Article Preview
TopIntroduction
Bangla is the 7th most spoken language in the world from 3500 languages and around 250 millions of people are using Bangla (Chowdhury, Khalil, & Mofazzal, 2000). It is the mother language of Bangladesh (Islam, 2003) and the second most spoken language in India (Olivet, 2015). Based on the economic survey - 2015, there are 62.30% literate people in Bangladesh where most of them are used to Bangla language only. Nowadays, many computerized contents are being developed in Bangla and online version of Bangla newspaper is also growing rapidly. So, electronic version of Bangla text is increasing without any bounds in the cyber world and people are overloaded with huge volume of texts. To alleviate this burden of large volume of text, very few research works have been conducted for Bangla (Islam & Masum, 2004; Sarkar, 2014; Haque, Pervin, & Begum, 2016). In this situation, the Bangla-speaking people have been deprived from the advantage of information technology. So, for this large community of Bangla-speaking people, more research work is very much necessary especially for Bangla information retrieval. But research works for Bangla language is difficult for the following issues:
- 1.
Based on our study, automatic procedures are hardly available for Bangla language to facilitate research work.
- 2.
For Bangla language, there is no lexical database like WordNet (Miller, 1995). Though a similar tool is ongoing to be developed it has limited features (Indian Statistical Institute, 2015).
- 3.
There is no database of ontological meanings for Bangla words that can be used programmatically.
- 4.
Subject and object of all sentences need to be identified for proper recognition of structures of sentences which is complex in Bangla than that of English. Because, the placement of subject in English sentence is generally before the verb phrase, auxiliary verb or it may appear after the word ‘by’ in passive voice but subject may be existed in several places in Bangla sentence.
Some other problems have been discussed in (Karim, Kaykobad, & Murshed, 2013; Zaman, 2015) about the research work on Bangla. Even, the scope of knowledge sharing is also limited as there are a few researchers in this arena. Despite these difficulties, a method has been presented here which focuses upon a problem in the output of Bangla information retrieval procedures.
In the output of information retrieval, some sentences may be available with dangling pronoun(s) where the corresponding noun is missing. These pronouns will make the information incoherent. So, the systems, which have been developed for burden minimization from large volume of text, may deliver wrong message. Other than receiving a direction, the user will often be misguided with misinformation.
The objective of this research work is to make the output of Bangla information retrieval procedures free from dangling pronoun. Otherwise, there is a huge probability to misunderstand the text by user because only a single dangling pronoun is enough to deliver wrong message. In these circumstances, a method has been proposed here to resolve this problem with the following major contributions:
- 1.
Detecting the nature of each word in the sentence of Bangla news document as noun, pronoun, verb, subject, object, numerical figure, acronym, name of people and places, etc. In this regard, words are tagged in two phases as general and special tagging.
- 2.
Dependency parsing of words to verify the nature of each word because a word may have dynamic nature for the effect of surrounding words. Some untagged words are also tagged using dependency parsing.
- 3.
Locating pronouns and distinguishing each of them as subject or object.
- 4.
Identifying the corresponding noun of the pronoun and replacing the pronoun in suitable format.