Computer Sc and Engg

We describe a datapath-scalable, minimalist cryptographic processor, called PAX, for mobile environments where the communication with the outside world is done on wireless connections. PAX is designed to fully utilize the high data rates of the newest and developing wireless technologies. Today, these rates exceed 2 Mbps for cellular/PCS connections, and 50 Mbps for WLAN connections. Future rates are expected to be about 20 Mbps and 100 Mbps for cellular/PCS and WLAN respectively.

In our proposed scheme, the data owner outsources huge volume of data to a cloud storage provider and the end users request data to the data owner. The data owner encrypts the data before sending it to the cloud service provider and does over-encryption proposed by Vimercati et al. [4] to the outsourced encrypted data before sending it to the users. We incorporate an existing Elliptic Curve Cryptography (ECC) based key management scheme in user hierarchy proposed by Nikooghadam et al. [11] in our scheme which classifies users in security classes and efficiently helps to derive the secret keys of the lower order security classes. The cloud storage component of our scheme will not have to perform any extra work except storing data and this reduces the cost of the data owner as per pay-per-use pricing policy of the cloud service provider. Our scheme incurs low overhead for key generation as well as for its storage and the end users can use wireless mobile devices. The scheme is useful in owner-write-users-read applications and it is secured from the adversaries.

Language usage over computer mediated discourses, like chats, emails and SMS texts, significantly differs from the standard form of the language. An urge towards shorter message length facilitating faster typing and the need for semantic clarity, shape the structure of this non-standard form known as the texting language. In this work we formally investigate the nature and type of compressions used in SMS texts, and based on the findings develop a word level model for the texting language. For every word in the standard language, we construct a Hidden Markov Model that succinctly represent all possible variations of that word in the texting language along with their associated observation probabilities. The structure of the HMM is novel and arrived at through linguistic analysis of the SMS data. The model parameters have been estimated from a word-aligned SMS and standard English parallel corpus, through machine learning techniques. Preliminary evaluation shows that the word-model can be used for decoding texting language words to their standard counterparts with more than 80% accuracy.

This paper describes our work on Bengali Part of Speech (POS) tagging using a corpus-based approach. There are several approaches for part of speech tagging. This paper deals with a model that uses a combination of supervised and unsupervised learning using a Hidden Markov Model (HMM). We make use of small tagged corpus and a large untagged corpus. We also make use of Morphological Analyzer. Bengali is a highly ambiguous and relatively free word order language. We have obtained an overall accuracy of 95%.

In this paper we discuss a technique for web site personalization. Connectivity analysis has been shown to be useful in identifying high quality web pages within a topic or domain specific graph of hyper linked documents. We have implemented a system that creates a view of a subset of a web site most relevant to a given user. This sort of personalization is useful for filtering a useful subset of a site so that the user gets a low volume of quality information. The essence of our approach is to augment a previous connectivity analysis with content analysis. We present an agent which assists the user when he browses and distills a personalized sub graph of the website based on his user profile.

Schwa deletion is an important issue in grapheme-to-phoneme conversion for Indo-Aryan languages (IAL). In this paper, we describe a syllable minimization based algorithm for dealing with this that outperforms the existing methods in terms of efficiency and accuracy. The algorithm is motivated by the fact that deletion of schwa is a diachronic and sociolinguistic phenomenon that facilitates faster communication through syllable economy. The contribution of the paper is not just a better algorithm for schwa deletion; rather we describe here a constrained optimization based framework that can partly model the evolution of languages, and hence, can be used for solving many problems in computational linguistics that call for diachronic explanations.

We investigate the role of learning in search-based systems for solving optimization problems. Many AI problem solving systems solve problems repeatedly from the same domain. If the problems come from the same distribution in the learning phase and the problem solving phase, the problem solver can acquire information while solving problems, which can be used to solve subsequent problems faster. We use a learning model, where the values of a set of features can be used to induce a clustering of the problem state space. The feasible set of h* values corresponding to each cluster is called h*set. If we relax the optimality guarantee, and tolerate a risk factor, the distribution of h*set can be used to expedite search and produce results within a given risk of suboptimality. The offline learning method consists of solving a batch of problems by using A* to learn the distribution of the h*set in the learning phase. This distribution can be used to solve the rest of the problems effectively. We show how the knowledge acquisition phase can be integrated with the problem solving phase. We present a continuous on-line learning scheme that uses an "anytime" algorithm to learn continuously while solving problems. The system starts with initial assumed distributions of h*set which are used to solve the initial problems. The results are used to update the distributions continuously and with time the distributions converge.

We describe our effort in developing a Named Entity Recognition (NER) system for Hindi using Maximum Entropy (Max-Ent) approach. We developed a NER annotated corpora for the purpose. We have tried to identify the most relevant features for Hindi NER task to enable us to develop an efficient NER from the limited corpora developed. Apart from the orthographic and collocation features, we have experimented on the efficiency of using gazetteer lists as features. We also worked on semi-automatic induction of context patterns and experimented with using these as features of the MaxEnt method. We have evaluated the performance of the system against a blind test set having 4 classes -Person, Organization, Location and Date. Our system achieved a f-value of 81.52%.

Named entity recognition is an extremely important and fundamental task of biomedical text mining. Biomedical named entities include mentions of proteins, genes, DNA, RNA, etc which often have complex structures, but it is challenging to identify and classify such entities. Machine learning methods like CRF, MEMM and SVM have been widely used for learning to recognize such entities from an annotated corpus. The identification of appropriate feature templates and the selection of the important feature values play a very important role in the success of these methods. In this paper, we provide a study on word clustering and selection based feature reduction approaches for named entity recognition using a maximum entropy classifier. The identification and selection of features are largely done automatically without using domain knowledge. The performance of the system is found to be superior to existing systems which do not use domain knowledge.

Annotating learning material with metadata allows easy reusability by different learning/tutoring systems. Several metadata standards have been developed to represent learning objects and courses. A learning system needs to use pedagogic attributes including document type, topic, coverage of concepts, and for each concept the significance and the role. Moreover, in order to have a flexible and reusable repository of e-learning materials, it is necessary that the annotation of the documents with such metadata be done in an automatic fashion as far as possible. This paper describes the attributes that represent some important pedagogic characteristics of learning materials. To reduce the overhead of manual annotation we have explored the feasibility of automatic annotation of learning materials with metadata. This facilitates the creation of an elearning open repository for storing these annotated learning materials, which can be used by learning systems. The automatic annotation is based on a domain knowledge base and a number of algorithms like standard classification algorithms, parsing and analysis of documents have been used for this purpose. The results show a fair degree of accuracy, which may be improved in future using more sophisticated algorithms.

The population with severe speech and motor impairments (SSMI) depend solely on the Augmentative and Alternative Communication Techniques (AAC) for their education and communication needs. Unfortunately, the AAC tools are expensive with respect to the Indian people and the language being English is not usable by the majority of the population. Also, the AAC technique being essentially icon-driven, the icons need to socioculturally relevant to the Indian population. In this paper, we present an overview of one education tool and a communication tool that caters to the multilingual need. We also present an access switch mechanism and a corresponding GUI tuned to the access mechanism, for replacing the mouse, which requires finer motor control -an ability absent in most of the people with SSMI. The proposed systems are indigenous and affordable as well.

This paper describes our work on building Part-of-Speech (POS) tagger for Bengali. We have use Hidden Markov Model (HMM) and Maximum Entropy (ME) based stochastic taggers. Bengali is a morphologically rich language and our taggers make use of morphological and contextual information of the words.

In this paper we describe a hybrid system that applies Maximum Entropy model (Max-Ent), language specific rules and gazetteers to the task of Named Entity Recognition (NER) in Indian languages designed for the IJCNLP NERSSEAL shared task. Starting with Named Entity (NE) annotated corpora and a set of features we first build a baseline NER system. Then some language specific rules are added to the system to recognize some specific NE classes. Also we have added some gazetteers and context patterns to the system to increase the performance. As identification of rules and context patterns requires language knowledge, we were able to prepare rules and identify context patterns for Hindi and Bengali only. For the other languages the system uses the MaxEnt model only. After preparing the one-level NER system, we have applied a set of rules to identify the nested entities. The system is able to recognize 12 classes of NEs with 65.13% f-value in Hindi, 65.96% f-value in Bengali and 44.65%, 18.74%, and 35.47% f-value in Oriya, Telugu and Urdu respectively.

In this paper, we propose a novel kernel function for support vector machines (SVM) that can be used for sequential labeling tasks like named entity recognition (NER). Machine learning methods like support vector machines, maximum entropy, hidden Markov model and conditional random fields are the most widely used methods for implementing NER systems. The features used in machine learning algorithms for NER are mostly string based features. The proposed kernel is based on calculating a novel distance function between the string based features. In tasks like NER, the similarity between the contexts as well as the semantic similarity between the words play an important role. The goal is to capture the context and semantic information in NER like tasks. The proposed distance function makes use of certain statistics primarily derived from the training data and hierarchical clustering information. The kernel function is applied to the Hindi and biomedical NER tasks and the results are quite promising.

Morphological synthesis is an essential part of any natural language generation system. bengali is a highly inflectional language with more than 160 different inflected forms for verbs and 36 different forms for nouns, and 24 different forms for pronouns. Moreover, the choice and ordering of suffixes for nouns and structural changes in verb morphology are dependent on the type (animate/inanimate etc) and syllable structure of the root respectively. We describe here a rulebased approach for Bengali morphological synthesis. For nouns and pronouns, the synthesis engine selects the appropriate suffixes, determines the ordering among them, and concatenates them according to the rules. The verb roots have been classified under 19 categories based on the syllable structure. An FSM is used to recognize the category of the root. Thereafter, a simple algorithm generates the inflected form of the verb based on suffix table and rule tables for the classes.

This paper presents a cross-language retrieval system for the retrieval of English documents in response to queries in Bengali and Hindi, as part of our participation in CLEF 1 2007 Ad-hoc bilingual track. We followed the dictionary-based Machine Translation approach to generate the equivalent English query out of Indian language topics. Our main challenge was to work with a limited coverage dictionary (of coverage ∼ 20%) that was available for Hindi-English, and virtually non-existent dictionary for Bengali-English. So we depended mostly on a phonetic transliteration system to overcome this. The CLEF results point to the need for a rich bilingual lexicon, a translation disambiguator, Named Entity Recognizer and a better transliterator for CLIR involving Indian languages. The best MAP values for Bengali and Hindi CLIR for our experiment were 7.26% and 4.77%, which are 20% and 13% of our best monolingual retrieval, respectively.

This report describes our work on Bengali Part-of-speech tagging (POS) for the NLPAI Machine Learning contest 2006. We use a Hidden Markov Model (HMM) based stochastic tagger. The tagger makes use of morphological and contextual information of words. Since only a small labeled training set is provided (41,000 words), a HMM based approach does not yield very good results. In this work, we have used a morphological analyzer to improve the performance of the tagger. Further, we have made use of semi-supervised learning by augmenting the small labeled training set provided with a larger unlabeled training set (100,000 words). The tagger has an accuracy of about 89% on the test data provided.

We report results of stylistic differences in blogging for gender and age group variation. The results are based on two mutually independent features. The first feature is the use of slang words which is a new concept proposed by us for Stylistic study of bloggers. For the second feature, we have analyzed the variation in average length of sentences across various age groups and gender. These features are augmented with previous study results reported in literature for stylistic analysis. The combined feature list enhances the accuracy by a remarkable extent in predicting age and gender. These machine learning experiments were done on two separate demographically tagged blog corpus. Gender determination is more accurate than age group detection over the data spread across all ages but the accuracy of age prediction increases if we sample data with remarkable age difference.