Gender bias in vision-language models (VLMs) can reinforce harmful stereotypes and discrimination. In this paper, we focus on mitigating gender bias towards vision-language tasks. We identify object hallucination as the essence of gender bias in VLMs. Existing VLMs tend to focus on salient or familiar attributes in images but ignore contextualized nuances. Moreover, most VLMs rely on the co-occurrence between specific objects and gender attributes to infer the ignored features, ultimately resulting in gender bias. We propose GAMA, a task-agnostic generation framework to mitigate gender bias. GAMA consists of two stages: narrative generation and answer inference. During narrative generation, GAMA yields all-sided but gender-obfuscated narratives, which prevents premature concentration on localized image features, especially gender attributes. During answer inference, GAMA integrates the image, generated narrative, and a task-specific question prompt to infer answers for different vision-language tasks. This approach allows the model to rethink gender attributes and answers. We conduct extensive experiments on GAMA, demonstrating its debiasing and generalization ability.

Clinical text summarization has proven successful in generating concise and coherent summaries. However, these summaries may include unintended text with hallucinations, which can mislead clinicians and patients. Existing methods for mitigating hallucinations can be categorized into task-specific and task-agnostic approaches. Task-specific methods lack versatility for real-world applicability. Meanwhile, task-agnostic methods are not model-agnostic, so they require retraining for different models, resulting in considerable computational costs. To address these challenges, we propose MEDAL, a model-agnostic framework designed to post-process medical hallucinations. MEDAL can seamlessly integrate with any medical summarization model, requiring no additional computational overhead. MEDAL comprises a medical infilling model and a hallucination correction model. The infilling model generates non-factual summaries with common errors to train the correction model. The correction model is incorporated with a self-examination mechanism to activate its cognitive capability. We conduct comprehensive experiments using 11 widely accepted metrics on 7 baseline models across 3 medical text summarization tasks. MEDAL demonstrates superior performance in correcting hallucinations when applied to summaries generated by pre-trained language models and large language models.

In Visually-rich Document Understanding (VrDU), recent advances of incorporating layout and image features into the pre-training language models have achieved significant progress. Existing methods usually developed complicated dedicated architectures based on pre-trained models and fine-tuned them with costly high-quality data to eliminate the inconsistency of knowledge distribution between the pre-training task and specialized downstream tasks. However, due to their huge data demands, these methods are not suitable for few-shot settings, which are essential for quick applications with limited resources but few previous works are presented. To solve these problems, we propose a unified Knowledge-aware prompt-tuning framework for Visual-rich Document Understanding (KnowVrDU) to enable broad utilization for diverse concrete applications and reduce data requirements. To model heterogeneous VrDU structures without designing task-specific architectures, we propose to reformulate various VrDU tasks into a single question-answering format with task-specific prompts and train the pre-trained model with the parameter-efficient prompt tuning method. To bridge the knowledge gap between the pre-training task and specialized VrDU tasks without additional annotations, we propose a prompt knowledge integration mechanism to leverage external open-source knowledge bases. We conduct experiments on several benchmark datasets in few-shot settings and the results validate the effectiveness of our method.

Relational triple extraction is a critical task for natural language processing. Existing methods mainly focused on capturing semantic information, but suffered from ignoring the syntactic structure of the sentence, which is proved in the relation classification task to contain rich relational information. This is due to the absence of entity locations, which is the prerequisite for pruning noisy edges from the dependency tree, when extracting relational triples. In this paper, we propose a unified framework to tackle this challenge and incorporate syntactic information for relational triple extraction. First, we propose to automatically contract the dependency tree into a core relational topology and eliminate redundant information with graph pooling operations. Then, we propose a symmetrical expanding path with graph unpooling operations to fuse the contracted core syntactic interactions with the original sentence context. We also propose a bipartite graph matching objective function to capture the reflections between the core topology and golden relational facts. Since our model shares similar contracting and expanding paths with encoder-decoder models like U-Net, we name our model as Relation U-Net (RelU-Net). We conduct experiments on several datasets and the results prove the effectiveness of our method.

Relational triple extraction is a critical task for constructing knowledge graphs. Existing methods focused on learning text patterns from explicit relational mentions. However, they usually suffered from ignoring relational reasoning patterns, thus failed to extract the implicitly implied triples. Fortunately, the graph structure of a sentence’s relational triples can help find multi-hop reasoning paths. Moreover, the type inference logic through the paths can be captured with the sentence’s supplementary relational expressions that represent the real-world conceptual meanings of the paths’ composite relations. In this paper, we propose a unified framework to learn the relational reasoning patterns for this task. To identify multi-hop reasoning paths, we construct a relational graph from the sentence (text-to-graph generation) and apply multi-layer graph convolutions to it. To capture the relation type inference logic of the paths, we propose to understand the unlabeled conceptual expressions by reconstructing the sentence from the relational graph (graph-to-text generation) in a self-supervised manner. Experimental results on several benchmark datasets demonstrate the effectiveness of our method.

Relational triple extraction is a crucial task for knowledge graph construction. Existing methods mainly focused on explicit relational triples that are directly expressed, but usually suffer from ignoring implicit triples that lack explicit expressions. This will lead to serious incompleteness of the constructed knowledge graphs. Fortunately, other triples in the sentence provide supplementary information for discovering entity pairs that may have implicit relations. Also, the relation types between the implicitly connected entity pairs can be identified with relational reasoning patterns in the real world. In this paper, we propose a unified framework to jointly extract explicit and implicit relational triples. To explore entity pairs that may be implicitly connected by relations, we propose a binary pointer network to extract overlapping relational triples relevant to each word sequentially and retain the information of previously extracted triples in an external memory. To infer the relation types of implicit relational triples, we propose to introduce real-world relational reasoning patterns in our model and capture these patterns with a relation network. We conduct experiments on several benchmark datasets, and the results prove the validity of our method.

Industrial dialogue systems such as Apple Siri and Google Now rely on large scale diverse and robust training data to enable their sophisticated conversation capability. Crowdsourcing provides a scalable and inexpensive way of data collection but collecting high quality data efficiently requires thoughtful orchestration of the crowdsourcing jobs. Prior study of this topic have focused on tasks only in the academia settings with limited scope or only provide intrinsic dataset analysis, lacking indication on how it affects the trained model performance. In this paper, we present a study of crowdsourcing methods for a user intent classification task in our deployed dialogue system. Our task requires classification of 47 possible user intents and contains many intent pairs with subtle differences. We consider different crowdsourcing job types and job prompts and analyze quantitatively the quality of the collected data and the downstream model performance on a test set of real user queries from production logs. Our observation provides insights into designing efficient crowdsourcing jobs and provide recommendations for future dialogue system data collection process.