Multilayer Neural Network Technique for Parsing the Natural Language Sentences

Multilayer Neural Network Technique for Parsing the Natural Language Sentences

Manu Pratap Singh (Dr. Bhimrao Ambedkar University, Agra, Uttar Pradesh, India), Sukrati Chaturvedi (Department of Physics and Computer Science, DEI, Agra, Uttar Pradesh, India) and Deepak D. Shudhalwar (PSS Central Institute of Vocational Education, NCERT, Bhopal, Madhya Pradesh, India)
DOI: 10.4018/IJAIML.2019070102

Abstract

In this article is presented an approach for parsing natural language sentences using neural networks. The pre-processing technique is applied to code the sentences into string of bits and after the training process is started, is formed into patterns available in the form of coded information. The multilayer feed forward networks are used here for training to classify the words into appropriate syntactical categories. The classified words represent the parsed information of the given sentences. The main function of the network is to assign the respective syntactical categories to each word of a sentence with a minimal error rate. The comparison between the two popular neural network approaches i.e. feed forward neural network and radial basis neural network is presented to analyze performance for the new and unknown sentences.
Article Preview
Top

1. Introduction

As we know that the language as spoken or used by native speakers in a day to day communication is known as the natural language. The natural language includes lots of ambiguities in the way. It is required to have common sense, reasoning capacity and experience to understand the natural language. This is the reason that humans are good to handle natural languages and computers, which lack common sense knowledge and reasoning capacity are poor in understanding. Their automatic understanding of natural language is a very difficult task for the reason that a natural language is intrinsically complicated and ambiguous (Marchesi et al., 1996). In spite of this, automatic understanding of natural language is always remained as the trust area in the computer science research as natural language processing.

Natural language Processing (NLP) is the field of study that deals with the interactions between human language and computers understanding for the natural language sentences. NLP systems have useful roles, such as converting speech to text, grammar correction and automatically translating between languages. Natural language systems take strings of words (sentences) as their input and produce structured representations capturing the meaning of those as their output. The nature of this output depends heavily on the task at hand (Archambault 1994). The natural sentences are mostly acquired by the machines as parsing tree.

In the linguistic context, parsing is the analysis of the relationship between parts as the words in a sentence. Thus, automatic parsing of natural language is an important task for many NLP applications. A number of approaches have been successfully applied to automatic parsing techniques like symbol, statistical connectionist etc. Symbolic approaches perform deep analysis of linguistic phenomenon and are based on explicit representation knowledge representation, schemes, and associated algorithms (Appolini et al., 1992). Symbolic approaches have been used for a decade in a variety of research areas and applications such as information extraction, text categorization, ambiguity resolution and lexical acquisition (Archambault 1994). Common methods for automatic parse trees those have been used are probabilistic grammars (Collobert et al., 2011) and sparse feature learning models such as perceptron (Skorzewski 2010). The parsing step has been divided into many steps; lexical analysis which highlights the basic constituents of a phrase, syntactic analysis which finds out the syntactic categories (noun, verb, adjective etc.) o0f such constituents and semantic analysis which tries to catch the meaning of the phrase often contributing to its disambiguation (Goldberg & Elhabad 2013). The most popular approach to parse natural language is to use the grammar which is able to describe linguistic rules, complemented by further rules which are able to disambiguate the meaning of words or sub phrases. Such rules can be applied by using an expert system approach, or using a pattern matching with the aid of a database of phrases already translated (Beardon & Holmes 1991).

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 10: 2 Issues (2020): 1 Released, 1 Forthcoming
Volume 9: 2 Issues (2019)
View Complete Journal Contents Listing