Fabio Rinaldi

Vani Kanjirangat, Ljiljana Dolamic, Fabio Rinaldi

This paper discusses our exploration of different data-efficient and parameter-efficient approaches to Arabic Dialect Identification (ADI). In particular, we investigate various soft-prompting strategies, including prefix-tuning, prompt-tuning, P-tuning, and P-tuning V2, as well as LoRA reparameterizations. For the data-efficient strategy, we analyze hard prompting with zero-shot and few-shot inferences to analyze the dialect identification capabilities of Large Language Models (LLMs). For the parameter-efficient PEFT approaches, we conducted our experiments using Arabic-specific encoder models on several major datasets. We also analyzed the n-shot inferences on open-source decoder-only models, a general multilingual model (Phi-3.5), and an Arabic-specific one(SILMA). We observed that the LLMs generally struggle to differentiate the dialectal nuances in the few-shot or zero-shot setups. The soft-prompted encoder variants perform better, while the LoRA-based fine-tuned models perform best, even surpassing full fine-tuning.

Lenz Furrer, Joseph Cornelius, Fabio Rinaldi

Background: Named Entity Recognition (NER) and Normalisation (NEN) are core components of any text-mining system for biomedical texts. In a traditional concept-recognition pipeline, these tasks are combined in a serial way, which is inherently prone to error propagation from NER to NEN. We propose a parallel architecture, where both NER and NEN are modeled as a sequence-labeling task, operating directly on the source text. We examine different harmonisation strategies for merging the predictions of the two classifiers into a single output sequence. Results: We test our approach on the recent Version 4 of the CRAFT corpus. In all 20 annotation sets of the concept-annotation task, our system outperforms the pipeline system reported as a baseline in the CRAFT shared task 2019. Conclusions: Our analysis shows that the strengths of the two classifiers can be combined in a fruitful way. However, prediction harmonisation requires individual calibration on a development set for each annotation set. This allows achieving a good trade-off between established knowledge (training set) and novel information (unseen concepts). Availability and Implementation: Source code freely available for download at https://github.com/OntoGene/craft-st. Supplementary data are available at arXiv online.

Sandra Mitrović, David Kletz, Ljiljana Dolamic et al.

Large Language Models (LLMs) display notable variation in multilingual behavior, yet the role of genealogical language structure in shaping this variation remains underexplored. In this paper, we investigate whether LLMs exhibit sensitivity to linguistic genera by extending prior analyses on the MultiQ dataset. We first check if models prefer to switch to genealogically related languages when prompt language fidelity is not maintained. Next, we investigate whether knowledge consistency is better preserved within than across genera. We show that genus-level effects are present but strongly conditioned by training resource availability. We further observe distinct multilingual strategies across LLMs families. Our findings suggest that LLMs encode aspects of genus-level structure, but training data imbalances remain the primary factor shaping their multilingual performance.

Vani Kanjirangat, Tanja Samardžić, Ljiljana Dolamic et al.

Dialectal data are characterized by linguistic variation that appears small to humans but has a significant impact on the performance of models. This dialect gap has been related to various factors (e.g., data size, economic and social factors) whose impact, however, turns out to be inconsistent. In this work, we investigate factors impacting the model performance more directly: we correlate Tokenization Parity (TP) and Information Parity (IP), as measures of representational biases in pre-trained multilingual models, with the downstream performance. We compare state-of-the-art decoder-only LLMs with encoder-based models across three tasks: dialect classification, topic classification, and extractive question answering, controlling for varying scripts (Latin vs. non-Latin) and resource availability (high vs. low). Our analysis reveals that TP is a better predictor of the performance on tasks reliant on syntactic and morphological cues (e.g., extractive QA), while IP better predicts performance in semantic tasks (e.g., topic classification). Complementary analyses, including tokenizer behavior, vocabulary coverage, and qualitative insights, reveal that the language support claims of LLMs often might mask deeper mismatches at the script or token level.

Sanghwan Kim, Farhad Nooralahzadeh, Morteza Rohanian et al.

Recent transformer-based models have made significant strides in generating radiology reports from chest X-ray images. However, a prominent challenge remains: these models often lack prior knowledge, resulting in the generation of synthetic reports that mistakenly reference non-existent prior exams. This discrepancy can be attributed to a knowledge gap between radiologists and the generation models. While radiologists possess patient-specific prior information, the models solely receive X-ray images at a specific time point. To tackle this issue, we propose a novel approach that leverages a rule-based labeler to extract comparison prior information from radiology reports. This extracted comparison prior is then seamlessly integrated into state-of-the-art transformer-based models, enabling them to produce more realistic and comprehensive reports. Our method is evaluated on English report datasets, such as IU X-ray and MIMIC-CXR. The results demonstrate that our approach surpasses baseline models in terms of natural language generation metrics. Notably, our model generates reports that are free from false references to non-existent prior exams, setting it apart from previous models. By addressing this limitation, our approach represents a significant step towards bridging the gap between radiologists and generation models in the domain of medical report generation.

K Vani, Sandra Mitrovic, Alessandro Antonucci et al.

Lexical semantic change detection (also known as semantic shift tracing) is a task of identifying words that have changed their meaning over time. Unsupervised semantic shift tracing, focal point of SemEval2020, is particularly challenging. Given the unsupervised setup, in this work, we propose to identify clusters among different occurrences of each target word, considering these as representatives of different word meanings. As such, disagreements in obtained clusters naturally allow to quantify the level of semantic shift per each target word in four target languages. To leverage this idea, clustering is performed on contextualized (BERT-based) embeddings of word occurrences. The obtained results show that our approach performs well both measured separately (per language) and overall, where we surpass all provided SemEval baselines.

Ensieh Davoodijam, Nasser Ghadiri, Maryam Lotfi Shahreza et al.

Automatic text summarization methods generate a shorter version of the input text to assist the reader in gaining a quick yet informative gist. Existing text summarization methods generally focus on a single aspect of text when selecting sentences, causing the potential loss of essential information. In this study, we propose a domain-specific method that models a document as a multi-layer graph to enable multiple features of the text to be processed at the same time. The features we used in this paper are word similarity, semantic similarity, and co-reference similarity, which are modelled as three different layers. The unsupervised method selects sentences from the multi-layer graph based on the MultiRank algorithm and the number of concepts. The proposed MultiGBS algorithm employs UMLS and extracts the concepts and relationships using different tools such as SemRep, MetaMap, and OGER. Extensive evaluation by ROUGE and BERTScore shows increased F-measure values.

Seyedmostafa Sheikhalishahi, Riccardo Miotto, Joel T Dudley et al.

Of the 2652 articles considered, 106 met the inclusion criteria. Review of the included papers resulted in identification of 43 chronic diseases, which were then further classified into 10 disease categories using ICD-10. The majority of studies focused on diseases of the circulatory system (n=38) while endocrine and metabolic diseases were fewest (n=14). This was due to the structure of clinical records related to metabolic diseases, which typically contain much more structured data, compared with medical records for diseases of the circulatory system, which focus more on unstructured data and consequently have seen a stronger focus of NLP. The review has shown that there is a significant increase in the use of machine learning methods compared to rule-based approaches; however, deep learning methods remain emergent (n=3). Consequently, the majority of works focus on classification of disease phenotype with only a handful of papers addressing extraction of comorbidities from the free text or integration of clinical notes with structured data. There is a notable use of relatively simple methods, such as shallow classifiers (or combination with rule-based methods), due to the interpretability of predictions, which still represents a significant issue for more complex methods. Finally, scarcity of publicly available data may also have contributed to insufficient development of more advanced methods, such as extraction of word embeddings from clinical notes. Further efforts are still required to improve (1) progression of clinical NLP methods from extraction toward understanding; (2) recognition of relations among entities rather than entities in isolation; (3) temporal extraction to understand past, current, and future clinical events; (4) exploitation of alternative sources of clinical knowledge; and (5) availability of large-scale, de-identified clinical corpora.