Abstract

This paper revisits the work of (Kuncham et al., 2015) which developed a statistical sandhi splitter (SSS) for agglutinative languages that was tested for Telugu and Malayalam languages. Handling compound words is a major challenge for Natural Language Processing (NLP) applications for agglutinative languages. Hence, in this paper we concentrate on testing the effect of SSS on the NLP applications like Machine Translation, Dialogue System and Anaphora Resolution and show that the accuracy of these applications is consistently improved by using SSS. We shall also discuss in detail the performance of SSS on these applications.

Abstract

The challenge of anaphora resolution has been taken up from long time. However, most of the work did not include for dialogues. In this paper we discuss the types of pronouns and anaphora in Telugu language and make an attempt to build a rule based pronominal anaphora resolution algorithm for human to human conversations. The model mainly consists of two parts, creating a knowledge base with a set of pronouns along with its morphological information and designing an algorithm which uses this knowledge base to give an output. In this process we have worked on normal pronominal anaphora and suggested a set of rules applicable for Telugu dialogues. However, since there was no corpus for the Telugu language, we built a corpus and tested the algorithm on it. Results show that the suggested algorithm produced an output with an accuracy of 61.1%.

Abstract

Sandhi splitting is a primary and an important step for any natural language processing (NLP) application for languages which have agglutinative morphology. This paper presents a statistical approach to build a sandhi splitter for agglutinative languages. The input to the model is a valid string in the language and the output is a split of that string into meaningful word/s. The approach adopted comprises of two stages namely Segmentation and Word generation, both of which use conditional random fields (CRFs). Our approach is robust and language independent. The results for two Dravidian languages viz. Telugu and Malayalam show an accuracy of 89.07% and 90.50% respectively.

Abstract

This paper discusses the handling of multisentence queries in a mixed-initiative dialogue system based on a hierarchically structured knowledge base, in a way that is domain independent. The system is rule-based and uses dependency relations and part-of-speech tags obtained from the Stanford Parser coupled with the hierarchical structure of the knowledge base to identify the user’s goal. The system was tested for its accuracy over answering questions, and also subjective testing was done to evaluate the dialogue flow; primarily over the books domain. We show examples of the system developed over the domains of books, movies and restaurants to demonstrate the domain independence.

Abstract

Earlier works has explored that a foreign language can be learnt from pictures than merely using native language word translations. These efforts on language learning using pictures were limited to colors, animals, birds, vehicles and common noun categories. In this paper, we present our research that learning of verbs is also possible by using right set of pictures and gestures, and we show that this is effective in second language learning. We report the acquisition of words in tourist domain in a second language by working with subjects who are between 14 and 48 years of age. From pre-learning and post-learning evaluations, we show that acquisition of vocabulary like nouns and verbs in a new language is better with the fusion of pictures and voice. We also show the subjects are able to generalize their learning towards phrase-level vocabulary without any additional training and efforts.

Abstract

Statistical morph analyzers have proved to be highly accurate while being comparatively easier to maintain than rule based approaches. Our morph analyzer (SMA++) is an improvement over the statistical morph analyzer (SMA) described in Malladi and Mannem (2013). SMA++ predicts the gender, number, person, case (GNPC) and the lemma (L) of a given token. We modified the SMA in Malladi and Mannem (2013), by adding some rich machine learning features. The feature set was chosen specifically to suit the characteristics of Indian Languages. In this paper we apply SMA++ to four Indian languages viz. Hindi, Urdu, Telugu and Tamil. Hindi and Urdu belong to the Indic1 language family. Telugu and Tamil belong to the Dravidian2 language family. We compare SMA++ with some state-of-art statistical morph analyzers viz. Morfette in Chrupała et al. (2008) and SMA in Malladi and Mannem (2013). In all four languages, our system performs better than the above mentioned state-of-art SMAs

Abstract

This paper proposes a novel approach to capture the concepts1 of an NL query. Given an NL query, the query is mapped to a tagset, which carries the concept information. The tagset was created by mapping every noun chunk to the attribute of a table (tableName.attributeName) and every verb chunk to a relation in the ER schema. The approach is discussed using the Courses Management domain of a University and can be extended to other domains. The tagset here was formed using the ER-schema of the Courses Management Portal of our university. We used the statistical approach to identify the concepts. We ourselves formed a tagged corpus with different types of NL queries. Conditional Random Field algorithm was used for the classification. The results are very promising and are compared to the rule based approach seen in Gupta et al. (2012) [1] .

Abstract

In this paper we explain the identification of karaka relations in an English sentence. We explain the genesis of the problem and present two different approaches, rule based and statistical. We briefly describe about rule based and focus more on statistical approach. We process a sentence through various stages and extract features at each stage. We train our data and identify Karaka relations using Support Vector Machines (SVM). We also explain the impact of our work on Natural Language Interfaces for Database systems.

Abstract

Online debate forums provide a rich collection of differing opinions on various topics. In dual-sided debates, users present their opinions or judge other’s opinions to support their stance. In this paper, we examine the use of users’ intentions and debate structure for stance classification of the debate posts. We propose a domain independent approach to capture users’ intent at sentence level using its dependency parse and sentiWordNet and to build the intention structure of the post to identify its stance. To aid the task of classification, we define the health of the debate structure and show that maximizing its value leads to better stance classification accuracies.

Abstract

Social networking sites provide users a virtual community interaction platform to share their thoughts, life experiences and opinions. Online debate forum is one such platform where people can take a stance and argue in support or opposition of debate topics. An important feature of such forums is that, they are dynamic and increase rapidly. In such situations, e↵ective opinion summarization approaches are needed so that readers need not go through the entire debate. This paper aims to summarize online debates by extracting highly topic relevant and sentiment rich sentences. The proposed approach takes into account topic relevant, document relevant and sentiment based features to capture topic opinionated sentences. ROUGE scores are used to evaluate our system. Our system significantly outperforms several baseline systems and show 5.2% (ROUGE-1), 7.3%(ROUGE-2) and 5.5% (ROUGE-L) improvement over the state-of-the-art opinion summarization system. The results verify that topic directed sentiment features are most important to generate e↵ective debate summaries.