Recent research has explored methods for updating and modifying factual knowledge in large language models, often focusing on specific multi-layer perceptron blocks. This study expands on this work by examining the effectiveness of existing knowledge editing methods across languages and delving into the role of attention mechanisms in this process. Drawing from the insights gained, we propose Mass-Editing Memory with Attention in Transformers (MEMAT), a method that achieves significant improvements in all metrics while requiring minimal parameter modifications. MEMAT delivers a remarkable 10% increase in magnitude metrics, benefits languages not included in the training data and also demonstrates a high degree of portability. Our code and data are at https://github.com/dtamayo-nlp/MEMAT.

We present the findings and results of our pseudonymisation system, which has been developed for a real-life use-case involving users and an informative chatbot in the context of the COVID-19 pandemic. Message exchanges between the two involve the former group providing information about themselves and their residential area, which could easily allow for their re-identification. We create a modular pipeline to detect PIIs and perform basic deidentification such that the data can be stored while mitigating any privacy concerns. The use-case presents several challenging aspects, the most difficult of which is the logistic challenge of not being able to directly view or access the data due to the very privacy issues we aim to resolve. Nevertheless, our system achieves a high recall of 0.99, correctly identifying almost all instances of personal data. However, this comes at the expense of precision, which only reaches 0.64. We describe the sensitive information identification in detail, explaining the design principles behind our decisions. We additionally highlight the particular challenges we’ve encountered.

The development of large language models (LLMs) relies heavily on extensive, high-quality datasets. Publicly available datasets focus predominantly on English, leaving other language communities behind. To address this issue, we introduce Community OSCAR, a multilingual dataset initiative designed to address the gap between English and non-English data availability. Through a collective effort, Community OSCAR covers over 150 languages with 45 billion documents, totaling over 345 TiB of data. Initial results indicate that Community OSCAR provides valuable raw data for training LLMs and enhancing the performance of multilingual models. This work aims to contribute to the ongoing advancements in multilingual NLP and to support a more inclusive AI ecosystem by making high-quality, multilingual data more accessible to those working with low-resource languages.

We present and describe two language resources in this paper: CATalog 1.0, the largest text corpus in Catalan to date, and CURATE (Corpus Utility for RAting TExt), a modular, parallelizable pipeline used for processing and scoring documents based on text quality that we have optimised to run in High Performance Cluster (HPC) environments. In the coming sections we describe our data preprocessing pipeline at length; traditional pipelines usually implement a set of binary filters such that a given document is either in or out. In our experience with Catalan, in lower-resource settings it is more practical to instead assign a document a soft score to allow for more flexible decision-making. We describe how the document score is calculated and highlight its interpretability by showing that it is significantly correlated with human judgements as obtained from a comparative judgement experiment. We additionally describe the different subcorpora that make up CATalog 1.0.

Collecting voice resources for speech recognition systems is a multifaceted challenge, involving legal, technical, and diversity considerations. However, it is crucial to ensure fair access to voice-driven technology across diverse linguistic backgrounds. We describe an ongoing effort to create an extensive, high-quality, publicly available voice dataset for future development of speech technologies in Catalan through the Mozilla Common Voice crowd-sourcing platform. We detail the specific approaches used to address the challenges faced in recruiting contributors and managing the collection, validation, and recording of sentences. This detailed overview can serve as a source of guidance for similar initiatives across other projects and linguistic contexts. The success of this project is evident in the latest corpus release, version 16.1, where Catalan ranks as the most prominent language in the corpus, both in terms of recorded hours and when considering validated hours. This establishes Catalan as a language with significant speech resources for language technology development and significantly raises its international visibility.

Current LLM-based applications are becoming steadily available for everyone with a reliable access to technology and the internet. These applications offer benefits to their users that leave those without access to them at a serious disadvantage. Given the vastly large amount of data needed to train LLMs, the gap between languages with access to such quantity of data and those without it is currently larger than ever. Aimed at saving this gap, the Aina Project was created to provide Catalan with the necessary resources to keep being relevant in the context of AI/NLP applications based on LLMs. We thus present a set of strategies to consider when improving technology support for a mid- or low-resource language, specially addressing sustainability of high-quality data acquisition and the challenges involved in the process. We also introduce a large amount of new annotated data for Catalan. Our hope is that those interested in replicating this work for another language can learn from what worked for us, the challenges that we faced, and the sometimes disheartening truth of working with mid- and low-resource languages.

Large language models have amply proven their great capabilities, both in downstream tasks and real-life settings. However, low- and mid-resource languages do not have access to the necessary means to train such models from scratch, and often have to rely on multilingual models despite being underrepresented in the training data. For the particular case of the Catalan language, we prove that continued pre-training with vocabulary adaptation is a better alternative to take the most out of already pre-trained models, even if these have not seen any Catalan data during their pre-training phase. We curate a 26B tokens corpus and use it to further pre-train BLOOM, giving rise to the FLOR models. We perform an extensive evaluation to assess the effectiveness of our method, obtaining consistent gains across Catalan and Spanish tasks. The models, training data, and evaluation framework are made freely available under permissive licenses.

Zero-shot text classification is a widely studied task that deals with a lack of annotated data. The most common approach is to reformulate it as a textual entailment problem, enabling classification into unseen classes. This work explores an effective approach that trains on a weakly supervised dataset generated from traditional classification data. We empirically study the relation between the performance of the entailment task, which is used as a proxy, and the target zero-shot text classification task. Our findings reveal that there is no linear correlation between both tasks, to the extent that it can be detrimental to lengthen the fine-tuning process even when the model is still learning, and propose a straightforward method to stop training on time. As a proof of concept, we introduce a domain-specific zero-shot text classifier that was trained on Microsoft Academic Graph data. The model, called SCIroShot, achieves state-of-the-art performance in the scientific domain and competitive results in other areas. Both the model and evaluation benchmark are publicly available on HuggingFace and GitHub.

This work presents the first large-scale biomedical Spanish language models trained from scratch, using large biomedical corpora consisting of a total of 1.1B tokens and an EHR corpus of 95M tokens. We compared them against general-domain and other domain-specific models for Spanish on three clinical NER tasks. As main results, our models are superior across the NER tasks, rendering them more convenient for clinical NLP applications. Furthermore, our findings indicate that when enough data is available, pre-training from scratch is better than continual pre-training when tested on clinical tasks, raising an exciting research question about which approach is optimal. Our models and fine-tuning scripts are publicly available at HuggingFace and GitHub.

Recently, various end-to-end architectures of Automatic Speech Recognition (ASR) are being showcased as an important step towards providing language technologies to all languages instead of a select few such as English. However many languages are still suffering due to the “digital gap,” lacking thousands of hours of transcribed speech data openly accessible that is necessary to train modern ASR architectures. Although Catalan already has access to various open speech corpora, these corpora lack diversity and are limited in total volume. In order to address this lack of resources for Catalan language, in this work we present ParlamentParla, a corpus of more than 600 hours of speech from Catalan Parliament sessions. This corpus has already been used in training of state-of-the-art ASR systems, and proof-of-concept text-to-speech (TTS) models. In this work we explain in detail the pipeline that allows the information publicly available on the parliamentary website to be converted to a speech corpus compatible with training of ASR and possibly TTS models.

Nested Named Entity Recognition (NER) is an information extraction task that aims to identify entities that may be nested within other entity mentions. Despite the availability of several corpora with nested entities in the Spanish clinical domain, most previous work has overlooked them due to the lack of models and a clear annotation scheme for dealing with the task. To fill this gap, this paper provides an empirical study of straightforward methods for tackling the nested NER task on two Spanish clinical datasets, Clinical Trials, and the Chilean Waiting List. We assess the advantages and limitations of two sequence labeling approaches; one based on Multiple LSTM-CRF architectures and another on Joint labeling models. To better understand the differences between these models, we compute task-specific metrics that adequately measure the ability of models to detect nested entities and perform a fine-grained comparison across models. Our experimental results show that employing domain-specific language models trained from scratch significantly improves the performance obtained with strong domain-specific and general-domain baselines, achieving state-of-the-art results in both datasets. Specifically, we obtained F1 scores of 89.21 and 83.16 in Clinical Trials and the Chilean Waiting List, respectively. Interestingly enough, we observe that the task-specific metrics and analysis properly reflect the limitations of the models when recognizing nested entities. Finally, we perform a case study on an aggregated NER dataset created from several clinical corpora in Spanish. We highlight how entity length and the simultaneous recognition of inner and outer entities are the most critical variables for the nested NER task.

One of the biomedical entity types of relevance for medicine or biosciences are chemical compounds and drugs. The correct detection these entities is critical for other text mining applications building on them, such as adverse drug-reaction detection, medication-related fake news or drug-target extraction. Although a significant effort was made to detect mentions of drugs/chemicals in English texts, so far only very limited attempts were made to recognize them in medical documents in other languages. Taking into account the growing amount of medical publications and clinical records written in Spanish, we have organized the first shared task on detecting drug and chemical entities in Spanish medical documents. Additionally, we included a clinical concept-indexing sub-track asking teams to return SNOMED-CT identifiers related to drugs/chemicals for a collection of documents. For this task, named PharmaCoNER, we generated annotation guidelines together with a corpus of 1,000 manually annotated clinical case studies. A total of 22 teams participated in the sub-track 1, (77 system runs), and 7 teams in the sub-track 2 (19 system runs). Top scoring teams used sophisticated deep learning approaches yielding very competitive results with F-measures above 0.91. These results indicate that there is a real interest in promoting biomedical text mining efforts beyond English. We foresee that the PharmaCoNER annotation guidelines, corpus and participant systems will foster the development of new resources for clinical and biomedical text mining systems of Spanish medical data.

Word embeddings are representations of words in a dense vector space. Although they are not recent phenomena in Natural Language Processing (NLP), they have gained momentum after the recent developments of neural methods and Word2Vec. Regarding their applications in medical and clinical NLP, they are invaluable resources when training in-domain named entity recognition systems, classifiers or taggers, for instance. Thus, the development of tailored word embeddings for medical NLP is of great interest. However, we identified a gap in the literature which we aim to fill in this paper: the availability of embeddings for medical NLP in Spanish, as well as a standardized form of intrinsic evaluation. Since most work has been done for English, some established datasets for intrinsic evaluation are already available. In this paper, we show the steps we employed to adapt such datasets for the first time to Spanish, of particular relevance due to the considerable volume of EHRs in this language, as well as the creation of in-domain medical word embeddings for the Spanish using the state-of-the-art FastText model. We performed intrinsic evaluation with our adapted datasets, as well as extrinsic evaluation with a named entity recognition systems using a baseline embedding of general-domain. Both experiments proved that our embeddings are suitable for use in medical NLP in the Spanish language, and are more accurate than general-domain ones.

The experiments presented here exploit the properties of the Apertium RDF Graph, principally cycle density and nodes’ degree, to automatically generate new translation relations between words, and therefore to enrich existing bilingual dictionaries with new entries. Currently, the Apertium RDF Graph includes data from 22 Apertium bilingual dictionaries and constitutes a large unified array of linked lexical entries and translations that are available and accessible on the Web (http://linguistic.linkeddata.es/apertium/). In particular, its graph structure allows for interesting exploitation opportunities, some of which are addressed in this paper. Two ‘massive’ experiments are reported: in the first one, the original EN-ES translation set was removed from the Apertium RDF Graph and a new EN-ES version was generated. The results were compared against the previously removed EN-ES data and against the Concise Oxford Spanish Dictionary. In the second experiment, a new non-existent EN-FR translation set was generated. In this case the results were compared against a converted wiktionary English-French file. The results we got are really good and perform well for the extreme case of correlated polysemy. This lead us to address the possibility to use cycles and nodes degree to identify potential oddities in the source data. If cycle density proves efficient when considering potential targets, we can assume that in dense graphs nodes with low degree may indicate potential errors.

The proliferation of different metadata schemas and models pose serious problems of interoperability. Maintaining isolated repositories with overlapping data is costly in terms of time and effort. In this paper, we describe how we have achieved a Linked Open Data version of metadata descriptions coming from heterogeneous sources, originally encoded in XML. The resulting model is much simpler than the original XSD schema and avoids problems typical of XML syntax, such as semantic ambiguity and order constraint. Moreover, the open world assumption of RDF/OWL allows to naturally integrate objects from different schemas and to add further extensions, facilitating merging of different models as well as linking to external data. Apart from the advantages in terms of interoperability and maintainability, the merged repository enables end-users to query multiple sources using a unified schema and is able to present them with implicit knowledge derived from the linked data. The approach we present here is easily scalable to any number of sources and schemas.

This paper describes on-going work for the construction of a new treebank for Spanish, The IULA Treebank. This new resource will contain about 60,000 richly annotated sentences as an extension of the already existing IULA Technical Corpus which is only PoS tagged. In this paper we have focused on describing the work done for defining the annotation process and the treebank design principles. We report on how the used framework, the DELPH-IN processing framework, has been crucial in the design principles and in the bootstrapping strategy followed, especially in what refers to the use of stochastic modules for reducing parsing overgeneration. We also report on the different evaluation experiments carried out to guarantee the quality of the already available results.

This paper describes on-going work for the construction of a new treebank for Spanish, The IULA Treebank. This new resource will contain about 60,000 richly annotated sentences as an extension of the already existing IULA Technical Corpus which is only PoS tagged. In this paper we have focused on describing the work done for defining the annotation process and the treebank design principles. We report on how the used framework, the DELPH-IN processing framework, has been crucial in the design principles and in the bootstrapping strategy followed, especially in what refers to the use of stochastic modules for reducing parsing overgeneration. We also report on the different evaluation experiments carried out to guarantee the quality of the already available results.

In this paper we present two real cases, in the fields of discourse analysis of newspapers and communication research which demonstrate the impact of Language Resources (LR) and NLP in the humanities. We describe our collaboration with (i) the Feminario research group from the UAB which has been investigating androcentric practices in Spanish general press since the 80s and whose research suggests that Spanish general press has undergone a dehumanization process that excludes women and men and (ii) the “Municipals'11 online” project which investigates the Spanish local election campaign in the blogosphere. We will see how NLP tools and LRs make possible the so called ‘e-Humanities research' as they provide Humanities with tools to perform intensive and automatic text analyses. Language technologies have evolved a lot and are mature enough to provide useful tools to researchers dealing with large amount of textual data. The language resources that have been developed within the field of NLP have proven to be useful for other disciplines that are unaware of their existence and nevertheless would greatly benefit from them as they provide (i) exhaustiveness -to guarantee that data coverage is wide and representative enough- and (ii) reliable and significant results -to guarantee that the reported results are statistically significant.

This paper reports our experience when integrating differ resources and services into a grid environment. The use case we address implies the deployment of several NLP applications as web services. The ultimate objective of this task was to create a scenario where researchers have access to a variety of services they can operate. These services should be easy to invoke and able to interoperate between one another. We essentially describe the interoperability problems we faced, which involve metadata interoperability, data interoperability and service interoperability. We devote special attention to service interoperability and explore the possibility to define common interfaces and semantic description of services. While the web services paradigm suits the integration of different services very well, this requires mutual understanding and the accommodation to common interfaces that not only provide technical solution but also ease the user‟s work. Defining common interfaces benefits interoperability but requires the agreement about operations and the set of inputs/outputs. Semantic annotation allows defining some sort of taxonomy that organizes and collects the set of admissible operations and types input/output parameters.

Despite of the importance of lexical resources for a number of NLP applications (Machine Translation, Information Extraction, Question Answering, among others), there has been a traditional lack of generic tools for the creation, maintenance and management of computational lexica. The most direct obstacle for the development of generic tools, independent of any particular application format, was the lack of standards for the description and encoding of lexical resources. The availability of the Lexical Markup Framework (LMF) has changed this scenario and has made it possible the development of generic lexical platforms. COLDIC is a generic platform for working with computational lexica. The system has been designed to let the user concentrate on lexicographical tasks, but still being autonomous in the management of the tools. The creation and maintenance of the database, which is the core of the tool, demand no specific training in databases. A LMF compliant schema implemented in a Document Type Definition (DTD) describing the lexical resources is taken by the system to automatically configure the platform. Besides, the most standard web services for interoperability are also generated automatically. Other components of the platform include build-in functions supporting the most common tasks of the lexicographic work.

The ISLE project is a continuation of the long standing EAGLES initiative, carried out under the Human Language Technology (HLT) programme in collaboration between American and European groups in the framework of the EU-US International Research Co-operation, supported by NSF and EC. In this paper we concentrate on the current position of the ISLE Computational Lexicon Working Group (CLWG), whose activities aim at defining a general schema for a multilingual lexical entry (MILE), as the basis for a standard framework for multilingual computational lexicons. The needs and features of existing Machine Translation systems provide the main reference points for the process of consensual definition of the MILE. The overall structure of the MILE will be illustrated with particular attention to some of the issues raised for multilingual lexicons by the need of expressing complex transfer conditions among translation equivalents