- elliptical constructions across the text. tion of an empty node in the syntactic structure, along In what follows, we report a few examples of reconwith the recovery of the relevant syntactic relations.11 struction of predicate ellipsis (subsection 3.1), of the However, since enhanced annotation is "an optional ad- treatment of orphan deprel in cases of nominal ellipsis dition to the basic representation",12 the guidelines for (subsection 3.2) and of ellipsis in copular constructions annotating such cases remain underspecified. (subsection 3.3). We then illustrate cases of ellipsis in predicative (subsection 3.4) and comparative structures (subsection 3.5), providing examples of how we decide 3. Ellipsis in Tacitus’ Germania whether or not to intervene. Finally, we present an example of nested ellipsis (subsection 3.6).
obj root 11For enhanced guidelines see: https://universaldependencies.org/
v2/enhanced.html 12https://universaldependencies.org/v2/enhanced.html 13The annotators–three students from the graduate program in Linguistic Computing and one in Theoretical and Applied Linguistics– developed their expertise in syntactic annotation in the context of a dedicated training workgroup (approximately 30 hours) held at the CIRCSE research centre at the Università Cattolica del Sacro Cuore, Milano. In a first stage, the annotators practised on basic annotation layer by reannotating portions of gold standard texts from the CIRCSE treebank, both from poetry and prose. In this 3.2. Nominal Ellipsis stage each dataset was annotated by all students. Their Inter Annotator Agreement (IAA), measured using Fleiss’ Kappa coeficient, Example 2 illustrates an example of nominal ellipsis yielded an average score of 0.86 for edges and 0.85 for deprels. marked with the deprel orphan in the basic layer. The (Our script for calculating IAA on CoNLL-U files is available at enhanced tree of this and of the following examples are shetctposn:/d/gsittahgueb,.tchoema/njonhontantyomrsofroecttuis/CseodNoLnL-thUe_Faldevisasn_cKeadplpaay)e.rI,nbya detailed in the appendix A. annotating ellipsis in Tacitus’ Germania. Due to time constraints, in this stage each annotator was responsible for the annotation of a single portion of the text. The annotation of the single subsets Example 2 were subsequently proof-checked by two expert annotators with Tac., Germ. 11,2 advanced competence in Latin language and literature and exper- nec dierum numerum ut nos sed noctium tise in the syntactic annotation of ellipsis. This process involved computant collaborative discussions during which the annotators presented They count not by days as we do, but by nights the issues they encountered. 14Two tokens out of 71 are annotated with orphan:missing due to a lacuna in the Opera Latina corpus [21], which is the source 4 and 6 tokens annotated as orphan, respectively, whereas the text of the treebank. average distribution of orphan equals to 1.44 tokens per sentence. 15Two sentences (Tac., Germ. 2,3 and 37,4) stand out, as they present 16For gapping in Latin, see, e.g., [23, p. 585-586; 610-611]. conj
In the enhanced UD annotation, the elided verb sumunt is introduced as an empty node, explicitly marking the omitted predicate. Consequently, in the enhanced tree graph representation, the empty node becomes the root, governing reges as its direct object (obj) and nobilitate as an oblique dependent (obl). Finally, the empty node governs sumunt as conj.
This sentence illustrates a simple case of nominal el- reference to. Unlike more typical cases of predicate or lipsis,17 where the elided element is the noun numerum comparative ellipsis (where the verb or noun is missing (“number”) before noctium (“of nights”). In the basic but recoverable from a parallel structure) here, no noun UD dependency tree, the ellipsis is captured by the pres- referring to e.g., “portion” appears in the clause, distinence of the orphan relation assigned to the dependent guishing this case from anaphoric ellipsis. Then, in the noctium, directly governed by the main verb computant absence of an antecedent, the reconstructed node in our (“they count”). enhanced annotation is just empty, and not a copy of
In the enhanced dependency graph, the elided noun something else. So, the new reshaped phrase would apnumerum is introduced as an empty node to resolve the pear like this: conferre principibus uel armentorum uel syntactic discontinuity. Thus, noctium is annotated as a frugum _ , with the empty node occupying the last posidependent of the reconstructed empty node numerum via tion.18 the nmod relation. This structure reproduces the expected Example 4 features an instance of ellipsis of the subject syntactic structure and mirrors the dependencies found which lacks an overt antecedent, requiring the reconin the first clause nec dierum numerum, where the root struction of an empty node not filled with any lexical or numerum governs dierum as its nominal modifier nmod. morphological information.
Example 3 is a an instance of nominal ellipsis where a missing antecedent led to creation of an empty node, which could be semantically inferred, but still does not have an actual linguistic counterpart in the phrase:
Tac., Germ. 15,1 mos est ciuitatibus ultro ac uiritim conferre principibus uel armentorum uel frugum quod pro honore acceptum etiam necessitatibus subuenit It is the custom in their states to bestow upon the chief unasked and man by man some portion of one’s cattle or crops: it is accepted as a compliment, but also serves his needs
In this sentence, the verb conferre (“to bestow”) governs an argument in dative (principibus, “upon the chiefs”) and semantically requires a direct object representing what is being bestowed. The coordinated genitives armentorum (“of cattle”) and frugum (“of crops”) function as modifiers of the verb, but not as arguments. They would imply a partitive relationship, and require an implicit head noun (e.g., pars, “a portion”) to form a semantically complete direct object depending on conferre.
In the basic UD representation, this ellipsis is captured through the presence of an orphan, while frugum is a conjunct. The orphan label here marks the gap in the basic annotation: the genitive lacks a head noun, and the sentence lacks a direct object for conferre. This missing noun is both semantically necessary and syntactically expected, and this motivates its reconstruction. So, in the enhanced annotation, we reconstruct the argument expectations of conferre and introduce: a new empty nominal node functioning as the direct object of conferre; armentorum as nmod of the nominal empty node and frugum as its conjunct; the elimination of the orphan relation.
Crucially, we add an empty node without storing any lexical content, as there is no explicit antecedent to make 17For nominal ellipsis in Latin, see, e.g., [24, p. 962].
Tac., Germ. 26,1 faenus agitare et in usuras extendere ignotum ideoque magis seruatur quam si uetitum esset To exploit capital and to increase it by interest are unknown, and the principle is accordingly better observed than if there had been actual prohibition
The first clause describes the unfamiliarity of usury among the Germans. The second clause has the verb seruatur (“observed”) as head: its implicit subject is understood as the negation of the previous clause, faenus non agitare neque in usuras extendere (“not to exploit capital and not to increase it”), whose negative meaning has to be inferred from ignotum (“are unknown”).
Following the interpretation adopted by [19] in their translation, we reconstruct a single empty node for the omitted subject, representing this implied negative concept that may be paraphrased as a generic “principle”. Hence, the reconstructed node is assigned the dependency relation nsubj:pass, under a nominal reading of the elided material. For reasons of accuracy, no lexical or morphological features are encoded, as no explicit antecedent is present in the text.
Example 3 and 4 serve to exemplify the procedure adopted in analogous cases: when no explicit antecedent is present in the text, the reconstructed node is left lexically and morphologically underspecified, and only the appropriate dependency relation is assigned. The absence of a textual antecedent is indicated in the MISC field of the CoNLL-U file by marking the source of interpretation as world knowledge (wk).
Example 5 illustrates a case of ellipsis of a copular construction. Such cases concern the omission of a predicate 18For a tentative recostruction of the elided part, see, e.g., [25, p. 41], who posits ellipsis of an indefinite pronoun, as head of uel armentorum uel frugum. formed by the nominal component and a form of the verb sum.
Tac., Germ. 27,2 feminis lugere honestum est uiris meminisse Lamentation becomes women: men must remember
This sentence illustrates a case of nominal predicate ellipsis, involving the omission of both the copula est and the nominal component honestum (“convenient”). The structure is parallel, contrasting actions appropriate for women and man in mourning. In the first clause, the argument feminis (“for women”) precedes the clausal subject lugere (“to mourn”) of the nominal predicate; in the second, uiris (“for men”) is followed by the infinitive meminisse (“to remember”), on which it depends, in basic annotation, via an orphan relation that marks the ellipsis of the predicate.
Hence, in the enhanced annotation, the ellipsis is resolved through the insertion of an empty node at the end of the second clause, mirroring the structure and the dependency relations of the first. In accordance with the proposal in [20], we reconstruct only the nominal part of the predicate, the head honestum, which carries the semantic content of the predicate and ensures crosslinguistic consistency.
As mentioned in section 2.3, ellipsis is annotated only when it creates syntactic discontinuities, specifically, when the absence of a word leaves dependents orphaned. This is most visible in constructions like gapping (section 3.1), while in other structures, an elided predicate does not result in unattached dependents or broken syntactic structure, and therefore there are cases of elliptical constructions which are not annotated as orphan in UD (section 2.3). Example 6 is a case of predicative constructions involving coordinated or juxtaposed clauses:
Tac., Germ. 13,1 ... scuto frameaque iuuenem ornant ... ante hoc domus pars uidentur, mox rei publicae ... [they] equip the young man with shield and spear ... up to this point they seem a part of the household, next a part of the state
We focus on the final segment of the sentence: ante hoc domus pars uidentur, mox rei publicae (“up to this point they seem a part of the household, next a part of the state”). The first clause ( domus pars uidentur) is a standard predicative construction consisting of a subject (understood as illi and referring to iuuenem), the verb the verb uidentur and the predicative noun pars are absent. Nevertheless, they can be clearly inferred and the intended structure formed by the two clauses is parallel: domus pars uidentur (“they seem a part of the household”) rei publicae pars uidentur (“they seem a part of the state”).
However, despite the interpretative clarity, as both the verbal head and its nominal predicate are missing, this is a clear case of predicate ellipsis that is not marked in the basic UD annotation, since no dependent is left syntactically orphaned.
So, in the enhanced UD annotation, we insert two reconstructed nodes: the verb uidentur as the verbal head and the predicative nominal pars, which has rei publicae as a dependent. Therefore, in the enhanced tree graph representation the structure would include: a new verbal node uidentur connected to the first verb with the deprel conj; a new nominal node pars connected as xcomp to the reconstructed uidentur; the nominal phrase rei publicae dependent of pars as nmod.
Another significant case of ellipsis in the Germania occurs in comparative clauses.19 The following case is one of this kind:
Tac., Germ. 14,3 nec arare terram aut exspectare annum tam facile persuaseris quam uocare hostem et uulnera mereri pigrum quin immo et iners uidetur sudore acquirere quod possis sanguine parare You will not so readily persuade them to plough the land and wait for the year’s returns as to challenge the enemy and earn wounds: besides, it seems limp and slack to get with the sweating of your brow what you can gain with the shedding of your blood
In this sentence, the verb persuaseris (“you will persuade”) governs the first clause of the comparative construction (with the two infinitive verbs arare, “to plough”, and expectare, “to wait”) but is not repeated in the second (verbs uocare, “to challenge”, and mereri, “to earn”), despite clearly being the intended meaning. This omission might seem to be a candidate for ellipsis annotation.
However, UD guidelines do not annotate such cases of comparative ellipsis for several theoretical and practical reasons, and we decide to do the same.
There is no orphan relation in the basic tree, so there is no ellipsis to resolve in the enhanced representation either. We see that there are no orphaned dependents in the second clause, as the infinitive verb uocare depends on persuaseris as advcl:cmp and mereri is its conjunct. Since all constituents are attached with appropriate uidentur (“they seem”), and the predicative nominal do- 19For ellipsis in comparative construction in Latin, see, e.g., [23, p. mus pars. In the second clause (mox rei publicae), both 721]. heads, no orphan relation is needed. Being the sharing of the predicate a common phenomenon in comparative constructions, especially in Latin, we accept this syntactically and semantically recoverable pattern, and choose not to annotate it. So, even by doing nothing, here the clause remains structurally intact, and no dependents are left without a head, although persuaseris is intuitively present in both parts of the comparative construction.
Among the knotty sentences to annotate, example 8 stands out: exemplifying Tacitus’ concise and condensed prose, it features a case of a nested ellipsis.
Tac., Germ. 43,2 e quibus Marsigni et Buri sermone cultuque Suebos referunt Cotinos Gallica Osos Pannonica lingua coarguit non esse Germanos et quod tributa patiuntur Among them the Marsigni and Buri in language and mode of life recall the Suebi: as for the Cotini and Osi, the Gallic tongue of the first and the Pannonian of the second prove them not to be
Germans; so does their submission to tribute
The two clauses Cotinos Gallica and Osos Pannonica lingua coarguit are in asyndeton, with no subordinating connective present: indeed, in accordance with the guidelines, their relationship has been annotated with the label conj. The clause Cotinos Gallica shows a predicate ellipsis (coarguit) and instantiates an example of gapping (section 3.1): the use of a singular verb would otherwise be inexplicable.
What makes the situation more complex is that the reconstructed coarguit implies, even in this first clause, non esse Germanos. Therefore, another empty node (Germanos) has been reconstructed after Cotinos. In the enhanced UD annotation, the reconstructed coarguit thus takes on the role that Gallica played in the previous annotation; Cotinos, on the other hand, becomes the nsubj of the reconstructed ccomp Germanos.20
This work describes the challenges encountered during manual annotation of ellipsis in Tacitus’ Germania. The treebank enhanced with this annotation will be included in the next UD release. Building on the proposal outlined in [20], it provides examples of ellipsis in Latin, thereby 20It should also be noted that this is a case of promotion: in the first clause Gallica is an adjective functioning as the subject, due to the ellipsis of lingua. In line with our annotation criteria, we do not reconstruct this ellipsis, since in the basic annotation the relation is not marked as orphan.
ofering material for comparison on the treatment of ellipsis across languages. Unlike the approach taken in [20], where the reconstruction focussed exclusively on predicate ellipsis and did not attempt to recover omitted arguments, thus deliberately excluding cases of nominal ellipsis, the present study has highlighted the need to consider nominal ellipsis as well, as illustrated in Section 3.2. This opens the possibility of expanding the domain of reconstruction to include a broader range of omitted elements. Accordingly, this work ofers insights into additional aspects that should be considered in the development of guidelines for ellipsis annotation within the UD framework, which remain currently underspecified.
For this work, we focussed on the description of the challenges and identified some directions for future works. First, a thorough examination of the position of the elided material shall pave the way for a study on the communicative reasons that may have guided the author in choosing an elliptical structure, such as topical focussing, among others. More research is needed to explore patterns of usage of elliptical constructions in the Germania and, more in general, in Tacitus’ oeuvre. Second, a classification of the types of ellipsis encountered in Tacitus’ Germania shall contribute to the ongoing discussion on ellipsis. The addition of other (Latin) treebanks enhanced with annotation focussing on ellipsis remains a desideratum, which will foster both research focussing on linguistic aspects of ellipsis and on stylistics. Such additions would, among other benefits, enable the training of NLP tools for ellipsis detection, thereby facilitating large-scale research into its frequency patterns and distribution across treebanks representing diferent genres and, for literary texts, across diferent authors. [6] J. Merchant, The syntax of silence: Sluicing, islands, guistics, Ljubljana, Slovenia, 2025, pp. 52–63. URL: and identity in ellipsis, University of California, https://aclanthology.org/2025.tlt-1.6/.
Santa Cruz, 1999. [21] J. Denooz, Opera Latina : une base de [7] C. Kennedy, Ellipsis and syntactic representation, données sur internet, Euphrosyne 32 (2004) in: The interfaces: Deriving and interpreting omit- 79–88. URL: https://www.brepolsonline.net/doi/10. ted structures, John Benjamins Publishing Com- 1484/J.EUPHR.5.125535. doi:10.1484/J.EUPHR. pany, 2003, pp. 29–53. 5.125535. [8] I. Ortega-Santos, M. Yoshida, C. Nakao, On ellipsis [22] K. Johnson, Gapping, in: M. Everaert, H. van Riemsstructures involving a wh-remnant and a non-wh- dijk (Eds.), The Blackwell Companion to Syntax, remnant simultaneously, Lingua (2014). volume II, Blackwell, 2006, pp. 407–435. [9] I. A. Mel’čuk, Dependency Syntax: Theory and Prac- [23] H. Pinkster, The Oxford Latin Syntax, volume 2, tice, SUNY Press, Albany, NY, 1988. Oxford University Press, Oxford, uk, 2021. [10] R. Hudson, An introduction to word grammar, Cam- [24] H. Pinkster, The Oxford Latin Syntax, volume 1, bridge University Press, 2010. Oxford University Press, Oxford, uk, 2015. [11] T. Osborne, Ellipsis, in: A Dependency Grammar [25] U. Zernial, Germania, Sammlung griechischer und of English, John Benjamins Publishing Company, lateinischer Schriftsteller mit deutschen Anmerkun2019, pp. 349–378. gen, 2. verbesserte aufl ed., Weidmannsche Buch[12] M. Marcus, B. Santorini, M. A. Marcinkiewicz, handlung, Berlin, 1897.
Building a large annotated corpus of english: The penn treebank, Computational linguistics 19 (1993) 313–330. [13] P. Osenova, K. Simov, The bulgarian hpsg treebank:
Specialization of the annotation scheme, in: Proceedings of The Second Workshop on Treebanks and Linguistic Theories; Växjö, Sweden, 2003. [14] C. Pollard, I. A. Sag, Head-driven phrase structure
grammar, University of Chicago Press, 1994. [15] J. Hajič, E. Hajičová, M. Mikulová, J. Mírovsky`,
Handbook on linguistic annotation, chapter prague dependency treebank, 2017. [16] M. Mikulová, Semantic representation of ellipsis in the prague dependency treebanks, in: Proceedings of the 26th Conference on Computational Linguistics and Speech Processing (ROCLING 2014), 2014, pp. 125–138. [17] D. Ćavar, Z. Tiganj, L. V. Mompelat, B. Dickson,
Computing ellipsis constructions: Comparing classical nlp and llm approaches, in: Proceedings of the Society for Computation in Linguistics 2024, 2024, pp. 217–226. [18] D. Ćavar, L. Mompelat, M. Abdo, The typology of ellipsis: a corpus for linguistic analysis and machine learning applications, in: Proceedings of the 6th Workshop on Research in Computational Linguistic
Typology and Multilingual NLP, 2024, pp. 46–54. [19] M. Hutton, E. H. Warmington, Agricola ; Germania ; Dialogus, Loeb classical library 62, rev. ed. ed.,
Harvard Univ. Press, Cambridge Mass, 1970. [20] C. Corbetta, F. Iurescia, M. C. Passarotti, «are you afraid of ghosts?» a proposal for busting predicate ellipsis in Universal Dependencies, in: S. Jablotschkin, S. Kübler, H. Zinsmeister (Eds.), Proceedings of the 23rd International Workshop on Treebanks and Linguistic Theories (TLT, SyntaxFest 2025), Association for Computational Lin
Example 2.1: Enhanced Tree root
conj cc nmod advcl:cmp
mark Example 3.1: Enhanced Tree
root nec dierum numerum ut nos sed noctium (numerum) computant neg day.gen.m.pl number.acc.m.sg as nom.1pl but night.gen.f.pl (number.acc.m.sg) count.3pl.prs cc conj
cc nmod advmod nsubj:pass obj conj advmod
cc conj obl:arg
cnsubj csubj mos ... conferre principibus uel armentorum uel frugum _ custom.nom.m.sg ... bestow.inf.prs chief.dat.m.pl or cattle.gen.n.pl or crop.gen.f.pl _
conj csubj cop
root faenus agitare ... ignotum ideo que _ magis seruatur quam si uetitum esset capital.acc.n.sg exploit.inf.prs ... unknown.nom.n.sg so and _ better observe.prs.pass.3sg than if prohibit.pst.pass.3sg aux advcl:cmp mark mark aux:pass obj obl:arg case csubj root
Example 6.1: Enhanced Tree root conj obl:tmod nmod
conj xcomp advmod amod Example 8.1: Enhanced Tree nmod conj advmod xcomp root
conj conj amod nsubj nsubj ccomp advmod:neg cop referunt Cotinos Gallica (coarguit) (Germanos) Osos Pannonica lingua coarguit non esse Germanos report.prs.3pl Cotini.acc.m.pl Gallic.nom.f.sg (prove.prs.3sg) (Germanacc.m.pl) Osi.acc.m.pl Pannonian.nom.f.sg tongue nom.f.sg prove.prs.3sg neg be.inf.prs German.acc.m.pl conj nsubj nsubj ccomp conj Declaration on Generative AI During the preparation of this work, the author(s) used ChatGPT (OpenAI) in order to: Paraphrase and reword and Grammar and spelling check. After using these tool(s)/service(s), the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s