The NLP community typically relies on performance of a model on a held-out test set to assess generalization. Performance drops observed in datasets outside of official test sets are generally attributed to “out-of-distribution” effects. Here, we explore the foundations of generalizability and study the factors that affect it, articulating lessons from clinical studies. In clinical research, generalizability is an act of reasoning that depends on (a) *internal validity* of experiments to ensure controlled measurement of cause and effect, and (b) *external validity* or transportability of the results to the wider population. We demonstrate how learning spurious correlations, such as the distance between entities in relation extraction tasks, can affect a model’s internal validity and in turn adversely impact generalization. We, therefore, present the need to ensure internal validity when building machine learning models in NLP. Our recommendations also apply to generative large language models, as they are known to be sensitive to even minor semantic preserving alterations. We also propose adapting the idea of *matching* in randomized controlled trials and observational studies to NLP evaluation to measure causation.

In this work, we measure the impact of affixal negation on modern English large language models (LLMs). In affixal negation, the negated meaning is expressed through a negative morpheme, which is potentially challenging for LLMs as their tokenizers are often not morphologically plausible. We conduct extensive experiments using LLMs with different subword tokenization methods, which lead to several insights on the interaction between tokenization performance and negation sensitivity. Despite some interesting mismatches between tokenization accuracy and negation detection performance, we show that models can, on the whole, reliably recognize the meaning of affixal negation.

In this short position paper, we highlight the importance of numbers in clinical text. We first present a taxonomy of number variants. We then perform corpus analysis to analyze characteristics of number use in several clinical corpora. Based on our findings of extensive use of numbers, and limited understanding of the impact of numbers on clinical NLP tasks, we identify the need for a public benchmark that will support investigation of numerical processing tasks for the clinical domain.

We explore different information extraction tools for annotation of interventions to support automated systematic reviews of preclinical AD animal studies. We compare two PICO (Population, Intervention, Comparison, and Outcome) extraction tools and two prompting-based learning strategies based on Large Language Models (LLMs). Motivated by the high recall of a dictionary-based approach, we define a two-stage method, removing false positives obtained from regexes with a pre-trained LM. With ChatGPT-based filtering using three-shot prompting, our approach reduces almost two-thirds of False Positives compared to the dictionary approach alone, while outperforming knowledge-free instructional prompting.

This paper describes the submissions of the Natural Language Processing (NLP) team from the Australian Research Council Industrial Transformation Training Centre (ITTC) for Cognitive Computing in Medical Technologies to the SemEval 2023 Task 7, i.e., multi-evidence natural language inference for clinical trial data (NLI4CT). More specifically, we were working on subtask 2 whose objective is to identify the relevant parts of the premise from clinical trial report that justify the truth of information in the statement. We approach the evidence retrieval problem as a document retrieval and sentence similarity task. Our results show that the task poses some challenges which involve dealing with complex sentences and implicit evidences.

Automatically rating the quality of published research is a critical step in medical evidence synthesis. While several methods have been proposed, their algorithmic fairness has been overlooked even though significant risks may follow when such systems are deployed in biomedical contexts. In this work, we study fairness on two systems along two sensitive attributes, participant sex and medical area. In some cases, we find important inequalities, leading us to apply various debiasing methods. Upon examining an interplay of systems’ predictive performance, fairness, as well as medically critical selective classification capabilities and calibration performance, we find that fairness can sometimes improve through debiasing, but at a cost in other performance measures.

BERT-based models have had strong performance on leaderboards, yet have been demonstrably worse in real-world settings requiring generalization. Limited quantities of training data is considered a key impediment to achieving generalizability in machine learning. In this paper, we examine the impact of training data quality, not quantity, on a model’s generalizability. We consider two characteristics of training data: the portion of human-adversarial (h-adversarial), i.e. sample pairs with seemingly minor differences but different ground-truth labels, and human-affable (h-affable) training samples, i.e. sample pairs with minor differences but the same ground-truth label. We find that for a fixed size of training samples, as a rule of thumb, having 10-30% h-adversarial instances improves the precision, and therefore F1, by up to 20 points in the tasks of text classification and relation extraction. Increasing h-adversarials beyond this range can result in performance plateaus or even degradation. In contrast, h-affables may not contribute to a model’s generalizability and may even degrade generalization performance.

Named entity recognition (NER) in clinical documentation is often hindered by the use of highly specialised terminology, variation in language used to express medical findings and general scarcity of high-quality data available for training. This short paper compares a Conditional Random Fields model to the previously established dictionary-based approach and evaluates its ability to extract information from a small corpus of annotated pathology reports. The results suggest that including token descriptors as well as contextual features significantly improves precision on several concept categories while maintaining the same level of recall.

Negation has been shown to be a major bottleneck for masked language models, such as BERT. However, whether this finding still holds for larger-sized auto-regressive language models (“LLMs”) has not been studied comprehensively. With the ever-increasing volume of research and applications of LLMs, we take a step back to evaluate the ability of current-generation LLMs to handle negation, a fundamental linguistic phenomenon that is central to language understanding. We evaluate different LLMs - including the open-source GPT-neo, GPT-3, and InstructGPT - against a wide range of negation benchmarks. Through systematic experimentation with varying model sizes and prompts, we show that LLMs have several limitations including insensitivity to the presence of negation, an inability to capture the lexical semantics of negation, and a failure to reason under negation.

Despite the subjective nature of semantic textual similarity (STS) and pervasive disagreements in STS annotation, existing benchmarks have used averaged human ratings as gold standard. Averaging masks the true distribution of human opinions on examples of low agreement, and prevents models from capturing the semantic vagueness that the individual ratings represent. In this work, we introduce USTS, the first Uncertainty-aware STS dataset with ∼15,000 Chinese sentence pairs and 150,000 labels, to study collective human opinions in STS. Analysis reveals that neither a scalar nor a single Gaussian fits a set of observed judgments adequately. We further show that current STS models cannot capture the variance caused by human disagreement on individual instances, but rather reflect the predictive confidence over the aggregate dataset.

Negation is a common linguistic feature that is crucial in many language understanding tasks, yet it remains a hard problem due to diversity in its expression in different types of text. Recent works show that state-of-the-art NLP models underperform on samples containing negation in various tasks, and that negation detection models do not transfer well across domains. We propose a new negation-focused pre-training strategy, involving targeted data augmentation and negation masking, to better incorporate negation information into language models. Extensive experiments on common benchmarks show that our proposed approach improves negation detection performance and generalizability over the strong baseline NegBERT (Khandelwal and Sawant, 2020).

Although multi-document summarisation (MDS) of the biomedical literature is a highly valuable task that has recently attracted substantial interest, evaluation of the quality of biomedical summaries lacks consistency and transparency. In this paper, we examine the summaries generated by two current models in order to understand the deficiencies of existing evaluation approaches in the context of the challenges that arise in the MDS task. Based on this analysis, we propose a new approach to human evaluation and identify several challenges that must be overcome to develop effective biomedical MDS systems.

Though recently there have been an increased interest in how pre-trained language models encode different linguistic features, there is still a lack of systematic comparison between languages with different morphology and syntax. In this paper, using BERT as an example of a pre-trained model, we compare how three typologically different languages (English, Korean, and Russian) encode morphology and syntax features across different layers. In particular, we contrast languages which differ in a particular aspect, such as flexibility of word order, head directionality, morphological type, presence of grammatical gender, and morphological richness, across four different tasks.

Procedural text contains rich anaphoric phenomena, yet has not received much attention in NLP. To fill this gap, we investigate the textual properties of two types of procedural text, recipes and chemical patents, and generalize an anaphora annotation framework developed for the chemical domain for modeling anaphoric phenomena in recipes. We apply this framework to annotate the RecipeRef corpus with both bridging and coreference relations. Through comparison to chemical patents, we show the complexity of anaphora resolution in recipes. We demonstrate empirically that transfer learning from the chemical domain improves resolution of anaphora in recipes, suggesting transferability of general procedural knowledge.

We present M3 (Multi-level dataset for Multi-document summarisation of Medical studies), a benchmark dataset for evaluating the quality of summarisation systems in the biomedical domain. The dataset contains sets of multiple input documents and target summaries of three levels of complexity: documents, sentences, and propositions. The dataset also includes several levels of annotation, including biomedical entities, direction, and strength of relations between them, and the discourse relationships between the input documents (“contradiction” or “agreement”). We showcase usage scenarios of the dataset by testing 10 generic and domain-specific summarisation models in a zero-shot setting, and introduce a probing task based on counterfactuals to test if models are aware of the direction and strength of the conclusions generated from input studies.

Negation is poorly captured by current language models, although the extent of this problem is not widely understood. We introduce a natural language inference (NLI) test suite to enable probing the capabilities of NLP methods, with the aim of understanding sub-clausal negation. The test suite contains premise–hypothesis pairs where the premise contains sub-clausal negation and the hypothesis is constructed by making minimal modifications to the premise in order to reflect different possible interpretations. Aside from adopting standard NLI labels, our test suite is systematically constructed under a rigorous linguistic framework. It includes annotation of negation types and constructions grounded in linguistic theory, as well as the operations used to construct hypotheses. This facilitates fine-grained analysis of model performance. We conduct experiments using pre-trained language models to demonstrate that our test suite is more challenging than existing benchmarks focused on negation, and show how our annotation supports a deeper understanding of the current NLI capabilities in terms of negation and quantification.

Scientific documents typically contain numerous entity mentions, while only a subset are directly relevant to the key contributions of the paper. Distinguishing these focus entities from background ones effectively could improve the recovery of relevant documents and the extraction of information from documents. To study the identification of focus entities, we developed two large datasets of disease-causing biological pathogens using MEDLINE, the largest collection of biomedical citations, and PubMed Central, a collection of full text articles. The focus entities were identified using human-curated indexing on these collections. Experiments with machine learning methods to identify focus entities show that transformer methods achieve high precision and recall and that document discourse information is relevant. The work lays the foundation for more targeted retrieval/summarisation of entity-relevant documents.

State-of-the-art classification and regression models are often not well calibrated, and cannot reliably provide uncertainty estimates, limiting their utility in safety-critical applications such as clinical decision-making. While recent work has focused on calibration of classifiers, there is almost no work in NLP on calibration in a regression setting. In this paper, we quantify the calibration of pre- trained language models for text regression, both intrinsically and extrinsically. We further apply uncertainty estimates to augment training data in low-resource domains. Our experiments on three regression tasks in both self-training and active-learning settings show that uncertainty estimation can be used to increase overall performance and enhance model generalization.

In this paper we report the experiments performed for the submission to the Multidocument summarisation for Literature Review (MSLR) Shared Task. In particular, we adopt Primera model to the biomedical domain by placing global attention on important biomedical entities in several ways. We analyse the outputs of 23 resulting models and report some patterns related to the presence of additional global attention, number of training steps and the inputs configuration.

Training with noisy labelled data is known to be detrimental to model performance, especially for high-capacity neural network models in low-resource domains. Our experiments suggest that standard regularisation strategies, such as weight decay and dropout, are ineffective in the face of noisy labels. We propose a simple noisy label detection method that prevents error propagation from the input layer. The approach is based on the observation that the projection of noisy labels is learned through memorisation at advanced stages of learning, and that the Pearson correlation is sensitive to outliers. Extensive experiments over real-world human-disagreement annotations as well as randomly-corrupted and data-augmented labels, across various tasks and domains, demonstrate that our method is effective, regularising noisy labels and improving generalisation performance.

We describe the work of the READ-BioMed team for the preparation of a submission to the SocialDisNER Disease Named Entity Recognition (NER) Task (Task 10) in 2022. We had developed a system for named entity recognition for identifying biomedical concepts in English MEDLINE citations and Spanish clinical text for the LivingNER 2022 challenge. Minimal adaptation of our system was required to perform named entity recognition in the Spanish tweets in the SocialDisNER task, given the availability of Spanish pre-trained language models and the SocialDisNER training data. Minor additions included treatment of emojis and entities in hashtags and Twitter account names.

In developing systems to identify focus entities in scientific literature, we face the problem of discriminating key entities of interest from other potentially relevant entities of the same type mentioned in the articles. We introduce the task of pathogen characterisation. We aim to discriminate mentions of biological pathogens, that are actively studied in the research presented in scientific publications. These are the pathogens that are the focus of direct experimentation in the research, rather than those that are referred to for context or as playing secondary roles. In this paper, we explore the hypothesis that these focus entities can be differentiated from other, non-actively studied, pathogens mentioned in articles through analysis of the patterns of mentions across different sections of a scientific paper, that is, using the discourse structure of the paper. We provide an indicative case study with the help of a small data set of PubMed abstracts that have been annotated with actively mentioned pathogens.

Public datasets are often used to evaluate the efficacy and generalizability of state-of-the-art methods for many tasks in natural language processing (NLP). However, the presence of overlap between the train and test datasets can lead to inflated results, inadvertently evaluating the model’s ability to memorize and interpreting it as the ability to generalize. In addition, such data sets may not provide an effective indicator of the performance of these methods in real world scenarios. We identify leakage of training data into test data on several publicly available datasets used to evaluate NLP tasks, including named entity recognition and relation extraction, and study them to assess the impact of that leakage on the model’s ability to memorize versus generalize.

Chemical patents contain rich coreference and bridging links, which are the target of this research. Specially, we introduce a novel annotation scheme, based on which we create the ChEMU-Ref dataset from reaction description snippets in English-language chemical patents. We propose a neural approach to anaphora resolution, which we show to achieve strong results, especially when jointly trained over coreference and bridging links.

Domain pretraining followed by task fine-tuning has become the standard paradigm for NLP tasks, but requires in-domain labelled data for task fine-tuning. To overcome this, we propose to utilise domain unlabelled data by assigning pseudo labels from a general model. We evaluate the approach on two clinical STS datasets, and achieve r= 0.80 on N2C2-STS. Further investigation reveals that if the data distribution of unlabelled sentence pairs is closer to the test data, we can obtain better performance. By leveraging a large general-purpose STS dataset and small-scale in-domain training data, we obtain further improvements to r= 0.90, a new SOTA.

We present our work on aligning the Unified Medical Language System (UMLS) to Wikipedia, to facilitate manual alignment of the two resources. We propose a cross-lingual neural reranking model to match a UMLS concept with a Wikipedia page, which achieves a recall@1of 72%, a substantial improvement of 20% over word- and char-level BM25, enabling manual alignment with minimal effort. We release our resources, including ranked Wikipedia pages for 700k UMLSconcepts, and WikiUMLS, a dataset for training and evaluation of alignment models between UMLS and Wikipedia collected from Wikidata. This will provide easier access to Wikipedia for health professionals, patients, and NLP systems, including in multilingual settings.

Efficient discovery and exploration of biomedical literature has grown in importance in the context of the COVID-19 pandemic, and topic-based methods such as latent Dirichlet allocation (LDA) are a useful tool for this purpose. In this study we compare traditional topic models based on word tokens with topic models based on medical concepts, and propose several ways to improve topic coherence and specificity.

In this paper, we apply pre-trained language models to the Semantic Textual Similarity (STS) task, with a specific focus on the clinical domain. In low-resource setting of clinical STS, these large models tend to be impractical and prone to overfitting. Building on BERT, we study the impact of a number of model design choices, namely different fine-tuning and pooling strategies. We observe that the impact of domain-specific fine-tuning on clinical STS is much less than that in the general domain, likely due to the concept richness of the domain. Based on this, we propose two data augmentation techniques. Experimental results on N2C2-STS 1 demonstrate substantial improvements, validating the utility of the proposed methods.

Identifying the reasons for antibiotic administration in veterinary records is a critical component of understanding antimicrobial usage patterns. This informs antimicrobial stewardship programs designed to fight antimicrobial resistance, a major health crisis affecting both humans and animals in which veterinarians have an important role to play. We propose a document classification approach to determine the reason for administration of a given drug, with particular focus on domain adaptation from one drug to another, and instance selection to minimize annotation effort.

Extracting chemical reactions from patents is a crucial task for chemists working on chemical exploration. In this paper we introduce the novel task of detecting the textual spans that describe or refer to chemical reactions within patents. We formulate this task as a paragraph-level sequence tagging problem, where the system is required to return a sequence of paragraphs which contain a description of a reaction. To address this new task, we construct an annotated dataset from an existing proprietary database of chemical reactions manually extracted from patents. We introduce several baseline methods for the task and evaluate them over our dataset. Through error analysis, we discuss what makes the task complex and challenging, and suggest possible directions for future research.

This paper focuses on a traditional relation extraction task in the context of limited annotated data and a narrow knowledge domain. We explore this task with a clinical corpus consisting of 200 breast cancer follow-up treatment letters in which 16 distinct types of relations are annotated. We experiment with an approach to extracting typed relations called window-bounded co-occurrence (WBC), which uses an adjustable context window around entity mentions of a relevant type, and compare its performance with a more typical intra-sentential co-occurrence baseline. We further introduce a new bag-of-concepts (BoC) approach to feature engineering based on the state-of-the-art word embeddings and word synonyms. We demonstrate the competitiveness of BoC by comparing with methods of higher complexity, and explore its effectiveness on this small dataset.

Chemical patents are an important resource for chemical information. However, few chemical Named Entity Recognition (NER) systems have been evaluated on patent documents, due in part to their structural and linguistic complexity. In this paper, we explore the NER performance of a BiLSTM-CRF model utilising pre-trained word embeddings, character-level word representations and contextualized ELMo word representations for chemical patents. We compare word embeddings pre-trained on biomedical and chemical patent corpora. The effect of tokenizers optimized for the chemical domain on NER performance in chemical patents is also explored. The results on two patent corpora show that contextualized word representations generated from ELMo substantially improve chemical NER performance w.r.t. the current state-of-the-art. We also show that domain-specific resources such as word embeddings trained on chemical patents and chemical-specific tokenizers, have a positive impact on NER performance.

In the fourth edition of the WMT Biomedical Translation task, we considered a total of six languages, namely Chinese (zh), English (en), French (fr), German (de), Portuguese (pt), and Spanish (es). We performed an evaluation of automatic translations for a total of 10 language directions, namely, zh/en, en/zh, fr/en, en/fr, de/en, en/de, pt/en, en/pt, es/en, and en/es. We provided training data based on MEDLINE abstracts for eight of the 10 language pairs and test sets for all of them. In addition to that, we offered a new sub-task for the translation of terms in biomedical terminologies for the en/es language direction. Higher BLEU scores (close to 0.5) were obtained for the es/en, en/es and en/pt test sets, as well as for the terminology sub-task. After manual validation of the primary runs, some submissions were judged to be better than the reference translations, for instance, for de/en, en/es and es/en.

We propose a novel neural network model for joint part-of-speech (POS) tagging and dependency parsing. Our model extends the well-known BIST graph-based dependency parser (Kiperwasser and Goldberg, 2016) by incorporating a BiLSTM-based tagging component to produce automatically predicted POS tags for the parser. On the benchmark English Penn treebank, our model obtains strong UAS and LAS scores at 94.51% and 92.87%, respectively, producing 1.5+% absolute improvements to the BIST graph-based parser, and also obtaining a state-of-the-art POS tagging accuracy at 97.97%. Furthermore, experimental results on parsing 61 “big” Universal Dependencies treebanks from raw texts show that our model outperforms the baseline UDPipe (Straka and Strakova, 2017) with 0.8% higher average POS tagging score and 3.6% higher average LAS score. In addition, with our model, we also obtain state-of-the-art downstream task scores for biomedical event extraction and opinion analysis applications. Our code is available together with all pre-trained models at: https://github.com/datquocnguyen/jPTDP

We investigate the incorporation of character-based word representations into a standard CNN-based relation extraction model. We experiment with two common neural architectures, CNN and LSTM, to learn word vector representations from character embeddings. Through a task on the BioCreative-V CDR corpus, extracting relationships between chemicals and diseases, we show that models exploiting the character-based word representations improve on models that do not use this information, obtaining state-of-the-art result relative to previous neural approaches.

We compare the use of LSTM-based and CNN-based character-level word embeddings in BiLSTM-CRF models to approach chemical and disease named entity recognition (NER) tasks. Empirical results over the BioCreative V CDR corpus show that the use of either type of character-level word embeddings in conjunction with the BiLSTM-CRF models leads to comparable state-of-the-art performance. However, the models using CNN-based character-level word embeddings have a computational performance advantage, increasing training time over word-based models by 25% while the LSTM-based character-level word embeddings more than double the required training time.

Machine translation enables the automatic translation of textual documents between languages and can facilitate access to information only available in a given language for non-speakers of this language, e.g. research results presented in scientific publications. In this paper, we provide an overview of the Biomedical Translation shared task in the Workshop on Machine Translation (WMT) 2018, which specifically examined the performance of machine translation systems for biomedical texts. This year, we provided test sets of scientific publications from two sources (EDP and Medline) and for six language pairs (English with each of Chinese, French, German, Portuguese, Romanian and Spanish). We describe the development of the various test sets, the submissions that we received and the evaluations that we carried out. We obtained a total of 39 runs from six teams and some of this year’s BLEU scores were somewhat higher that last year’s, especially for teams that made use of biomedical resources or state-of-the-art MT algorithms (e.g. Transformer). Finally, our manual evaluation scored automatic translations higher than the reference translations for German and Spanish.

We describe SemEval–2017 Task 3 on Community Question Answering. This year, we reran the four subtasks from SemEval-2016: (A) Question–Comment Similarity, (B) Question–Question Similarity, (C) Question–External Comment Similarity, and (D) Rerank the correct answers for a new question in Arabic, providing all the data from 2015 and 2016 for training, and fresh data for testing. Additionally, we added a new subtask E in order to enable experimentation with Multi-domain Question Duplicate Detection in a larger-scale scenario, using StackExchange subforums. A total of 23 teams participated in the task, and submitted a total of 85 runs (36 primary and 49 contrastive) for subtasks A–D. Unfortunately, no teams participated in subtask E. A variety of approaches and features were used by the participating systems to address the different subtasks. The best systems achieved an official score (MAP) of 88.43, 47.22, 15.46, and 61.16 in subtasks A, B, C, and D, respectively. These scores are better than the baselines, especially for subtasks A–C.

Systems that locate mentions of concepts from ontologies in free text are known as ontology concept recognition systems. This paper describes an approach to the evaluation of the workings of ontology concept recognition systems through use of a structured test suite and presents a publicly available test suite for this purpose. It is built using the principles of descriptive linguistic fieldwork and of software testing. More broadly, we also seek to investigate what general principles might inform the construction of such test suites. The test suite was found to be effective in identifying performance errors in an ontology concept recognition system. The system could not recognize 2.1% of all canonical forms and no non-canonical forms at all. Regarding the question of general principles of test suite construction, we compared this test suite to a named entity recognition test suite constructor. We found that they had twenty features in total and that seven were shared between the two models, suggesting that there is a core of feature types that may be applicable to test suite construction for any similar type of application.