基于听感知的 音频信息隐藏研究 - VideoLecturesNET.PPT

Given an input character sequence, it is encoded by a finite state acceptor FSAinput. For example, the input “合成分子时 ”(while synthesizing molecule) is represented as Figure a. Figure b shows a toy dictionary with 6 words, which is represented by a transducer with empty weights We adopt an n-gram language model based on word classes for word segmentation. Figure c shows a toy bigram with three words. The word classes in defined in the table. For both POS tagging and named entity recognition, the Hidden Markov Models are used. Each HMM is represented with two WFSTs. Taking the POS tagging as an example, Figure a models the generation of words by POS. Figure b models the transitions between POS tags. For named entity recognition, the topology of WFST can be obtained by substituting the POS tags with 30 named entity role tags such as surname, the first character of a given name with two characters, the first character of a location name. Some rules, which can be encoded into WFSTs, are responsible for the transformation from a role sequence to corresponding name class. * 1m * Xinhao Wang, Jiazhong Nie, Dingsheng Luo, and Xihong Wu Speech and Hearing Research Center, Department of Machine Intelligence, Peking University September 18th, 2008 A Joint Segmenting and Labeling Approach for Chinese Lexical Analysis ECML PKDD 2008, Antwerp Cascaded Subtasks in NLP Chunking and Parsing Word Segmentation and Named Entity Recognition POS Tagging Word Sense Disambiguation Drawbacks: Errors introduced by earlier subtasks propagate through the pipeline and will never be recovered in downstream subtasks. The information sharing among different subtasks is prohibited by this pipeline manner. Researchers’ Efforts on Joint Processing Reranking (Shi, 2007; Sutton, 2005; Zhang, 2003) As an approximation of joint processing, it may miss the true optimal result, which often lies out of the k-best list. Taking multiple subtasks as a single one (Luo, 2003; Miller, 2000; Yi, 2005; Nakaga