Leveraging LLMs for Event Extraction in Italian Documents: a Roadmap for Future Research Federica Rollo* , Giovanni Bonisoli and Laura Po "Enzo Ferrari" Engineering Department, University of Modena and Reggio Emilia, MO 41121 Italy Abstract Event extraction is a task of significant interest in the field of Natural Language Processing (NLP) and plays a vital role in various applications, such as information retrieval and document summarization. Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. In this paper, we present a roadmap for the application of LLMs for event extraction from Italian documents, aiming to address the gap in research and resources for event extraction in non-English languages. We first discuss the challenges of event extraction and the current state-of-the-art approaches based on LLMs. Next, we present potential Italian datasets suitable for adapting linguistic models to the domain of event extraction. Furthermore, we outline future research directions and potential areas for improvement in this evolving field. Keywords event extraction, Large Language Model, Italian language 1. Introduction efficient access to relevant information by automatically identifying text spans containing the desired answers to The recent development of Large Language Models specific questions. While other models can be provided (LLMs) poses significant promise for advancing several with detailed instructions to extract specific data from natural language-based tasks, including event extrac- the text. Integrating these models into NLP pipelines can tion from lengthy text. LLMs such as GPT models [1] streamline the process of real-time event analysis, allow- have demonstrated remarkable capabilities in under- ing for timely and efficient extraction of event-related standing and generating natural language text. The appli- information from textual data. This paper explores the cation of LLMs for event extraction offers several advan- role of LLMs in advancing event extraction from lengthy tages. Firstly, these models can process vast amounts of text. In particular, we focus on the Italian language and text data, enabling comprehensive analysis of events de- we explore the resources available for adapting and eval- scribed in natural language. Secondly, LLMs can capture uating LLMs to event extraction on Italian documents. In complex linguistic structures and contextual nuances typ- the end, we define possible future directions for research ical of different kinds of documents, enhancing the accu- in this dynamic field. racy of extracted event details. The continuous learning ability of LLMs allows them to adapt to different writing styles and language conventions. 2. Event Extraction However, challenges persist in leveraging LLMs for event extraction in languages other than English, par- 2.1. Task formulation ticularly in languages with limited available resources Event extraction aims at identifying and categorizing such as Italian. Fine-tuning requires curated datasets events described within a text, including the recognition that accurately represent the diversity of language and of the entities involved in the event (such as individu- scenarios, and the annotation of different event-related als, organizations, or locations), and the extraction of data. temporal references and any elements that are relevant Despite these challenges, the potential of LLMs to rev- for the event. This task has gained significant popu- olutionize event extraction is substantial. For instance, larity in recent years due to its broad applicability and Question Answering (QA) models can facilitate rapid and practical utility in various real-world scenarios. Figure 1 Ital-IA 2024: 4th National Conference on Artificial Intelligence, orga- shows an example of the results of event extraction from nized by CINI, May 29-30, 2024, Naples, Italy a document describing an air crash. In addition to the * Corresponding author. identification of the event type, different event roles have $ federica.rollo@unimore.it (F. Rollo); been annotated, e.g., the date of the event occurrence, giovanni.bonisoli@unimore.it (G. Bonisoli); laura.po@unimore.it (L. Po) the aircraft agency.  0000-0002-3834-3629 (F. Rollo); 0000-0001-8538-8347 (G. Bonisoli); 0000-0002-3345-176X (L. Po) © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR ceur-ws.org Workshop ISSN 1613-0073 Proceedings Figure 1: Example of event extraction. 2.2. Challenges 2.3. Large Language Models based Due to the complexity of the natural language, event approaches extraction poses several challenges that require sophisti- Several approaches have been proposed for event extrac- cated techniques to address effectively. tion in recent surveys, from traditional methods which The first challenge consists of detecting multiple rely on the use of linguistic rules for pattern identifica- events described in the same document and understand- tion within the text to more advanced solutions such as ing which are the references to each event. Natural lan- machine learning and deep learning algorithms able to guage often contains ambiguous expressions that can learn patterns after training on annotated data, and the refer to multiple events or entities. This ambiguity, along use of pre-trained language models [2, 3]. LLMs based ap- with the use of coreference, further complicates the task proaches have emerged as a promising avenue for event of accurately extracting event data from text since resolv- extraction in recent years. These models leverage the ing ambiguity requires contextual understanding and power of machine learning and deep learning algorithms disambiguation techniques. as they are pre-trained on vast amounts of text data and Identifying relevant elements for each event requires then fine-tuned for specific tasks. By encoding contextual distinguishing between event triggers (words or phrases information and capturing semantic relationships within that indicate the occurrence of an event) and background the text, LLMs seem to be promising in identifying and information and noise. Another complexity is given by extracting events from various sources. the variability in language usage, writing styles, syntactic We identified three main approaches based on the use structures, and document length. Indeed, event extrac- of LLMs that could reach good performance in event ex- tion can be performed on short text like tweets, longer traction: sequence labeling models, extractive Question documents such as news articles, and lengthy documents Answering (QA) models and instruction-tuned models. such as investigative reports or government documents. All these factors require the use of techniques able of Sequence Labeling models In Sequence labeling accommodating these variations to achieve accurate and each token in a sequence is assigned a label based on reliable results across diverse text types and genres. its role or category within the context of the sequence. Two of the key aspects of events are the time and the Sequence labeling models can be used to identify those space, i.e., when the event took place and where. The text spans reporting relevant information within a text. recognition and standardization of temporal and spatial Therefore, it is widely employed for several classical NLP expressions could be complex since temporal reference tasks like part-of-speech (POS) tagging, named entity can be expressed in various formats (such as dates, times, recognition (NER), text chunking. part of the day). In addition, a document describing an Sequence labeling models are suitable for the scenario event can refer to the location of the event providing of event extraction, where they can identify and classify information at different granularity, for example indicat- those parts of text reporting information about events. ing the name of the city, specifying the address, and/or Indeed, some works in literature have already treated describing the type of the place like an apartment, a shop, event extraction as a sequence labeling or NER problem, or a park. During event extraction, the references to all [4, 5], also for Italian Language [6]. these locations should be identified. Extractive Question Answering The goal of extrac- next-word prediction objective of LLMs and the users’ tive QA models is to understand an input question in objective of following their instructions helpfully and natural language and extract the answer as a span from safely. Instruction-tuning involves a fine-tuning of Auto- an input text. QA models can facilitate rapid and effi- Regressive LLMs with input-output pairs, where input cient access to event-related information by automati- denotes the human instructions, and output denotes the cally identifying text spans containing the desired an- desired output that follows the instruction. The results swers to specific questions. For instance, the question of this process are the Instruction-Tuned LLMs, designed “When did the event take place?” (Q1) can be formulated specifically to provide appropriate results based on in- to retrieve the date of the event. struction inputs. This ability is also enhanced as a cross- The results of these models depend significantly on the task generalization, leading Instruction-Tuned LLMs to quality of the input documents, as well as the structure better performances on novel tasks. of the questions provided to the models. Prior knowledge Instruction-Tuned LLMs can be employed to solve a about the kind of event described in the document allows wide range of NLP tasks through various techniques of to formulate ad hoc questions. For instance, considering prompt engineering [13], i.e., the process of designing the document in Figure 1, the question “When did the air task-specific instructions to guide model output. There- crash take place?” (Q2) should provide more accurate an- fore, the utilization of these models can also yield benefits swers than Q1. In addition, questions should be enriched for event extraction. by other details about the event after a partial process of Currently, there are several Instruction-Tuned LLMs event extraction. For example, the question “When did capable of understanding and generating text. For those, the Flight 345 crash?” (Q3) contain the reference to the Italian represents a minority percentage in the training flight number and should help the QA models to select data compared to more widely used languages on the the correct context for the extraction of the date. web such as English. Among these, there are proprietary Within the QA models, distinctions arise between models like GPT-3.5 and GPT-4 from OpenAI, Gemini Single-Span QA (SQA) and Multi-Span QA (MQA). While from Google, and open-source families of LLMs like Mis- the former identifies a single text segment for each ques- tral [14] and Mixtral [15] from Mistral AI and Llama tion, the latter locates answers even when distributed [11] and Llama 2 [16] from Meta. From this last family, across non-consecutive text segments, potentially located Llamantino [17] has been derived through a language far apart within a document. Given the prevalence of adaptation process to the Italian Language. such scenarios, especially in complex inquiries and de- tailed documents, the limitations of SQA models are ev- ident. An example is the annotation of “causalities and 3. Italian datasets losses” in Figure 1. The recent surge in MQA model Currently, there are few Italian datasets suitable for event development [7, 8, 9] underscores a notable interest. extraction. Some of them provide a comprehensive an- In the current state-of-the-art, the only Italian dataset notation of event-related data, while in other cases, only properly designed for training QA models is SQuAD-it one type of information (e.g., the temporal references) is [10], derived from the automatic translation of the En- annotated. glish SQuAD dataset, consisting of a list of pairs question- answer. However, this dataset can be used only for SQA, therefore it is unsuitable for complex tasks like event 3.1. EVENTI extraction which requires the ability to retrieve multiple The EVENTI1 corpus was built in 2014 for the evalua- spans for one question. tion of Temporal Information Processing systems of the EVENTI evaluation exercise [18] in the EVALITA work- Instruction-Tuned models Among LLMs, Auto- shop. The corpus consists of three datasets: the Main Regressive models such as GPT [1] or Llama [11] series task training data (274 documents) and test data (92 doc- stand out. These models leverage advanced deep learn- uments) of contemporary news articles and the Pilot task ing techniques to predict the subsequent word based on (10 documents) test data of historical news articles. The an input text. This prediction process is repeated mul- annotation guidelines involve the use of four tags to an- tiple times, with each predicted word being added to notate different elements within news texts: the EVENT the original text. By training on vast amounts of text tag is used to annotate all the mentions of events includ- data, Auto-Regressive LLMs effectively capture complex ing verbs, nouns, prepositional phrases and adjectives; patterns and structures in language, leading them to gen- the TIMEX3 tag is used for temporal expressions; the erate full and coherent text which is contextually relevant SIGNAL tag identifies textual items which encode a re- to input text. lation either between EVENTs, or TIMEX3s or both; the The research in recent years has led to the development of instruction tuning [12] to bridge the gap between the 1 https://sites.google.com/site/eventievalita2014/data-tools TLINK tag is used for temporal dependencies between [24, 25]. The news articles underwent automated NLP EVENTs and/or Temporal Expressions. processes to extract temporal references, entities, and cor- responding DBpedia resources. Duplicates are annotated 3.2. NewsReader MEANTIME to identify news articles referring to the same crime event. The theft-related news articles are annotated manually The NewsReader MEANTIME (Multilingual Event ANd following a sophisticated annotation schema to identify TIME) is a multilingual semantically annotated corpus of stolen items (What), crime locations (Where), references 480 Wikinews articles in four languages: English, Italian, to authors and victims, and their sociodemographic char- Spanish, and Dutch [19]. The corpus was released in 2016 acteristics (Who). The annotation provided in the dataset and derives from the NewsReader Project2 [20] which is multi-span since it involves identifying and linking aims at extracting information about what happened to multiple text spans within the document. whom, when, and where, processing a large volume of financial and economic data. The corpus is enriched with 3.5. EventNet-ITA annotations that span multiple levels, including entities, entity mentions, events, temporal information, semantic EventNet-ITA4 [26] is an Italian corpus for Frame Parsing roles, and intra-document and cross-document event and applied to events released in 2024. Semantic Frame Pars- entity coreference. ing is a task which aims at identifying semantic frames within textual data. A semantic frame [27] is a cognitive 3.3. De Gasperi structure that organizes and represents knowledge about a concept or situation. It consists of a set of intercon- The De Gasperi corpus [21] is a collection of historical nected elements such as roles, attributes, and relations, documents by Alcide De Gasperi, the first Prime Minister which collectively define the meaning and typical fea- of the Italian Republic. The corpus was released in 2019 tures of that concept or situation. Frames help humans and includes 2,762 documents published between 1901 understand and interpret language by providing a mental and 1954, originally released in an oral or written form. In framework for comprehending and categorizing informa- addition to the raw text, a set of meta-data and additional tion. semiautomatically annotated information are available. EventNet-ITA is built upon the idea of enabling frame The corpus contains different kinds of documents, like parsing for event extraction. It is composed of 53,854 daily press written by De Gasperi when he worked as a sentences manually annotated with 205 semantic frames journalist for newspapers in Trentino, and speeches in of events and covers different domains, like conflictual, institutional venues when he was a Member of the Italian social, communication, legal, geopolitical, economic and Parliament. In each document, references to persons and biographical events. places are annotated. 3.4. DICE 4. Future directions DICE3 [22] is a collection of 10,395 Italian news articles Automated information extraction from documents con- describing crime events that happened in the Modena tinues to captivate the scientific community due to its province between 2011 and 2021. The news articles are manifold advantages, facilitating improved information extracted from one of the most popular local newspapers, accessibility across various domains. By leveraging LLMs “Gazzetta di Modena”, following the approach described and exploiting annotated datasets, researchers can de- in [23]. Thanks to an agreement between the University velop robust event extraction systems capable of achiev- of Modena and Reggio Emilia and the Gazzetta di Mod- ing high accuracy and efficiency across a wide range of ena, DICE was released online in 2023, free to redistribute text sources. As the field continues to advance, further and transform without encountering legal copyright is- research into LLMs and their applications in event ex- sues under an Attribution-NonCommercial-ShareAlike traction is expected to drive continued innovation and 4.0 International (CC BY-NC-SA 4.0). progress in this area. Along with the data related to the title, the text, and Future directions will focus on three key aspects: the publication date of each news article that are crawled • Definition of an Italian benchmark: while from the newspaper’s webpage, several annotations are we have identified five Italian datasets suitable available on the data. The crime event category (e.g., for event extraction, further efforts are needed to theft, robbery) is assigned to each news article using text expand their annotation and support comprehen- categorization approaches based on word embeddings sive event extraction tasks. This entails defining 2 http://www.newsreader-project.eu/ 3 4 https://github.com/federicarollo/Italian-Crime-News https://huggingface.co/datasets/mrovera/eventnet-ita a standardized benchmark for evaluating event [2] G. Frisoni, G. Moro, A. Carbonaro, A survey on extraction systems. Such a benchmark would event extraction for natural language understand- serve as a common evaluation dataset, enabling ing: Riding the biomedical literature wave, IEEE comparisons between different approaches and Access 9 (2021) 160721 – 160757. doi:10.1109/ fostering the development of more accurate and ACCESS.2021.3130956. reliable event extraction models. [3] W. Xiang, B. Wang, A survey of event extraction • Evaluation of LLMs on the benchmark: de- from text, IEEE Access 7 (2019) 173111–173137. spite the limited literature on Italian event extrac-doi:10.1109/ACCESS.2019.2956831. tion, our preliminary evaluation of three BERT- [4] A. Ramponi, R. van der Goot, R. Lombardo, B. Plank, based QA models on the DICE dataset revealed Biomedical event extraction as sequence labeling, promising results [22]. However, challenges per- in: B. Webber, T. Cohn, Y. He, Y. Liu (Eds.), Pro- sist, particularly related to the size and quality ceedings of the 2020 Conference on Empirical of the annotated data. Once the benchmark is Methods in Natural Language Processing (EMNLP), defined, future efforts will focus on evaluating Association for Computational Linguistics, On- and comparing various approaches outlined in line, 2020, pp. 5357–5367. doi:10.18653/v1/2020. Section 2.3. The evaluation will include the recent emnlp-main.431. Minerva models that represent the first family of[5] S. Pongpaichet, B. Sukosit, C. Duangtanawat, LLMs trained from scratch on Italian documents J. Jamjongdamrongkit, C. Mahacharoensuk, developed by Sapienza NLP. K. Matangkarat, P. Singhajan, T. Noraset, S. Tuarob, • Creation of a synthetic annotated dataset: Camelon: A system for crime metadata extraction since manual annotation is a time and resource- and spatiotemporal visualization from online consuming process, new strategies will be studied news articles, IEEE Access 12 (2024) 22778–22802. to automate the process of annotation. Employ- doi:10.1109/ACCESS.2024.3363879. ing LLMs for data augmentation (i.e., to expand [6] N. Viani, T. A. Miller, D. Dligach, S. Bethard, the annotated dataset) is now the most promising C. Napolitano, S. G. Priori, R. Bellazzi, L. Sacchi, approach, especially focusing on text generation G. K. Savova, Recurrent neural network architec- models. Given a list of desired annotations, i.e., tures for event extraction from italian medical re- the spans to extract from the text (like “May 14th” ports, in: A. ten Teije, C. Popow, J. H. Holmes, as the date of the event), the LLM is asked to cre- L. Sacchi (Eds.), Artificial Intelligence in Medicine, ate a document with that span with the expected Springer International Publishing, Cham, 2017, pp. role in the event described (like “create a docu- 198–202. ment describing an event that occurred on May [7] H. Li, M. Tomko, M. Vasardani, T. Baldwin, Mul- 14th”). This methodology allows for obtaining a tispanqa: A dataset for multi-span question an- synthetic dataset that is also already annotated. swering, in: M. Carpuat, M. de Marneffe, I. V. M. Furthermore, this approach offers control over Ruíz (Eds.), Proceedings of the 2022 Conference text generation and ensures fairness in dataset of the North American Chapter of the Associa- composition, ultimately contributing to the devel- tion for Computational Linguistics: Human Lan- opment of balanced and unbiased datasets essen- guage Technologies, NAACL 2022, Seattle, WA, tial for training accurate and equitable AI models. United States, July 10-15, 2022, Association for Computational Linguistics, 2022, pp. 1250–1260. doi:10.18653/V1/2022.NAACL-MAIN.90. References [8] E. Segal, A. Efrat, M. Shoham, A. Globerson, J. Be- rant, A simple and effective model for answering [1] T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. multi-span questions, 2020, p. 3074 – 3080. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, [9] M. Zhu, A. Ahuja, D. Juan, W. Wei, C. K. Reddy, G. Sastry, A. Askell, S. Agarwal, A. Herbert-Voss, Question answering with long multiple-span an- G. Krueger, T. Henighan, R. Child, A. Ramesh, swers, in: T. Cohn, Y. He, Y. Liu (Eds.), D. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, Findings of the Association for Computational E. Sigler, M. Litwin, S. Gray, B. Chess, J. Clark, Linguistics: EMNLP 2020, Online Event, 16-20 C. Berner, S. McCandlish, A. Radford, I. Sutskever, November 2020, volume EMNLP 2020 of Findings D. Amodei, Language models are few-shot learners, of ACL, Association for Computational Linguis- in: H. Larochelle, M. Ranzato, R. Hadsell, M. Bal- tics, 2020, pp. 3840–3849. doi:10.18653/V1/2020. can, H. Lin (Eds.), Advances in Neural Information FINDINGS-EMNLP.342. Processing Systems, volume 33, Curran Associates, [10] D. Croce, A. Zelenanska, R. Basili, Neural learning Inc., 2020, pp. 1877–1901. for question answering in italian, Lecture Notes in Computer Science (including subseries Lecture clicit201425. Notes in Artificial Intelligence and Lecture Notes [19] A.-L. Minard, M. Speranza, R. Urizar, B. Altuna, in Bioinformatics) 11298 LNAI (2018) 389 – 402. M. van Erp, A. Schoen, C. van Son, MEANTIME, the doi:10.1007/978-3-030-03840-3_29. NewsReader multilingual event and time corpus, in: [11] H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Proceedings of the Tenth International Conference Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Ham- on Language Resources and Evaluation (LREC’16), bro, F. Azhar, A. Rodriguez, A. Joulin, E. Grave, European Language Resources Association (ELRA), G. Lample, Llama: Open and efficient foundation Portorož, Slovenia, 2016, pp. 4417–4422. language models, 2023. arXiv:2302.13971. [20] P. Vossen, R. Agerri, I. Aldabe, A. Cybulska, M. van [12] S. Zhang, L. Dong, X. Li, S. Zhang, X. Sun, S. Wang, Erp, A. Fokkens, E. Laparra, A.-L. Minard, A. P. J. Li, R. Hu, T. Zhang, F. Wu, G. Wang, Instruction Aprosio, G. Rigau, et al., Newsreader: Using knowl- tuning for large language models: A survey, 2024. edge resources in a cross-lingual reading machine arXiv:2308.10792. to generate more knowledge from massive streams [13] P. Sahoo, A. K. Singh, S. Saha, V. Jain, S. Mondal, of news, Knowledge-Based Systems 110 (2016) 60– A. Chadha, A systematic survey of prompt engi- 85. neering in large language models: Techniques and [21] S. Tonelli, R. Sprugnoli, G. Moretti, Prendo la parola applications, 2024. arXiv:2402.07927. in questo consesso mondiale: A multi-genre 20th [14] A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bam- century corpus in the political domain, volume ford, D. S. Chaplot, D. de las Casas, F. Bressand, 2481, 2019. G. Lengyel, G. Lample, L. Saulnier, L. R. Lavaud, [22] G. Bonisoli, M. P. di Buono, L. Po, F. Rollo, DICE: M.-A. Lachaux, P. Stock, T. L. Scao, T. Lavril, a dataset of italian crime event news, in: H. Chen, T. Wang, T. Lacroix, W. E. Sayed, Mistral 7b, 2023. W. E. Duh, H. Huang, M. P. Kato, J. Mothe, B. Poblete arXiv:2310.06825. (Eds.), Proceedings of the 46th International ACM [15] A. Q. Jiang, A. Sablayrolles, A. Roux, A. Mensch, SIGIR Conference on Research and Development in B. Savary, C. Bamford, D. S. Chaplot, D. de las Information Retrieval, SIGIR 2023, Taipei, Taiwan, Casas, E. B. Hanna, F. Bressand, G. Lengyel, July 23-27, 2023, ACM, 2023, pp. 2985–2995. doi:10. G. Bour, G. Lample, L. R. Lavaud, L. Saulnier, M.-A. 1145/3539618.3591904. Lachaux, P. Stock, S. Subramanian, S. Yang, S. An- [23] F. Rollo, L. Po, Crime event localization and dedu- toniak, T. L. Scao, T. Gervet, T. Lavril, T. Wang, plication, in: J. Z. Pan, V. Tamma, C. d’Amato, T. Lacroix, W. E. Sayed, Mixtral of experts, 2024. K. Janowicz, B. Fu, A. Polleres, O. Seneviratne, L. Ka- arXiv:2401.04088. gal (Eds.), The Semantic Web – ISWC 2020, Springer [16] H. Touvron, L. Martin, K. Stone, P. Albert, A. Alma- International Publishing, Cham, 2020, pp. 361–377. hairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhar- [24] F. Rollo, G. Bonisoli, L. Po, A comparative analysis gava, S. Bhosale, D. Bikel, L. Blecher, C. C. Ferrer, of word embeddings techniques for italian news M. Chen, G. Cucurull, D. Esiobu, J. Fernandes, J. Fu, categorization, IEEE Access 12 (2024) 25536 – 25552. W. Fu, B. Fuller, C. Gao, V. Goswami, N. Goyal, doi:10.1109/ACCESS.2024.3367246. A. Hartshorn, S. Hosseini, R. Hou, H. Inan, M. Kar- [25] F. Rollo, G. Bonisoli, L. Po, Supervised and unsu- das, V. Kerkez, M. Khabsa, I. Kloumann, A. Ko- pervised categorization of an imbalanced italian renev, P. S. Koura, M.-A. Lachaux, T. Lavril, J. Lee, crime news dataset, Lecture Notes in Business In- D. Liskovich, Y. Lu, Y. Mao, X. Martinet, T. Mihaylov, formation Processing 442 LNBIP (2022) 117 – 139. P. Mishra, I. Molybog, Y. Nie, A. Poulton, J. Reizen- doi:10.1007/978-3-030-98997-2_6. stein, R. Rungta, K. Saladi, A. Schelten, R. Silva, E. M. [26] M. Rovera, EventNet-ITA: Italian frame parsing Smith, R. Subramanian, X. E. Tan, B. Tang, R. Tay- for events, in: Y. Bizzoni, S. Degaetano-Ortlieb, lor, A. Williams, J. X. Kuan, P. Xu, Z. Yan, I. Zarov, A. Kazantseva, S. Szpakowicz (Eds.), Proceedings of Y. Zhang, A. Fan, M. Kambadur, S. Narang, A. Ro- the 8th Joint SIGHUM Workshop on Computational driguez, R. Stojnic, S. Edunov, T. Scialom, Llama 2: Linguistics for Cultural Heritage, Social Sciences, Open foundation and fine-tuned chat models, 2023. Humanities and Literature (LaTeCH-CLfL 2024), As- arXiv:2307.09288. sociation for Computational Linguistics, St. Julians, [17] P. Basile, E. Musacchio, M. Polignano, L. Siciliani, Malta, 2024, pp. 77–90. G. Fiameni, G. Semeraro, Llamantino: Llama 2 mod- [27] C. J. Fillmore, C. F. Baker, Frame semantics for els for effective text generation in italian language, text understanding, in: Proceedings of WordNet 2023. arXiv:2312.09993. and Other Lexical Resources Workshop, NAACL, [18] T. Caselli, R. Sprugnoli, M. Speranza, M. Mona- volume 6, 2001. chini, Eventi evaluation of events and temporal information at evalita 2014, 2014. doi:10.12871/