[Heavy Report] Service Robot Core Technology and Module Analysis (2)

Third, the interaction module: voice reaches the commercial threshold, semantic understanding needs to be improved

1 Intelligent voice technology has reached the commercial threshold

The development of speech semantics has gone through three stages, and the progress of the rules has been minimal. The first stage of the statistical period broke out, and deep learning was the second outbreak. From the 1950s to the 1970s, the rule of speech recognition was dominated by rules. The bottleneck could not break the slow development. The recognition of hundreds of words in IBM was 70%. From the 1970s to the end of the 20th century, rapid development, statistics and regular struggles, and Gradually solve the problems of speech recognition, part of speech analysis and syntactic analysis; at the beginning of the 21st century, due to the breakthrough in computing power enhancement speech technology, from 2006 to now, deep learning continues to be perfect in the field of speech recognition.

   

2 Semantic understanding still takes time, waiting for deep learning algorithm breakthrough

Natural Language Processing (NLP): The lexical and syntactical solutions are basically solved, and the semantics are currently only shallow processing. NLP analysis techniques are broadly divided into three levels: lexical analysis, syntactic analysis, and semantic analysis.

1) Lexical analysis

Lexical analysis includes word segmentation, part-of-speech tagging, named entity recognition, and word sense disambiguation. Word segmentation and part-of-speech tagging are well understood. The task of identifying a named entity is to identify named entities such as person names, place names, and institution names in the sentence. Each named entity is made up of one or more words. Word sense disambiguation is to judge the true meaning of each or some words according to the context of the sentence.

2) Syntactic analysis

Syntactic analysis is to change the input sentence from sequence form to tree structure, so that it can capture the collocation or modification relationship between words within the sentence. This step is a key step in NLP. At present, there are two mainstream syntactic analysis methods in the research community: the phrase structure syntax system and the dependency structure syntax system. The dependency syntax system has now become a hot topic in the study of syntactic analysis. Dependent grammatical representations are concise, easy to understand and annotate, which can easily represent the semantic relationship between words, such as the relationship between sentence components can constitute things, things, time and so on. This semantic relationship makes it easy to apply fish semantic analysis and information extraction. Dependencies can also implement decoding algorithms more efficiently. The syntactic structure obtained by syntactic analysis can help the semantic analysis of the upper layer, as well as some applications, such as machine translation, question and answer, text mining, information retrieval and so on.

3) Semantic analysis

The ultimate goal of semantic analysis is to understand the true semantics of sentence expression. But what form to use to represent semantics has not been well resolved. Semantic role labeling is a relatively mature shallow semantic analysis technique. Given a predicate in a sentence, the task of semantic role labeling is to mark the parameters of the predicate's acting, receiving, time, and location from the sentence. Semantic role labeling is generally done on the basis of syntactic analysis, and the syntactic structure is crucial for the performance of semantic role labeling.

Difficulties in natural language processing: word sense disambiguation is the bottleneck, and Chinese is more difficult than English. One: cutting words, Chinese and English natural language processing has a first step, which is to decompose the input string into lexical units; second: word class annotation; third: grammar theory; fourth: word sense disambiguation.

CPG Generator 2 Series:power Range1200KWe-3900KWe/50&60HZ

Generator,1200KWe-3900KWe

CCSN POWER GENERATION INC.(Engine is a subsidiary of CCSN) , https://www.ccsnengine.com