The field of natural language processing (NLP) has recently seen a large change towards using pre-trained language models for solving almost any task. Despite showing great improvements in benchmark datasets for various tasks, these models often perform sub-optimal in non-standard domains like the clinical domain where a large gap between pre-training documents and target documents is observed. In this paper, we aim at closing this gap with domain-specific training of the language model and we investigate its effect on a diverse set of downstream tasks and settings. We introduce the pre-trained CLIN-X (Clinical XLM-R) language models and show how CLIN-X outperforms other pre-trained transformer models by a large margin for ten clinical concept extraction tasks from two languages. In addition, we demonstrate how the transformer model can be further improved with our proposed task- and language-agnostic model architecture based on ensembles over random splits and cross-sentence context. Our studies in low-resource and transfer settings reveal stable model performance despite a lack of annotated data with improvements of up to 47 F1 points when only 250 labeled sentences are available. Our results highlight the importance of specialized language models as CLIN-X for concept extraction in non-standard domains, but also show that our task-agnostic model architecture is robust across the tested tasks and languages so that domain- or task-specific adaptations are not required.
Natural language inference (NLI) requires models to learn and apply commonsense knowledge. These reasoning abilities are particularly important for explainable NLI systems that generate a natural language explanation in addition to their label prediction. The integration of external knowledge has been shown to improve NLI systems, here we investigate whether it can also improve their explanation capabilities. For this, we investigate different sources of external knowledge and evaluate the performance of our models on in-domain data as well as on special transfer datasets that are designed to assess fine-grained reasoning capabilities. We find that different sources of knowledge have a different effect on reasoning abilities, for example, implicit knowledge stored in language models can hinder reasoning on numbers and negations. Finally, we conduct the largest and most fine-grained explainable NLI crowdsourcing study to date. It reveals that even large differences in automatic performance scores do neither reflect in human ratings of label, explanation, commonsense nor grammar correctness.
In this paper, we explore possible improvements of transformer models in a low-resource setting. In particular, we present our approaches to tackle the first two of three subtasks of the MEDDOPROF competition, i.e., the extraction and classification of job expressions in Spanish clinical texts. As neither language nor domain experts, we experiment with the multilingual XLM-R transformer model and tackle these low-resource information extraction tasks as sequence-labeling problems. We explore domain- and language-adaptive pretraining, transfer learning and strategic datasplits to boost the transformer model. Our results show strong improvements using these methods by up to 5.3 F1 points compared to a fine-tuned XLM-R model. Our best models achieve 83.2 and 79.3 F1 for the first two tasks, respectively.
When humans solve complex problems, they rarely come up with a decision right-away. Instead, they start with an intuitive decision, reflect upon it, spot mistakes, resolve contradictions and jump between different hypotheses. Thus, they create a sequence of ideas and follow a train of thought that ultimately reaches a conclusive decision. Contrary to this, today's neural classification models are mostly trained to map an input to one single and fixed output. In this paper, we investigate how we can give models the opportunity of a second, third and $k$-th thought. We take inspiration from Hegel's dialectics and propose a method that turns an existing classifier's class prediction (such as the image class forest) into a sequence of predictions (such as forest $\rightarrow$ tree $\rightarrow$ mushroom). Concretely, we propose a correction module that is trained to estimate the model's correctness as well as an iterative prediction update based on the prediction's gradient. Our approach results in a dynamic system over class probability distributions $\unicode{x2014}$ the thought flow. We evaluate our method on diverse datasets and tasks from computer vision and natural language processing. We observe surprisingly complex but intuitive behavior and demonstrate that our method (i) can correct misclassifications, (ii) strengthens model performance, (iii) is robust to high levels of adversarial attacks, (iv) can increase accuracy up to 4% in a label-distribution-shift setting and (iv) provides a tool for model interpretability that uncovers model knowledge which otherwise remains invisible in a single distribution prediction.
The performance of relation extraction models has increased considerably with the rise of neural networks. However, a key issue of neural relation extraction is robustness: the models do not scale well to long sentences with multiple entities and relations. In this work, we address this problem with an enriched attention mechanism. Attention allows the model to focus on parts of the input sentence that are relevant to relation extraction. We propose to enrich the attention function with features modeling knowledge about the relation arguments and the shortest dependency path between them. Thus, for different relation arguments, the model can pay attention to different parts of the sentence. Our model outperforms prior work using comparable setups on two popular benchmarks, and our analysis confirms that it indeed scales to long sentences with many entities.
In low-resource settings, model transfer can help to overcome a lack of labeled data for many tasks and domains. However, predicting useful transfer sources is a challenging problem, as even the most similar sources might lead to unexpected negative transfer results. Thus, ranking methods based on task and text similarity may not be sufficient to identify promising sources. To tackle this problem, we propose a method to automatically determine which and how many sources should be exploited. For this, we study the effects of model transfer on sequence labeling across various domains and tasks and show that our methods based on model similarity and support vector machines are able to predict promising sources, resulting in performance increases of up to 24 F1 points.
The recognition and normalization of clinical information, such as tumor morphology mentions, is an important, but complex process consisting of multiple subtasks. In this paper, we describe our system for the CANTEMIST shared task, which is able to extract, normalize and rank ICD codes from Spanish electronic health records using neural sequence labeling and parsing approaches with context-aware embeddings. Our best system achieves 85.3 F1, 76.7 F1, and 77.0 MAP for the three tasks, respectively.
Current developments in natural language processing offer challenges and opportunities for low-resource languages and domains. Deep neural networks are known for requiring large amounts of training data which might not be available in resource-lean scenarios. However, there is also a growing body of works to improve the performance in low-resource settings. Motivated by fundamental changes towards neural models and the currently popular pre-train and fine-tune paradigm, we give an overview of promising approaches for low-resource natural language processing. After a discussion about the definition of low-resource scenarios and the different dimensions of data availability, we then examine methods that enable learning when training data is sparse. This includes mechanisms to create additional labeled data like data augmentation and distant supervision as well as transfer learning settings that reduce the need for target supervision. The survey closes with a brief look into methods suggested in non-NLP machine learning communities, which might be beneficial for NLP in low-resource scenarios
Certain embedding types outperform others in different scenarios, e.g., subword-based embeddings can model rare words well and domain-specific embeddings can better represent in-domain terms. Therefore, recent works consider attention-based meta-embeddings to combine different embedding types. We demonstrate that these methods have two shortcomings: First, the attention weights are calculated without knowledge of word properties. Second, the different embedding types can form clusters in the common embedding space, preventing the computation of a meaningful average of different embeddings and thus, reducing performance. We propose to solve these problems by using feature-based meta-embeddings learned with adversarial training. Our experiments and analysis on sentence classification and sequence tagging tasks show that our approach is effective. We set the new state of the art on various datasets across languages and domains.