Detecting completeness in natural language (NL) requirements traditionally involves domain modeling to establish reference models, which requires substantial manual effort and proves to be time- and labor-intensive. Large Language Models (LLMs), with an extensive knowledge base and advanced capabilities for contextual understanding and reasoning, present an opportunity for automated completeness detection in NL requirements. To better complete this task, we first propose the CMD model for implied completeness, which formally characterizes the essential nature of external completeness in NL requirements. We further establish formal definitions for requirements compositions and incompleteness dimensions, providing a qualitative analytical foundation for completeness detection. We then present ReCompGPT (Requirements Completing GPT), an LLM-powered method for detecting and repairing requirements completeness. Grounded in the proposed definitions, we construct a fine-grained chain-of-thought augmented by an action diffusion approach that provides the model with multi-perspective insights into requirements. It eliminates the need for a reference model, verifying the completeness of requirements specifications from natural language to natural language (NL2NL). Simultaneously, we develop a new dataset dedicated to detecting the external completeness in NL requirements, on which our method achieves an overall detection success rate of 93.8%, significantly enhancing performance relative to the vanilla prompts and the vanilla chain-of-thought, respectively. Overall, ReCompGPT demonstrates superior capabilities in both simple scenarios and complex cases characterized by implied requirements incompleteness. Our method proves to be effective and flexible.

Named entity recognition (NER) is an essential subtask in natural language processing field. Recent studies have demonstrated that character-word lattice models are efficient for Chinese NER, which can leverage useful word boundary information to enhance the representation of characters. However, previous models only consider the integration of local matched word features and neglect the semantic interactions with long-range matched words. Moreover, prior methods solely achieve superficial fusion in the character-word feature space with simple methods, such as feature concatenation, but fail to implement fine-grained semantic fusion. In this paper, we propose a mixed-lattice self-attention network (MLSAN) to integrate richer word boundary information, which can explicitly capture the fine-grained correlations across characters and long-range matched words and achieve the integration of global word features. In addition, we design an end-to-end method for incorporating lexical information at the bottom layer of BERT. Compared with existing methods, our model achieves a deeper lexical knowledge fusion, which makes MLSAN perform well on cases of a small number of samples. Experimental results on four Chinese NER datasets show that our model obtains competitive performance.