Search

Word embeddings are used to convert human language into a numerical form by encoding the semantic properties of words. Using it each word can be transformed to a set of N-dimensional vectors. It plays a vital role in processing of linguistic applications like natural language inference, information retrieval, sentiment analysis, etc. The goal of word embedding is to capture the meaning of words in their context. And it also find the semantic relationships and similarities between words. The aim of this work is to summarize the existing embedding techniques for words and available corpus for Malayalam language. Since Malayalam is a resource-constrained Indian language, this paper is expected to help NLP researchers in Malayalam to identify the existing resources and to improve the current research trend.

Reation Extraction(RE) is the subprocess of Information Extraction(IE) which focuses on determining and extracting the reation between two participating entities. Most of the past work focus on extracting relations within a sentence. Nowadays, research on relation extraction focuses on identifying and determining relationship between participating entities across sentences. This paper proposes a bi-directional GRU model with self attention mechanism for inter-sentential relation extraction. First, a bi-directional GRU with self attention mechanism is used to capture the information about the relation from intermediary terms between two entities. Then a bi-directional GRU is used to capture the information represented by entities, which plays a vital role in relation extraction. Finally, the proposed model combines both word embeddings and entity embeddings for extracting a relation. Experimental results show that the proposed Bi-directional GRU model can deliver state-of-the-art results on relation classification. Application of self attention mechanism on intermediary terms improves the performance of relation extraction. Experimental results show that F-measure of the proposed inter-sentential relation extraction is 0.75, which is better than state-of-the-art systems of inter-sentential relation extraction with the same dataset.

Most of the past work on relation extraction(RE) has focused on identifying relationships between entities within a sentence. Nowadays, most of the research in the field of RE has got interested in relation extraction between entity pairs across sentence boundaries. This paper proposes a Bi-directional LSTM model for for inter-sentential RE. Experimental results show that the proposed Bi-LSTM model can achieve better results on relation classification by capturing the information hidden in long-distance relation patterns.

The goal of an information retrieval system is to provide the information that is relevant to the user's query. In some cases the information relevant to the user request may not exist in the user's native language. Situations may also arise where the user is able to read documents in languages different from the native one, but might have difficulty in formulating queries in them. The main intention behind Multilingual Information Retrieval is to find the relevant information available irrespective of the language used in the query.

The morphological richness and the agglutinative nature of Malayalam make it necessary to retrieve the root word from its inflected form in most of the NLP tasks. This paper presents an approach to identify the suffixes of Malayalam words using MBLP approach. The idea here is to use Memory Based Language Processing (MBLP) algorithm for Malayalam suffix identification. MBLP is an approach to language processing based on exemplar storage during learning and analogical reasoning during processing. Sandhi splitting is essential for morphological analysis, document indexing and topic modeling. Suffix separation improves the quality of machine translated text. Training instances created from words are manually annotated for their segmentation and the system is trained using TiMBL (Tilberg Memory Based Learner). The paper presents memory-based model of Malayalam suffix identification and its generalization accuracy.

Word Sense Disambiguation (WSD) is the task of identifying the correct sense of a word in a specific context when the word has multiple meaning. WSD is very important as an intermediate step in many Natural Language Processing (NLP) tasks, especially in Information Extraction(IE), Machine Translation(MT) and Question/Answering Systems. Word sense ambiguity arises when a particular word has more than one possible sense. The peculiarity of any language is that it includes a lot of ambiguous words. Since the sense of a word depends on its context of use, disambiguation process requires the understanding of word knowledge. Automatic WSD systems are available for structured languages like English, Chinese, etc. But Indian languages are morphologically rich and thus the processing task is very complex. The aim of this work is to develop a WSD system for Malayalam, a language spoken in India, predominantly used in the state of Kerala. The proposed system uses a corpus which is collected from various Malayalam web documents. For each possible sense of the ambiguous word, a relatively small set of training examples (seed sets) are identified which represents the sense. Collocations and most co-occurring words are considered as training examples. Seed set expansion module extends the seed set by adding most similar words to the seed set elements. These extended sets act as sense clusters. The most similar sense cluster to the input text context is considered as the sense of the target word.

The information contained in the document can be retrieved from its most significant paragraph, rather than by reading the whole document. The proposed work ranks the paragraphs of a text document using eigen analysis and returns the most important paragraph of a document. The importance of each paragraph is determined based on the correlation between the paragraphs. The proposed method explores the use of fuzzy graphs in capturing the inter-paragraph correlation of text documents. This approach models the document as a fuzzy graph where a node refers to a paragraph and an edge indicates the relationship between the paragraphs. The correlation between paragraphs is measured by extracting their semantic similarity. The importance of each node is determined based on this correlation. Subsequently the system ranks the paragraphs according to their importance. The proposed system is evaluated using DUC 2001 data set. The ROUGE scores show that the significant paragraph suggested by the proposed method covers relatively a good amount of relevant information in the document.

As the global energy consumption is rising dramatically, wind energy is a prominent one among the renewable energy sources. The penetration of wind energy into grid is increasing day by day. In order to carry huge amount of wind power during the grid integration of large scale wind farms, high transmission line capability is demanded. In order to improve the power carrying capability of the transmission line and to improve the stability of the system, series compensation is the best practical solution. Series compensation can result in Sub-Synchronous Resonance (SSR) oscillations in the electrical system which will lead to damages in the system such as shaft failure. In this paper, a novel idea of using the Real Time Recurrent Learning (RTRL) based adaptive neuro controller is proposed for damping SSR oscillations in grid connected windfarms. The controller is trained in real time without a reference model. The effectiveness of the proposed controller is tested under varying series compensation, wind speeds and grid impedance conditions and it has been proved that the proposed controller performs far better than any other linear controllers.

Question answering (QA) system aims at retrieving precise information from a large collection of documents against a query. This paper describes the architecture of a Natural Language Question Answering (NLQA) system for a specific domain based on the ontological information, a step towards semantic web question answering. The proposed architecture defines four basic modules suitable for enhancing current QA capabilities with the ability of processing complex questions. The first module was the question processing, which analyses and classifies the question and also reformulates the user query. The second module allows the process of retrieving the relevant documents. The next module processes the retrieved documents, and the last module performs the extraction and generation of a response. Natural language processing techniques are used for processing the question and documents and also for answer extraction. Ontology and domain knowledge are used for reformulating queries and identifying the relations. The aim of the system is to generate short and specific answer to the question that is asked in the natural language in a specific domain. We have achieved 94 % accuracy of natural language question answering in our implementation.

Words are the important building blocks of every language. Morphological generator is used to get the inflected form of a word, given its root word and a set of properties such as lexical category and morphological properties. Morphological Generation and analysis are necessary for developing computational grammars as well as machine translation systems. This paper presents a morphological generator for Malayalam nouns using Memory Based Language Processing (MBLP) approach. MBLP is an approach to language processing based on exemplar storage during learning, and analogical reasoning during processing. For training the system, a training corpus is created. It contains the basic examples of root words and their features. The feature set for this Malayalam noun generation system includes number, case, and the last syllable of the root word. Tilburg Memory based Learner (TiMBL) is used for training the system. The system doesn't require a dictionary or rules for its working. It gives a satisfactory result, having an accuracy of 93.68%

Text chunking consists of dividing a text into syntactically correlated parts of words. Given the words and their morphosyntactic class, a chunker will decide which words can be grouped as chunks. Malayalam is a free word order language and has relatively unrestricted phrase structures that make the problem of chunking quite challenging. This paper aims to develop a text chunker for Malayalam using Memory-Based Learning (MBL) approach. Memory-Based Learning is a machine learning methodology based on the idea that the direct reuse of examples using analogical reasoning is more suited for solving language processing problems than the application of rules extracted from those examples. The chunker was trained using the tool Memory-Based Tagger (MBT) with words and their POS tags as features. The chunker demonstrated an accuracy of 97.14%.

Transliteration forms an essential part of transcription which converts text from one writing system to another. The need for translating data has become larger than before as the world is getting together through social media. Machine transliteration has emerged as a part of information retrieval and machine translation projects to translate named entities, that are not registered in the dictionary, based on phonemes and graphemes. This paper proposes a machine learning technique that performs transliteration from Tamil to Malayalam, two languages that belong to Dravidian family. Transliteration can be used to supplement machine translation process by handling the issues that can happen due to the presence of named entities.

In view of the exponential growth of online document corpora, even perfect retrieval will fetch too much material for a user to cope with. One way to reduce this problem is automatic domain-specific summarization tailored to user's needs, which is a kind of high-level data cleaning. This requires some method of discovering classes of similar items that may be grouped into predetermined domains. We explore whether there exists a synergic relation between systems for classification and those for summarization by way of composing those subsystems. In other words, we examine whether prior summarization will increase the performance of the classifier system and vice versa. In both cases, the answer is affirmative, as we show in this paper. We propose a text-mining framework in which these subsystems are treated as constituents of a knowledge discovery process for text corpora.

This paper presents an approach to morphological analysis of Malayalam words as a classification Problem. The idea here is to use Memory Based Language Processing (MBLP) algorithm for Malayalam morphological analysis. MBLP is an approach to language processing based on exemplar storage during learning, and analogical reasoning during processing. The aim of the system is to find the citation forms (or meaningful parts) of words rather than a detailed morphological analysis. Training instances created from words are manually annotated for their segmentation and the system is trained using TiMBL (Tilburg Memory based Learner). The paper presents memory based model of Malayalam morphological analysis and its generalization accuracy.

In this paper, Sentence level Sentiment Analysis of Malayalam movie reviews is done by classifying the polarity of opinions obtained from the user as positive, negative and neutral. Sentiment analysis is an application of Natural Language Processing and text analysis which helps to identify the emotions in a given context. In this work a hybrid approach for Sentiment Analysis is used in which Machine Learning method is used for tagging and Fuzzy Logic is used to find the membership of the review in each sentiment class. Fuzzy logic can be used to handle the vagueness in natural language. Manually tagged data are trained using TnT and Fuzzy rules are incorporated for identifying and classifying the emotions. Certain other rules are also incorporated to handle certain special cases. The fuzzy rules yield output that varies in degree between 0 and 1.

TnT is an efficient statistical Parts-of-speech (POS) Tagger based on Hidden Markov Model. TnT stands for Trigrams‘n’Tags. Viterbi algorithm is used for finding the best tag sequence for a given observation sequence of words. TnT performs well on known word sequences. But, the performance degrades with increase in the number of unknown words. In this paper, we propose a method to overcome this performance degradation using fuzzy rules. Fuzzy rule based model is designed to provide TnT with sufficient information about the tag of unknown words without degrading the performance of TnT. When TnT with fuzzy rule based learning encounters an unknown word, the TnT generates a set of possible tags for the given word, based on the fuzzy rules matched by the word. If the word does not match any fuzzy rule, then the model depends upon the probability distribution of the suffix. This approach guarantees that the performance of TnT will only be improved from its normal performance.

In this paper, we present the details of a phonotactic language identification (LID) system developed for five Indian languages, English (Indian), Hindi, Malayalam, Tamil and Kan-nada. Since there are no publicly available speech databases for English, Malayalam and Kannada, we developed the database for each of the target languages by downloading the audio files from YouTube videos and removing the non-speech signals manually. The system was tested using a test data set consisting of 40 utterances with duration of 30, 10, and 3 sees, in each of 5 target languages. The performance evaluation was done separately accordingly to the NIST benchmarking sessions, for 30s, 10s and 3s segments separately. For the baseline system, we got an overall EER of 10.41 %, 19.56 % and 31.45 % for 30, 10, and 3 sees segments when tested with a 3-gram language model. The use of 4-gram language model has helped enhance the performance of the LID system to 9.81 %, 19.38 % and 32.77% respectively for 30,10 and 3 sees test segments. Further, by using the n-gram smoothing, we were able to improve the EER of the LID system, 9.02 %, 18.70 % and 29.24 % for 3-gram language models and 8.88 %, 16.46 % and 32.03 % for 4-gram language models, respectively for 30,10, and 3 sec test segments. The study shows that the use of 4-gram language models can help enhance the performance of LID systems for Indian languages.

Speech/non-speech detection (SND) distinguishes between speech and non-speech segments in recorded audio and video documents. SND systems can help reduce the storage space required when only speech segments from the audio documents are required, for example content analysis, spoken language identification, etc. In this work, we experimented with the use of time domain, frequency domain and cepstral domain features for short time frames of 20 ms. size along with their mean and standard deviation for segments of size 200 ms. We then analysed if selecting a subset of the features can help improve the performance of the SND system. Towards this, we experimented with different feature selection algorithms, and observed that correlation based feature selection gave the best results. Further, we experimented with different decision tree classification algorithms, and note that random forest algorithm outperformed other decision tree algorithms. We further improved the SND system performance by smoothing the decisions over 5 segments of 200 ms. each. Our baseline system has 272 features, a classification accuracy of 94.45 % and the final system with 8 features has a classification accuracy of 97.80 %.

This paper studies the contribution of different phones in speech data towards improving the performance of text/language independent speaker recognition systems. This work is motivated by the fact that the removal of silence segments from the speech data improves the system performance significantly as it does not contain any speaker-specific information. It is also clear from the literature that not all the phones in the speech data contains equal amount of speaker-specific information in it and the performance of the speaker recognition systems depends on this information. In addition to the silence segments, our work empirically finds 18 other diluent phones that has minimum speaker discrimination capability. We propose to use a preprocessing stage that identifies all non-informative set of phones recursively and removes them along with silence segments. Results show that using phones removed preprocessed data in state-of-the-art i-vector system outperforms the baseline i-vector system. We report absolute improvements of 1%, 1%, 2%, 2% and 1% in EER for test set collected through channels of Digital Voice Recorder, Headset, Mobile Phone 1, Mobile Phone 2 and Tablet PC respectively on IITG-MV database.

Malayalam, a classical language in India, is spoken by over 40 million people. This paper proposes an effective information retrieval system for Malayalam, which retrieve Malayalam documents relevant to the user's information need. The proposed system improves effectiveness by considering synonyms and negations of the terms specified in the query. Though the evaluation of the system is performed over a small corpus, the results are promising. The proposed system is thus relevant for various natural language processing tasks in an highly agglutinative language like Malayalam.

Stemming is the process of removing the affixes from inflections and to return the root form. Malayalam is highly agglutinative in nature and hundreds of inflections are possible for each word. An effective stemmer in Malayalam is not yet implemented. This paper presents a lightweight stemmer for Malayalam, which conflates terms by suffix removal. The proposed stemmer is both computationally inexpensive and domain independent and will serve as a vital part in many areas of Malayalam Language Computing.

In natural language processing applications, string matching is the main time-consuming operation due to the large size of lexicon. Data dependence is minimal in string matching operations, and hence it is ideal for parallelization. A dedicated hardware for string matching that uses memory interleaving and parallel processing techniques can relieve the host CPU from this burden, thereby making the system suitable for real-time applications. This paper reports the FPGA design of such a system with m parallel matching units. The time complexity of the proposed algorithm is O (log2 n), where n is the total number of lexical entries. This has been achieved by a proper selection of the value of m. A special memory organization technique, which reduces the storage space by nearly 70%, has been adopted for storing lexical entries. The techniques used for matching and storage of lexical entries make the system language independent

Present day information retrieval systems largely ignore the issues of lexical and compositional semantics, and rely mainly on some statistical measures for choosing or evolving an indexing scheme. This has been the reason for the decreasing precision in their responses, given an exponentially increasing number of Web pages. The work reported in this paper addresses this issue from a linguistic point of view. We show that the detection of domain-specific phrases can capture the task-specific semantics of documents. We introduce the notion of n*-gram formalism to characterize the domain-specific phrases and their variants, taking a few sample domains. A method to construct a phrase grammar from a small set of documents is proposed. A method of conceptual indexing based on the phrase grammar has also been proposed. In order to demonstrate the effectiveness of the proposed method, we have designed a versatile system that can perform concept-based retrieval, in addition to several document-processing tasks, such as text classification, extraction-based summarization, context tracking, and semantic tagging. Collectively, the system can address the semantic content of documents. Considering the fact that an average user prefers highly relevant results in the top-ranked subset to an exhaustively retrieved set, it is shown that the proposed system performs better in that it retrieves documents that are more conceptually relevant than those retrieved by Google, and at 95% confidence level. Copyright ? 2005 Taylor & Francis Inc.

In Natural Language Processing applications, string matching is the main time-consuming operation. A dedicated co-processor for string matching that uses memory interleaving and parallel processing techniques can relieve the host CPU from this burden. This paper reports the FPGA design of such a system with m parallel matching units. It has been shown to improve the performance by a factor of nearly m, without increasing the chip area by more than 45% The time complexity of the proposed algorithm is O(log/sub 2/ n), where n is the number of lexical entries. The memory used by the lexicon has been efficiently organized and the space saving achieved is about 67%.

A story may be analyzed to identify the main characters and to extract the relationship between them. Relation extraction problems are generally solved either through supervised or unsupervised learning algorithms. In the former, there should be a text corpus for which the entities and their relation types are already known. Such algorithms typically learn to classify new entity pairs into any of the relation types it has already seen, based on some recurring patterns. On the other hand, the unsupervised learning approach is used when there is no such marked up corpus. Such algorithms typically identify patterns relevant to the relation extraction task, occurring within the corpus and then use these patterns to group entities such that the entities within a group share similar relationships. The proposed method is a hybrid approach which combines the features of unsupervised and supervised learning methods. It also uses some rules to extract relationships. The method identifies the main characters and collects the sentences related to them. Then these sentences are analyzed and classified to extract relationships. The main applications are story summarization and analysis of the major characters in stories.