Instruction tuning has become an integral part of training pipelines for Large Language Models (LLMs) and has been shown to yield strong performance gains. In an orthogonal line of research, Annotation Error Detection (AED) has emerged as a tool for detecting quality problems in gold standard labels. So far, however, the application of AED methods has been limited to classification tasks. It is an open question how well AED methods generalize to language generation settings, which are becoming more widespread via LLMs. In this paper, we present a first and novel benchmark for AED on instruction tuning data: Donkii.It comprises three instruction-tuning datasets enriched with error annotations by experts and semi-automatic methods. We also provide a novel taxonomy of error types for instruction-tuning data.We find that all three datasets contain clear errors, which sometimes propagate directly into instruction-tuned LLMs. We propose four AED baselines for the generative setting and evaluate them extensively on the newly introduced dataset. Our results show that the choice of the right AED method and model size is indeed crucial and derive practical recommendations for how to use AED methods to clean instruction-tuning data.
Manually annotated datasets are crucial for training and evaluating Natural Language Processing models. However, recent work has discovered that even widely-used benchmark datasets contain a substantial number of erroneous annotations. This problem has been addressed with Annotation Error Detection (AED) models, which can flag such errors for human re-annotation. However, even though many of these AED methods assume a final curation step in which a human annotator decides whether the annotation is erroneous, they have been developed as static models without any human-in-the-loop component. In this work, we propose ActiveAED, an AED method that can detect errors more accurately by repeatedly querying a human for error corrections in its prediction loop. We evaluate ActiveAED on eight datasets spanning five different tasks and find that it leads to improvements over the state of the art on seven of them, with gains of up to six percentage points in average precision.
Automatic extraction of event structures from text is a promising way to extract important facts from the evergrowing amount of biomedical literature. We propose BEEDS, a new approach on how to mine event structures from PubMed based on a question-answering paradigm. Using a three-step pipeline comprising a document retriever, a document reader, and an entity normalizer, BEEDS is able to fully automatically extract event triples involving a query protein or gene and to store this information directly in a knowledge base. BEEDS applies a transformer-based architecture for event extraction and uses distant supervision to augment the scarce training data in event mining. In a knowledge base population setting, it outperforms a strong baseline in finding post-translational modification events consisting of enzyme-substrate-site triples while achieving competitive results in extracting binary relations consisting of protein-protein and protein-site interactions.
Large-scale language modeling and natural language prompting have demonstrated exciting capabilities for few and zero shot learning in NLP. However, translating these successes to specialized domains such as biomedicine remains challenging, due in part to biomedical NLP’s significant dataset debt – the technical costs associated with data that are not consistently documented or easily incorporated into popular machine learning frameworks at scale. To assess this debt, we crowdsourced curation of datasheets for 167 biomedical datasets. We find that only 13% of datasets are available via programmatic access and 30% lack any documentation on licensing and permitted reuse. Our dataset catalog is available at: https://tinyurl.com/bigbio22.
Deriving and modifying graphs from natural language text has become a versatile basis technology for information extraction with applications in many subfields, such as semantic parsing or knowledge graph construction. A recent work used this technique for modifying scene graphs (He et al. 2020), by first encoding the original graph and then generating the modified one based on this encoding. In this work, we show that we can considerably increase performance on this problem by phrasing it as graph extension instead of graph generation. We propose the first model for the resulting graph extension problem based on autoregressive sequence labelling. On three scene graph modification data sets, this formulation leads to improvements in accuracy over the state-of-the-art between 13 and 24 percentage points. Furthermore, we introduce a novel data set from the biomedical domain which has much larger linguistic variability and more complex graphs than the scene graph modification data sets. For this data set, the state-of-the art fails to generalize, while our model can produce meaningful predictions.
This paper describes our contribution for the MEDIQA-2021 Task 1 question summarization competition. We model the task as conditional generation problem. Our concrete pipeline performs a finetuning of the large pretrained generative transformers PEGASUS (Zhang et al.,2020a) and BART (Lewis et al.,2020). We used the resulting models as strong baselines and experimented with (i) integrating structured knowledge via entity embeddings, (ii) ensembling multiple generative models with the generator-discriminator framework and (iii) disentangling summarization and interrogative prediction to achieve further improvements. Our best performing model, a fine-tuned vanilla PEGASUS, reached the second place in the competition with an ROUGE-2-F1 score of 15.99. We observed that all of our additional measures hurt performance (up to 5.2 pp) on the official test set. In course of a post-hoc experimental analysis which uses a larger validation set results indicate slight performance improvements through the proposed extensions. However, further analysis is need to provide stronger evidence.
Biomedical event extraction from natural text is a challenging task as it searches for complex and often nested structures describing specific relationships between multiple molecular entities, such as genes, proteins, or cellular components. It usually is implemented by a complex pipeline of individual tools to solve the different relation extraction subtasks. We present an alternative approach where the detection of relationships between entities is described uniformly as questions, which are iteratively answered by a question answering (QA) system based on the domain-specific language model SciBERT. This model outperforms two strong baselines in two biomedical event extraction corpora in a Knowledge Base Population setting, and also achieves competitive performance in BioNLP challenge evaluation settings.
Rule-based models are attractive for various tasks because they inherently lead to interpretable and explainable decisions and can easily incorporate prior knowledge. However, such systems are difficult to apply to problems involving natural language, due to its large linguistic variability. In contrast, neural models can cope very well with ambiguity by learning distributed representations of words and their composition from data, but lead to models that are difficult to interpret. In this paper, we describe a model combining neural networks with logic programming in a novel manner for solving multi-hop reasoning tasks over natural language. Specifically, we propose to use an Prolog prover which we extend to utilize a similarity function over pretrained sentence encoders. We fine-tune the representations for the similarity function via backpropagation. This leads to a system that can apply rule-based reasoning to natural language, and induce domain-specific natural language rules from training data. We evaluate the proposed system on two different question answering tasks, showing that it outperforms two baselines – BiDAF (Seo et al., 2016a) and FastQA( Weissenborn et al., 2017) on a subset of the WikiHop corpus and achieves competitive results on the MedHop data set (Welbl et al., 2017).