Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 15
Encontrada(s) 1 página(s)

RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Revisiting Supertagging for faster HPSG parsing

  • Zamaraeva, Olga
  • Gómez-Rodríguez, Carlos
Presented at: Conference on Empirical Methods in Natural Language Processing, Miami, Florida, USA,12-16 Nov. 2024., [Abstract]: We present new supertaggers trained on English HPSG-based treebanks and test the effects of the best tagger on parsing speed and accuracy. HPSG treebanks are produced automatically by large manually built grammars and feature high-quality annotation based on a well-developed linguistic theory. The English Resource Grammar treebanks include diverse and challenging test datasets, beyond the usual WSJ section 23 and Wikipedia data. HPSG supertagging has previously relied on MaxEnt-based models. We use SVM and neural CRF- and BERT-based methods and show that both SVM and neural supertaggers achieve considerably higher accuracy compared to the baseline and lead to an increase not only in the parsing speed but also the parser accuracy with respect to gold dependency structures. Our fine-tuned BERT-based tagger achieves 97.26% accuracy on 950 sentences from WSJ23 and 93.88% on the out-of-domain technical essay The Cathedral and the Bazaar. We present experiments with integrating the best supertagger into an HPSG parser and observe a speedup of a factor of 3 with respect to the system which uses no tagging at all, as well as large recall gains and an overall precision gain. We also compare our system to an existing integrated tagger and show that although the well-integrated tagger remains the fastest, our experimental system can be more accurate. Finally, we hope that the diverse and difficult datasets we used for evaluation will gain more popularity in the field: we show that results can differ depending on the dataset, even if it is an in-domain one. We contribute the complete datasets reformatted for Huggingface token classification., We acknowledge the European Union’s Horizon Europe Framework Programme which funded this research under the Marie Skłodowska-Curie postdoctoral fellowship grant HORIZON-MSCA-2021-PF-01 (GAUSS, grant agreement No 101063104); and the European Research Council (ERC), which has funded this research under the Horizon Europe research and innovation programme (SALSA, grant agreement No 101100615). We also acknowledge grants SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; and TSI-100925-2023-1 funded by Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR; as well as funding by Xunta de Galicia (ED431C 2024/02), and Centro de Investigación de Galicia “CITIC”, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS)., Xunta de Galicia; ED431C 2024/02
RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

Document-level event extraction from Italian crime news using minimal data

  • Bonisoli, Giovanni
  • Vilares, David
  • Rollo, Federica
  • Po, Laura
Financiado para publicación en acceso aberto: Universidade da Coruña/CISUG, [Abstract]: Event extraction from unstructured text is a critical task in natural language processing, often requiring substantial annotated data. This study presents an approach to document-level event extraction applied to Italian crime news, utilizing large language models (LLMs) with minimal labeled data. Our method leverages zero-shot prompting and in-context learning to effectively extract relevant event information. We address three key challenges: (1) identifying text spans corresponding to event entities, (2) associating related spans dispersed throughout the text with the same entity, and (3) formatting the extracted data into a structured JSON. The findings are promising: LLMs achieve an F1-score of approximately 60% for detecting event-related text spans, demonstrating their potential even in resource-constrained settings. This work represents a significant advancement in utilizing LLMs for tasks traditionally dependent on extensive data, showing that meaningful results are achievable with minimal data annotation. Additionally, the proposed approach outperforms several baselines, confirming its robustness and adaptability to various event extraction scenarios, This work has received support by Grant GAP (PID2022-139308OA-I00) funded by MCIN/AEI/10.13039/501100011033/ and by ERDF, EU; by Grant SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033/; by Xunta de Galicia (ED431C 2024/02); and by Centro de Investigación de Galicia “CITIC”, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centers of the Galician University System (CIGUS); by Mobility of higher education students and staff supported by internal policy funds (2022-1-IT02-KA131-HED-000064316) funded by EU through the Erasmus+ Programme and by UNIMORE; by Ministry for Digital Transformation and Civil Service and ‘Next-GenerationEU’/PRTR under Grant TSI-100925-2023-1., Xunta de Galicia; ED431C 2024/02
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Nested Named Entity Recognition as Single-Pass Sequence Labeling

  • Muñoz-Ortiz, Alberto
  • Vilares, David
  • Corro, Caio
  • Gómez-Rodríguez, Carlos
Presented at: Conference on Empirical Methods in Natural Language Processing (EMNLP 2025), Suzhou, China, November 4th to November 9th, 2025., [Abstract]: We cast nested named entity recognition (NNER) as a sequence labeling task by leveraging prior work that linearizes constituency structures, effectively reducing the complexity of this structured prediction problem to straightforward token classification. By combining these constituency linearizations with pretrained encoders, our method captures nested entities while performing exactly n tagging actions. Our approach achieves competitive performance compared to less efficient systems, and it can be trained using any off-the-shelf sequence labeling library., Xunta de Galicia; ED431C 2024/02, Xunta de Galicia; ED431C 2024/02, Xunta de Galicia; ICTS-2019-02-CESGA-3, Xunta de Galicia; CESG15-DE-3114, France. Agence nationale de la recherche; ANR-23-IAS1-0004, France. Agence nationale de la recherche; ANR-23-CE23-0005
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Parsing the Switch: LLM-Based UD Annotation for Complex Code-Switched and Low-Resource Languages

  • Kellert, Olga
  • Tyagi, Nemika
  • Imran, Muhammad
  • Licona-Guevara, Nelvin
  • Gómez-Rodríguez, Carlos
Presented at: Conference on Empirical Methods in Natural Language Processing (EMNLP 2025), Suzhou, China, November 4th to November 9th, 2025., [Abstract]: Code-switching presents a complex challenge for syntactic analysis, especially in low-resource language settings where annotated data is scarce. While recent work has explored the use of large language models (LLMs) for sequence-level tagging, few approaches systematically investigate how well these models capture syntactic structure in code-switched contexts. Moreover, existing parsers trained on monolingual treebanks often fail to generalize to multilingual and mixed-language input. To address this gap, we introduce the BiLingua Pipeline, an LLM-based annotation pipeline designed to produce Universal Dependencies (UD) annotations for code-switched text. First, we develop a prompt-based framework for Spanish-English and Spanish-Guaraní data, combining few-shot LLM prompting with expert review. Second, we release two annotated datasets, including the first Spanish-Guaraní UD-parsed corpus. Third, we conduct a detailed syntactic analysis of switch points across language pairs and communicative contexts. Experimental results show that BiLingua Pipeline achieves up to 95.29% LAS after expert revision, significantly outperforming prior baselines and multilingual parsers. These results show that LLMs, when carefully guided, can serve as practical tools for bootstrapping syntactic resources in under-resourced, code-switched environments., We thank the anonymous annotators and reviewers for their constructive suggestions and help. We ex- tend our gratitude to the Research Computing (RC) and Enterprise Technology at ASU for providing computing resources and access to the ChatGPT en- terprise version for experiments. We acknowledge grants GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU. CITIC, as a center accredited for excellence within the Galician University System and a member of the CIGUS Network, receives subsidies from the Department of Education, Sci- ence, Universities, and Vocational Training of the Xunta de Galicia. Additionally, it is co-financed by the EU through the FEDER Galicia 2021-27 operational program (Ref. ED431G 2023/01). Furthermore, this research was supported by the International, Interdisciplinary and Intersectoral Information and Communications Technology PhD programme (3-i ICT) granted to CITIC and supported by the European Union through the Horizon 2020 research and innovation programme under a Marie Skłodowska-Curie agreement (H2020-MSCA-COFUND), GA 101034261., Xunta de Galicia; ED431G 2023/01
RUC. Repositorio da Universidade da Coruña
Publicación Proyecto fin de carrera (BachelorThesis)

Dinámicas de entrenamiento para la caracterización de datos sintéticos en NLP

  • Lago Pato, Irene
[Resumen] La creciente utilización de los Large Language Models (LLMs) para la generación de datos sintéticos ha abierto nuevas posibilidades para reducir la dependencia de la anotación manual en tareas de Procesamiento del Lenguaje Natural (NLP). Sin embargo, su utilidad depende de su capacidad para representar la diversidad y complejidad presentes en datos humanos, lo que hace necesario desarrollar metodologías que permitan caracterizarlos y evaluarlos. En este trabajo se propone un marco de análisis basado en las dinámicas de entrenamiento de modelos supervisados para estudiar las propiedades de los datos sintéticos. Para ello, se generan conjuntos de datos correspondientes a distintas tareas de NLP mediante diferentes modelos de lenguaje y estrategias de generación, comparando su comportamiento con datos de referencia elaborados por humanos. Además, se estudia la robustez del marco propuesto utilizando distintos modelos para extraer las dinámicas de entrenamiento, con el objetivo de comprobar si la caracterización obtenida se mantiene estable independientemente del modelo empleado. Finalmente, se analiza cómo diferentes estrategias de selección de datos basadas en estas dinámicas afectan al entrenamiento de modelos supervisados. Los resultados muestran que este marco permite identificar diferencias relevantes entre datos sintéticos y humanos, así como analizar la influencia que ejercen la familia y el tamaño del LLM sobre las características de los datos generados. De este modo, este Trabajo de Fin de Grado busca contribuir al estudio de la calidad de los datos generados por LLMs en tareas críticas de NLP, evaluando hasta qué punto dichos datos pueden complementar o incluso sustituir el esfuerzo humano y definiendo una metodología para evaluar estos corpus e identificar las muestras más valiosas para el entrenamiento de modelos., [Abstract] The increasing use of Large Language Models (LLMs) for synthetic data generation has opened new opportunities to reduce the reliance on manual annotation in Natural Language Processing (NLP) tasks. However, their usefulness depends on their ability to capture the diversity and complexity of human-generated data, making it necessary to develop methodologies for their characterization and evaluation. This work proposes an analysis framework based on the training dynamics of supervised models to study the properties of synthetic data. To this end, datasets corresponding to different NLP tasks are generated using various language models and generation strategies, and their behavior is compared with that of human-created reference datasets. Furthermore, the robustness of the proposed framework is evaluated by employing different models to extract training dynamics, with the aim of determining whether the resulting characterization remains stable regardless of the underlying model. Finally, the impact of different data selection strategies based on these dynamics on the training of supervised models is analyzed. The results show that this framework can effectively identify relevant differences between synthetic and human-generated data, while also revealing the influence of LLM family and model size on the characteristics of the generated data. Thus, this Final Degree Project aims to contribute to the study of the quality of data generated by LLMs in critical NLP tasks, by analyzing the extent to which such data can complement or even replace human effort, and by defining a methodology to evaluate these corpora and identify the most valuable samples for model training., Este trabajo se ha desarrollado en el marco del proyecto (GAP; PID2022-139308OA-I00) financiado por MICIU/AEI/10.13039/501100011033 y por FEDER, UE. Asimismo, este proyecto de investigación ha sido posible gracias al acceso concedido por el Centro de Supercomputación de Galicia (CESGA) a su infraestructura de supercomputación. El supercomputador FinisTerrae III y su sistema de almacenamiento permanente de datos han sido financiados por el Plan de Recuperación, Transformación y Resiliencia 2021 de NextGenerationEU, ICT2021-006904, así como por el Programa Operativo Plurirregional de España 2014-2020 del Fondo Europeo de Desarrollo Regional (FEDER), ICTS-2019-02-CESGA-3, y por el Programa Estatal de Fomento de la Investigación Científica y Técnica de Excelencia del Plan Estatal de Investigación Científica y Técnica y de Innovación 2013-2016 (Subprograma Estatal de Infraestructuras Científicas y Técnicas y Equipamiento) del FEDER, CESG15-DE-3114., Xunta de Galicia; ICTS-2019-02-CESGA-3, Xunta de Galicia; CESG15-DE-3114, Traballo fin de grao (UDC.FIC). Intelixencia Artificial. Curso 2025/2026
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Hierarchical Bracketing Encodings for Dependency Parsing as Tagging

  • Ezquerro, Ana
  • Vilares, David
  • Yli-Jyrä, Anssi
  • Gómez-Rodríguez, Carlos
Presented at the 63rd Annual Meeting of the Association for Computational Linguistics, July 2025, Vienna, Austria., [Abstract]: We present a family of encodings for sequence labeling dependency parsing, based on the concept of hierarchical bracketing. We show that the existing 4-bit projective encoding belongs to this family, but it is suboptimal in the number of labels used to encode a tree. We derive an optimal hierarchical bracketing, which minimizes the number of symbols used and encodes projective trees using only 12 distinct labels (vs. 16 for the 4-bit encoding). We also extend optimal hierarchical bracketing to support arbitrary non-projectivity in a more compact way than previous encodings. Our new encodings yield competitive accuracy on a diverse set of treebanks., We acknowledge grants SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; and TSI-100925-2023-1 funded by Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR; as well as funding by Xunta de Galicia (ED431C 2024/02), and CITIC, as a center accredited for excellence within the Galician University System and a member of the CIGUS Network, receives subsidies from the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia. Additionally, it is co-financed by the EU through the FEDER Galicia 2021-27 operational program (Ref. ED431G 2023/01). We also extend our gratitude to CESGA, the supercomputing center of Galicia, for granting us access to its resources. Furthermore, we acknowledge the Faculty of Agriculture and Forestry of the University of Helsinki, as well as projects "Theory of Computational Logics" (352420) and "XAILOG" (345612, 345633) funded by the Research Council of Finland for the continued support of the third author during the multistage writing process., Xunta de Galicia; ED431C 2024/02, Xunta de Galicia; ED431G 2023/01, Finland. Research Council of Finland; 352420, Finland. Research Council of Finland; 345612, Finland. Research Council of Finland; 345633
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Bringing Emerging Architectures to Sequence Labeling in NLP

  • Ezquerro, Ana
  • Gómez-Rodríguez, Carlos
  • Vilares, David
Presented at: 19th Conference of the European Chapter of the Association for Computational Linguistics, March 24–29, 2026, Rabat, Morocco, [Abstract]: Pretrained Transformer encoders are the dominant approach to sequence labeling. While some alternative architectures-such as xLSTMs, structured state-space models, diffusion models, and adversarial learning-have shown promise in language modeling, few have been applied to sequence labeling, and mostly on flat or simplified tasks. We study how these architectures adapt across tagging tasks that vary in structural complexity, label space, and token dependencies, with evaluation spanning multiple languages. We find that the strong performance previously observed in simpler settings does not always generalize well across languages or datasets, nor does it extend to more complex structured tasks., We acknowledge grants GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; and TSI-100925-2023-1 funded by Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR; as well as funding by Xunta de Galicia (ED431C 2024/02), and CITIC, as a center accredited for excellence within the Galician University System and a member of the CIGUS Network, receives subsidies from the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia. Additionally, it is co-financed by the EU through the FEDER Galicia 2021-27 operational program (Ref. ED431G 2023/01). This research project was made possible through the access granted by the Galician Supercomputing Center (CESGA) to its supercomputing infrastructure. The supercomputer FinisTerrae III and its permanent data storage system have been funded by the NextGeneration EU 2021 Recovery, Transformation and Resilience Plan, ICT2021-006904, and also from the Pluriregional Operational Programme of Spain 2014-2020 of the European Regional Development Fund (ERDF), ICTS-2019-02-CESGA-3, and from the State Programme for the Promotion of Scientific and Technical Research of Excellence of the State Plan for Scientific and Technical Research and Innovation 2013-2016 State subprogramme for scientific and technical infrastructures and equipment of ERDF, CESG15-DE-3114., Xunta de Galicia; ED431C 2024/02, Xunta de Galicia; ED431G 2023/01, Xunta de Galicia; ICTS-2019-02-CESGA-3, Xunta de Galicia; CESG15-DE-3114
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Better Benchmarking LLMs for Zero-Shot Dependency Parsing

  • Ezquerro, Ana
  • Gómez-Rodríguez, Carlos
  • Vilares, David
Presented at: Joint 25th Nordic Conference on Computational Linguistics and 11th Baltic Conference on Human Language Technologies (NoDaLiDa/Baltic-HLT 2025), pages 121–135, March 3-4, 2025 ©2025 University of Tartu Library, Código asociado: https://github.com/anaezquerro/naipar, [Abstract]: While LLMs excel in zero-shot tasks, their performance in linguistic challenges like syntactic parsing has been less scrutinized. This paper studies state-of-the-art openweight LLMs on the task by comparing them to baselines that do not have access to the input sentence, including baselines that have not been used in this context such as random projective trees or optimal linear arrangements. The results show that most of the tested LLMs cannot outperform the best uninformed baselines, with only the newest and largest versions of LLaMA doing so for most languages, and still achieving rather low performance. Thus, accurate zero-shot syntactic parsing is not forthcoming with open LLMs., We acknowledge grants SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; and TSI-100925-2023-1 funded by Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR; as well as funding by Xunta de Galicia (ED431C 2024/02), and Centro de Investigación de Galicia “CITIC”, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS). We also extend our gratitude to CESGA, the supercomputing center of Galicia, for granting us access to its resources., Xunta de Galicia; ED431C 2024/02
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Automated Code Generation for Tabular Question Answering in Zero-Shot Settings

  • Santos-Ríos, Roi
  • López Gude, Adrián
  • Prado-Valiño, Francisco
  • Ezquerro, Ana
  • Vilares, Jesús
Presentado en: VIII Congreso Xove TIC: impulsando el talento científico. Octubre, 2025, A Coruña., [Abstract] Our work presents an effective and robust approach to Tabular Question Answering based on a zero-shot pipeline that uses a Large Language Model to generate executable code for extracting information from tables. Our system includes a modular architecture: a code generation module supported by column selection and type inference components to guide accurate code synthesis, together with an iterative repair mechanism for runtime errors into revised prompts. Our approach is successfully tested both in English and Spanish., We acknowledge grants SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; CIDMEFEO funded by the Spanish National Statistics Institute (INE); as well as funding by Xunta de Galicia (ED431C 2024/02). CITIC, as a center accredited for excellence within the Galician University System and a member of the CIGUS Network, receives subsidies from the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia. Additionally, it is co-financed by the EU through the FEDER Galicia 2021-27 operational program (ED431G 2023/01)., Xunta de Galicia; ED431G 2023/01, Xunta de Galicia; ED431C 2024/02
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Enhancing Transformer-Based Sentiment Analysis for the Rest-Mex 2025 Challenge: A Hybrid Strategy with Oversampling, Back-Translation, and Transformers

  • Imran, Muhammad
  • Rasheed, Tayyab
  • Gómez-Rodríguez, Carlos
Presented at: Researching on Evaluating Sentiment and Textual Instances Selection for Mexican Magical Towns (REST-MEX 2025), at IberLEF 2025: Natural Language Processing Challenges for Spanish and other Iberian Languages, September 2025, Zaragoza, Spain The code is available in the following GitHub repository. https://github.com/chimran135/Sentiment-Analysis-Rest-Mex-2025, [Abstract]: This paper presents a sentiment analysis framework for the Rest-Mex 2025 challenge, focused on Spanish-language reviews of Mexican Magical Towns. The task involves predicting sentiment polarity (1-5), classifying attraction type (Hotel, Restaurant, Attraction), and identifying the correct town from a list of 60. To address class imbalance, we propose a hybrid augmentation approach combining oversampling and back-translation using both structurally similar and dissimilar languages. Two transformer-based models roberta-base-bne and twitter-xlm-roberta-base are fine-tuned on the augmented datasets. The hybrid strategy, particularly with the multilingual model, achieved the best results, demonstrating improved performance and generalization across all subtasks. Our system achieved 4th place in the overall sentiment analysis track of the Rest-Mex 2025 shared task competing against 35 participating teams which demonstrates the robustness of our approach in sentiment classification across multiple subtasks. © 2025 Copyright for this paper by its authors., We acknowledge grants GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; and TSI-100925-2023-1 funded by Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR; as well as funding by Xunta de Galicia (ED431C 2024/02), and CITIC, as a center accredited for excellence within the Galician University System and a member of the CIGUS Network, receives subsidies from the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia. Additionally, it is co-financed by the EU through the FEDER Galicia 2021-27 operational program (Ref. ED431G 2023/01)., Xunta de Galicia; ED431C 2024/02, Xunta de Galicia; ED431G 2023/01
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Spanish Resource Grammar Version 2023

  • Zamaraeva, Olga
  • Suárez Allegue, Lorena
  • Gómez-Rodríguez, Carlos
Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), held in Torino (Italia) 20-25 May, 2024., [Abstract]: We present the latest version of the Spanish Resource Grammar (SRG), a grammar of Spanish implemented in the HPSG formalism. Such grammars encode a complex set of hypotheses about syntax making them a resource for empirical testing of linguistic theory. They also encode a strict notion of grammaticality which makes them a resource for natural language processing applications in computer-assisted language learning. This version of the SRG uses the recent version of the Freeling morphological analyzer and is released along with an automatically created, manually verified treebank of 2,291 sentences. We explain the treebanking process, emphasizing how it is different from treebanking with manual annotation and how it contributes to empirically-driven development of syntactic theory. The treebanks’ high level of consistency and detail makes them a resource for training high-quality semantic parsers and generally systems that benefit from precise and detailed semantics. Finally, we present the grammar’s coverage and overgeneration on 100 sentences from a learner corpus, a new research line related to developing methodologies for robust empirical evaluation of hypotheses in second language acquisition., We acknowledge the European Union’s Horizon Europe Framework Programme which funded this research under the Marie Skłodowska-Curie post-doctoral fellowship grant HORIZON-MSCA-2021-PF-01 (GAUSS, grant agreement No 101063104); the European Research Council (ERC), which has funded this research under the Horizon Europe research and innovation programme (SALSA, grant agreement No 101100615); Grant SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; Xunta de Galicia (ED431C 2020/11); and Centro de Investigación de Galicia “CITIC”, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS). We also acknowledge grant GAP (PID2022-139308OA-I00) funded by MCIN/AEI/10.13039/501100011033/ and by ERDF “A way of making Europe”., Xunta de Galicia; ED431C 2020/11
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Comparing LLM-generated and human-authored news text using formal syntactic theory

  • Zamaraeva, Olga
  • Flickinger, Dan
  • Bond, Francis
  • Gómez-Rodríguez, Carlos
The Congress took place in Vienna, Austria from July 27 to August 1st, 2025, [Abstract]: This study provides the first comprehensive comparison of New York Times-style text generated by six large language models against real, human-authored NYT writing. The comparison is based on a formal syntactic theory. We use Head-driven Phrase Structure Grammar (HPSG) to analyze the grammatical structure of the texts. We then investigate and illustrate the differences in the distributions of HPSG grammar types, revealing systematic distinctions between human and LLM-generated writing. These findings contribute to a deeper understanding of the syntactic behavior of LLMs as well as humans, within the NYT genre., We thank Ann Copestake and Emily M. Bender, and more generally the DELPH-IN community for the useful discussion related to the paper. We acknowledge grants SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF, EU; LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF, EU; and TSI-100925-2023-1 funded by Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR; as well as funding by Xunta de Galicia (ED431C 2024/02). CITIC, as a center accredited for excellence within the Galician University System and a member of the CIGUS Network, receives subsidies from the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia. Additionally, it is co-financed by the EU through the FEDER Galicia 2021-27 operational program (Ref. ED431G 2023/01)., Xunta de Galicia; ED431C 2024/02, Xunta de Galicia; ED431G 2023/01
RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

On the impact of syntactic infusion for gender categorization across contextual dimensions, Sobre el impacto de la integración sintáctica en la categorización de género a través de dimensiones contextuales

  • Veiga Menéndez, Inés María
  • Muñoz-Ortiz, Alberto
  • Vilares, David
[Abstract]: This paper investigates how incorporating syntactic information can enhance the categorization of text into multiple gender dimensions, defined by our own identity (as category), the person we are addressing (to category), or the individual we are discussing (about category). Specifically, we explore the use of dependency grammars to integrate explicit syntactic embeddings while leveraging the strengths of pre-trained masked language models (MLMs). Our goal is to determine if dependency grammars add value beyond the implicit syntactic understanding already captured by MLMs. We begin by establishing a baseline using standard MLMs. Next, we propose a neural architecture that explicitly integrates dependency-based structures into this baseline, enabling a comparative analysis of performance and variations. Finally, in addition to evaluating the results, we analyzed the training dynamics of the two proposed variants to provide additional insights into their behavior during the fine-tuning stage. Explicit syntactic information boosts performance in single-task setups, though its gains fade in multitask scenarios., [Resumen]: Este artículo investiga cómo incorporar información sintáctica puede mejorar la clasificación de textos en múltiples dimensiones de género, definidas por nuestra propia identidad (categoría as), la persona a la que nos dirigimos (categoría to) o el individuo del que se habla (categoría about). En concreto, exploramos el uso de gramáticas de dependencias para integrar representaciones sintácticas explícitas, complementando las representaciones de modelos de lenguaje enmascarados preentrenados (MLMs). Nuestro objetivo es determinar si las gramáticas de dependencias aportan algo más allá de la comprensión sintáctica implícita ya capturada por los MLMs. Para ello, primero establecemos un modelo base usando un MLM estándar. A continuación, proponemos una arquitectura neuronal que integra en este modelo estructuras basadas en dependencias de forma explícita, permitiendo comparar el rendimiento y las variaciones. Finalmente, evaluamos los resultados y analizamos las dinámicas de entrenamiento de las dos variantes propuestas para ofrecer información adicional sobre su comportamiento durante la etapa de ajuste fino. La información sintáctica explícita mejora el rendimiento en configuraciones de tarea única, aunque sus beneficios disminuyen en escenarios multitarea., This work was supported by GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and by ERDF, EU; SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; Xunta de Galicia (ED431C 2024/02); Grant PRE2021-097001 funded by MICIU/AEI/10.13039/501100011033 and by ESF+ (predoctoral training grant associated to project PID2020-113230RB-C21); LATCHING (PID2023- 147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF; and Centro de Investigación de Galicia “CITIC”, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS)., Xunta de Galicia; ED431C 2024/02
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Evaluating Pixel Language Models on Non-Standardized Languages

  • Muñoz-Ortiz, Alberto
  • Blaschke, Verena
  • Plank, Barbara
Trabajo presentado a: 31st International Conference on Computational Linguistics - COLING, January 19–24, 2025., [Abstract]: We explore the potential of pixel-based models for transfer learning from standard languages to dialects. These models convert text into images that are divided into patches, enabling a continuous vocabulary representation that proves especially useful for out-of-vocabulary words common in dialectal data. Using German as a case study, we compare the performance of pixel-based models to token-based models across various syntactic and semantic tasks. Our results show that pixel-based models outperform token-based models in part-of-speech tagging, dependency parsing and intent detection for zero-shot dialect evaluation by up to 26 percentage points in some scenarios, though not in Standard German. However, pixel-based models fall short in topic classification. These findings emphasize the potential of pixel-based models for handling dialectal data, though further research should be conducted to assess their effectiveness in various linguistic contexts., This work was funded by the European Research Council (ERC) Consolidator Grant DIALECT 101043235; SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; Xunta de Galicia (ED431C 2024/02); GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and by ERDF, EU; Grant PRE2021-097001 funded by MICIU/AEI/10.13039/501100011033 and by ESF+ (predoctoral training grant associated to project PID2020-113230RB-C21); LATCH- ING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF; and Centro de Investigación de Galicia “CITIC”, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS)., Xunta de Galicia; ED431C 2024/02
RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

A Syntax-Injected Approach for Faster and More Accurate Sentiment Analysis

  • Imran, Muhammad
  • Kellert, Olga
  • Gómez-Rodríguez, Carlos
The following information was supplied regarding data availability: The OpeNERen, OpeNERes, UD_English-EWT and UD_Spanish-AnCora datasets are available at Zenodo: Muhammad Imran. (2025). chimran135/syntax-injected-sentimentanalysis: Syntax-injected Sentiment Analysis (1.0.0). Zenodo. https://doi.org/10.5281/zenodo.15323755. The Rest-Mex 2023 shared task website provides access to the training data for registered participants. The data are held privately for assessment purposes and can be obtained from Rest-Mex 2023 organizers https://sites.google.com/cimat.mx/rest-mex2023 by contacting Miguel Ángel Álvarez Carmona at miguel.alvarez@cimat.mx. The Rest-Mex 2023 dataset was used under license for this study., [Abstract]: Sentiment Analysis (SA) is a crucial aspect of Natural Language Processing (NLP), focusing on identifying and interpreting subjective assessments in textual content. Syntactic parsing is useful in SA as it improves accuracy and provides explainability; however, it often becomes a computational bottleneck due to slow parsing algorithms. This article proposes a solution to this bottleneck by using a Sequence Labeling Syntactic Parser (SELSP) to integrate syntactic information into SA via a rule-based sentiment analysis pipeline. By reformulating dependency parsing as a sequence labeling task, we significantly improve the efficiency of syntax-based SA. SELSP is trained and evaluated on a ternary polarity classification task, demonstrating greater speed and accuracy compared to conventional parsers like Stanza and heuristic approaches such as Valence Aware Dictionary and sEntiment Reasoner (VADER). The combination of speed and accuracy makes SELSP especially attractive for sentiment analysis applications in both academic and industrial contexts. Moreover, we compare SELSP with Transformer-based models trained on a 5-label classification task. In addition, we evaluate multiple sentiment dictionaries with SELSP to determine which yields the best performance in polarity prediction. The results show that dictionaries accounting for polarity judgment variation outperform those that ignore it. Furthermore, we show that SELSP outperforms Transformer-based models in terms of speed for polarity prediction., The European Research Council (ERC) funded this research under the Horizon Europe research and innovation programme (SALSA, grant agreement No. 101100615), SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033, LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF (EU), project GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and ERDF (EU), Ministry for Digital Transformation and Civil Service and “NextGenerationEU” PRTR under grant TSI-100925-2023-1, Xunta de Galicia (ED431C 2024/02), and Galician Research Center “CITIC”, funded by Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS). Furthermore, this research was supported by the International, Interdisciplinary and Intersectoral Information and Communications Technology PhD programme (3-i ICT) granted to CITIC and supported by the European Union through the Horizon 2020 research and innovation programme under a Marie Skłodowska-Curie agreement (H2020-MSCA-COFUND), GA 101034261. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript., Xunta de Galicia; ED431C 2024/02