We present LatinPipe, the winning submission to the EvaLatin 2024 Dependency Parsing shared task. Our system consists of a fine-tuned concatenation of base and large pre-trained LMs, with a dot-product attention head for parsing and softmax classification heads for morphology to jointly learn both dependency parsing and morphological analysis. It is trained by sampling from seven publicly available Latin corpora, utilizing additional harmonization of annotations to achieve a more unified annotation style. Before fine-tuning, we train the system for a few initial epochs with frozen weights. We also add additional local relative contextualization by stacking the BiLSTM layers on top of the Transformer(s). Finally, we ensemble output probability distributions from seven randomly instantiated networks for the final submission. The code is available at https://github.com/ufal/evalatin2024-latinpipe.

The paper proposes a novel data representation inspired by Universal Dependencies (UD) syntactic trees, which are extended to capture the internal morphological structure of word forms. As a result, morphological segmentation is incorporated within the UD representation of syntactic dependencies. To derive the proposed data structure we leverage existing annotation of UD treebanks as well as available resources for segmentation, and we select 10 languages to work with in the presented case study. Additionally, statistical analysis reveals a robust correlation between morphs and sets of morphological features of words. We thus align the morphs to the observed feature inventories capturing the morphological meaning of morphs. Through the beneficial exploitation of cross-lingual correspondence of morphs, the proposed syntactic representation based on morphological segmentation proves to enhance the comparability of sentence structures across languages.

This paper explores the possibility to exploit different Pretrained Language Models (PLMs) to assist in a manual annotation task consisting in assigning the appropriate sense to verbal predicates in a Latin text. Indeed, this represents a crucial step when annotating data according to the Uniform Meaning Representation (UMR) framework, designed to annotate the semantic content of a text in a cross-linguistic perspective. We approach the study as a Word Sense Disambiguation task, with the primary goal of assessing the feasibility of leveraging available resources for Latin to streamline the labor-intensive annotation process. Our methodology revolves around the exploitation of contextual embeddings to compute token similarity, under the assumption that predicates sharing a similar sense would also share their context of occurrence. We discuss our findings, emphasizing applicability and limitations of this approach in the context of Latin, for which the limited amount of available resources poses additional challenges.

This paper presents the harmonisation process carried out on the five treebanks available for Latin in Universal Dependencies, with the aim of eliminating the discrepancies in their annotation styles. Indeed, this is the first issue to be addressed when parsing Latin, as significant drops in parsing accuracy on different Latin treebanks have been repeatedly observed. Latin syntactic variability surely accounts for this, but parsing results are as well affected by divergent annotation choices. By analysing where annotations differ, we propose a Python-based alignment of the five UD treebanks. Consequently, the impact of annotation choices on accuracy scores is assessed by performing parsing experiments with UDPipe and Stanza.

This paper focuses on the process of harmonising the five Latin treebanks available in Universal Dependencies with respect to morphological annotation. We propose a workflow that allows to first spot inconsistencies and missing information, in order to detect to what extent the annotations differ, and then correct the retrieved bugs, with the goal of equalising the annotation of morphological features in the treebanks and producing more consistent linguistic data. Subsequently, we present some experiments carried out with UDPipe and Stanza in order to assess the impact of such harmonisation on parsing accuracy.

This paper is framed in the context of the SSHOC project and aims at exploring how Language Technologies can help in promoting and facilitating multilingualism in the Social Sciences and Humanities (SSH). Although most SSH researchers produce culturally and societally relevant work in their local languages, metadata and vocabularies used in the SSH domain to describe and index research data are currently mostly in English. We thus investigate Natural Language Processing and Machine Translation approaches in view of providing resources and tools to foster multilingual access and discovery to SSH content across different languages. As case studies, we create and deliver as freely, openly available data a set of multilingual metadata concepts and an automatically extracted multilingual Data Stewardship terminology. The two case studies allow as well to evaluate performances of state-of-the-art tools and to derive a set of recommendations as to how best apply them. Although not adapted to the specific domain, the employed tools prove to be a valid asset to translation tasks. Nonetheless, validation of results by domain experts proficient in the language is an unavoidable phase of the whole workflow.