NFLAT: Non-Flat-Lattice Transformer for Chinese Named Entity Recognition

Recently, FLAT has achieved great success in Chinese Named Entity Recognition (NER). This method achieves lexical enhancement by constructing a flat lattice, which mitigates the difficulties posed by blurred word boundaries and the lack of word semantics. To this end, FLAT uses the position information of the starting and ending characters to connect the matching words. However, this method is likely to match more words when dealing with long texts, resulting in very long input sequences. Therefore, it increases the memory used by self-attention and computational costs. To deal with this issue, we advocate a novel lexical enhancement method, InterFormer, that effectively reduces the amount of computational and memory costs by constructing the non-flat-lattice. Furthermore, we implement a complete model, namely NFLAT, for the Chinese NER task. NFLAT decouples lexicon fusion and context feature encoding. Compared with FLAT, it reduces unnecessary attention calculations in "word-character" and "word-word". This reduces the memory usage by about 50\% and can use more extensive lexicons or higher batches for network training. The experimental results obtained on several well-known benchmarks demonstrate the superiority of the proposed method over the state-of-the-art character-word hybrid models.