Corpus generation device and method, human-machine interaction system

A corpus generation method, comprising the steps of:

• segmenting each sentence in at least one monolingual parallel corpus into words, and tagging the segmented words by a knowledge-driven approach;
• identifying the knowledge-driven sentence, and classifying sentences having different tag sequences but the same meaning into the same sentence cluster;
• analyzing the sentences in each sentence cluster of each monolingual parallel corpus, determining the categories of sentence structures of all the sentences in the sentence cluster, determining, recording and storing a mapping mode for transforming tags between corresponding sentence structures when different categories of sentence structures in the same sentence cluster are transformed;
• searching for a first category of sentence structures in each sentence cluster of all the monolingual parallel corpora, and mapping, according to a first mapping mode between the first category of sentence structures in one of the sentence clusters and other categories of sentence structures in the same sentence cluster, the first category of sentence structures in other sentence clusters in accordance with the mapping mode to generate corresponding categories of sentence structures;
• and
• nesting a word corresponding to the sequence tag of a sentence structure in the sentence cluster into the generated new sentence structures to generate a new monolingual parallel corpus.

A human-machine interaction system, comprising a microphone, an analog-to-digital converter, a voice identification processor, a natural language processor, a voice synthesizer, a power amplifier, a loudspeaker, a corpus server, and a corpus generation processor, wherein a plurality of monolingual parallel corpora are stored in the corpus server, wherein:

• the microphone, the analog-to-digital converter, the voice identification processor, the natural language processor, the voice synthesizer, the power amplifier and the loudspeaker are sequentially connected;
• the corpus server is connected to the natural language processor;
• and the corpus generation processor is connected to the corpus server;
• the microphone is configured to acquire a voice signal from a user when the user and a robot are dialoging;
• the analog-to-digital converter is configured to convert the voice signal into voice digital information;
• the voice identification processor is configured to convert the voice digital information into text information, and input the text information into the natural language processor;
• the natural language processor is configured to analyze and determine the text information, generate an appropriate reply message by combining the corpora in the corpus server, transmit the reply message to the voice synthesizer, and store the text information and the reply message in the corpus server;
• the voice synthesizer is configured to generate a voice signal according to the reply message, amplify the voice signal via the power amplifier, and play the amplified voice signal via the loudspeaker;
• and
• the corpus generation processor is configured to execute programs to realize the following steps:
• segmenting each sentence in at least one monolingual parallel corpus into words, and tagging the segmented words by a knowledge-driven approach;
• identifying the knowledge-driven sentence, and classifying sentences having different tag sequences but the same meaning into the same sentence cluster;
• analyzing the sentences in each sentence cluster of each monolingual parallel corpus, determining the categories of sentence structures of all the sentences in the sentence cluster, determining, recording and storing a mapping mode for transforming tags between corresponding sentence structures when different categories of sentence structures in the same sentence cluster are transformed;
• searching for a first category of sentence structures in each sentence cluster of all the monolingual parallel corpora, and mapping, according to a first mapping mode between the first category of sentence structures in one of the sentence clusters and other categories of sentence structures in the same sentence cluster, the first category of sentence structures in other sentence clusters in accordance with the mapping mode to generate corresponding categories of sentence structures;
• and
• nesting a word corresponding to the sequence tag of a sentence structure in the sentence cluster into the generated new sentence structures to generate a new monolingual parallel corpus, and storing the new monolingual parallel corpus in the corpus server.