Despite significant strides in multimodal tasks, Multimodal Large Language Models (MLLMs) are plagued by the critical issue of hallucination. The reliable detection of such hallucinations in MLLMs has, therefore, become a vital aspect of model evaluation and the safeguarding of practical application deployment. Prior research in this domain has been constrained by a narrow focus on singular tasks, an inadequate range of hallucination categories addressed, and a lack of detailed granularity. In response to these challenges, our work expands the investigative horizons of hallucination detection. We present a novel meta-evaluation benchmark, MHaluBench, meticulously crafted to facilitate the evaluation of advancements in hallucination detection methods. Additionally, we unveil a novel unified multimodal hallucination detection framework, UNIHD, which leverages a suite of auxiliary tools to validate the occurrence of hallucinations robustly. We demonstrate the effectiveness of UNIHD through meticulous evaluation and comprehensive analysis. We also provide strategic insights on the application of specific tools for addressing various categories of hallucinations.

Many NLP tasks involve sentence-level annotation yet the relevant information is not encoded at sentence level but at some relevant parts of the sentence. Such tasks include but are not limited to: sentiment expression annotation, product feature annotation, and template annotation for Q&A systems. However, annotation of the full corpus sentence by sentence is resource intensive. In this paper, we propose an approach that iteratively extracts frequent parts of sentences for annotating, and compresses the set of sentences after each round of annotation. Our approach can also be used in preparing training sentences for binary classification (domain-related vs. noise, subjectivity vs. objectivity, etc.), assuming that sentence-type annotation can be predicted by annotation of the most relevant sub-sentences. Two experiments are performed to test our proposal and evaluated in terms of time saved and agreement of annotation.