Producing high-dimensional semantic spaces from lexical co-occurrence

In the light of Quantum Physics, probabilistic technique is said to support a backbone in the domain of Quantum Information Retrieval (QIR). This well known domain is based on the Quantum Probability Theory, which is used to examine the probability of relevance of a document given a user's query. The Quantum Probability Ranking Principle (QPRP), captures dependencies between documents in absolute way via “quantum interference” by considering a single vector i.e., one-dimension subspace. The HAL semantic space can be used to capture meaning of the document of information needs by representing the document into a high dimensional semantic space. This paper presents an investigation of the Quantum Probability Ranking Principle with the concept of HAL semantic space such that which can act as a bridge the gap between Quantum Probability Theory and HAL semantic space, that can results to agreeable performance in an ad-hoc retrieval task of document ranking.

Predicting Noun and Verb Latencies: Influential Variables and Task Effects Natalie Kacinik (kacinn01@student.ucr.edu) Christine Chiarello (christine.chiarello@ucr.edu) University of California, Riverside Life Sciences Psychology Building Riverside, CA 92521-0426, U.S.A Abstract Natural language comprehension involves processing a multitude of words that vary along many dimensions, some of which may reflect statistical regularities in language. These variables may differ in their relative importance across various types of words and tasks. This study used a multiple regression approach to investigate potentially important predictors of noun-verb processing across naming, grammatical categorization, and sentence completion tasks. Although there were some indications of different predictors for nouns vs verb processing, the strongest predictors of response latencies were primarily determined by the types of processing most important for a given task. One variable of particular interest was the newly created Noun-Verb Distributional Difference (NVDD) metric developed by Chiarello et al. (1999). NVDD values reflect statistical regularities in language regarding the typicality of the contexts in which nouns and verbs tend to occur. The results suggest that although noun-verb typicality, as assessed via the NVDD, is a valid measure of regularities in noun-verb contexts within a linguistic corpus, individuals may not be very sensitive to this dimension in standard psycholinguistic processing tasks. Introduction Single word recognition is a central component of language processing. The typical approach has been to use a naming or lexical decision (LD) task and a factorial design to investigate the processing effect of one or more variables such as familiarity or imageability while holding other potentially confounding variables constant. In addition, most single word recognition research has tended to use words of different parts of speech without considering grammatical class (e.g., nouns vs. verbs), or has focused on concrete, imageable nouns. Natural language comprehension, however, involves processing a multitude of words varying along many dimensions. These dimensions may be relatively more or less important for various word types, and their relative importance is likely to vary across different forms of language processing (e.g., word pronunciation vs grammatical identification vs sentence integration). With a few exceptions (e.g., Balota & Chumbley, 1984; Balota, Cortese, & Pilotti, 1999), there have been few attempts to investigate the relative importance of various orthographic and semantic dimensions for responding to words across tasks using multiple regression procedures. This approach provides the opportunity to study many variables simultaneously, to determine which lexical dimensions account for the greatest amount of variance in reaction time (RT) and accuracy for a particular task, and to assess whether the variance accounted for is unique, or is shared by other variables. Such a regression approach was used in the present study to investigate the relative importance of different lexical dimensions across three language tasks. To our knowledge no prior regression study has examined whether various predictor variables are equally applicable to words of different grammatical class. This is an important issue because neuropsychological research has shown that nouns and verbs appear to be processed differently in the brain (e.g., Daniele et al., 1994; Koenig & Lehmann, 1996; Sereno, 1999). It is unclear whether these differences are due to neurally separate noun and verb processing systems, or whether these processing differences are mainly due to different semantic dimensions that covary with word class. Investigating several potentially relevant dimensions using a regression approach may be informative regarding these processing differences between nouns and verbs. One possible reason word recognition research has generally been limited to concrete, imageable nouns, is the lack of word norming corpora available for other word types. A recent study by Chiarello, Shears, and Lund (1999), however, provides imageability ratings, frequency values from the Usenet text corpus of the Lund and Burgess (1996) Hyperspace Analog to Language (HAL) model, and a new measure of noun-verb distributional typicality (the Noun- Verb Distributional Difference, NVDD, metric), for a set of 1197 words: 555 “pure” nouns, 427 “pure” verbs, and 215 words “balanced” for noun-verb usage, as classified by the Francis and Kucera (FK, 1982) norms. Noun-Verb Distributional Typicality The new measure of noun-verb usage developed by Chiarello et al. (1999) uses context vectors from the Lund and Burgess (1996) HAL model, where words occurring in similar phrasal and sentential contexts are nearby in high dimensional context space. Context distances were computed between each word and each of the 555 “pure” nouns (according to Francis & Kucera, 1982) and averaged to get a mean noun context distance score. Mean verb distance scores were similarly obtained by computing and

High-dimensional semantic space accounts of priming

The question papers of many academic institutions contain a significant part of descriptive type questions. Generally, these question papers follow a fixed pattern. Normally, the descriptive type answers are evaluated manually with the possibilities of deviation in the evaluation of the same answer-script by different examiners with almost the same amount of experience and academic qualifications. The present work focuses on the online evaluation of descriptive type answers which will eliminate the discrepancy in manual evaluation. Evaluation of students' answers in online can be performed using Hyperspace Analog to Language (HAL) procedure and Self-Organizing Map (SOM) method. The answer is given as input to HAL. HAL represents a procedure that processes a corpus of text and produce numeric vectors containing information about its meanings. The clustering method of Kohonen Self-Organizing Maps is applied to the vector. SOM is neural based technique. It takes the vectors as inputs and forms a document map in such a way that nearby neurons contain a similar document. The proposed method will be tested with short answers written by learners of our department. This method has a number of benefits like increased reliability of results, reduced time and effort, reduced burden on the faculty and efficient use of resources.

Vector space models (VSMs) represent word meanings as points in a high dimensional space. VSMs are typically created using a large text corpora, and so represent word semantics as observed in text. We present a new algorithm (JNNSE) that can incorporate a measure of semantics not previously used to create VSMs: brain activation data recorded while people read words. The resulting model takes advantage of the complementary strengths and weaknesses of corpus and brain activation data to give a more complete representation of semantics. Evaluations show that the model 1) matches a behavioral measure of semantics more closely, 2) can be used to predict corpus data for unseen words and 3) has predictive power that generalizes across brain imaging technologies and across subjects. We believe that the model is thus a more faithful representation of mental vocabularies.

Buntonia is a common genus in the Maastrichtian to Late Eocene basins of Africa. Analysis of similarity (ANOSIM), hierarchical clustering (HC), and non-metric multidimensional scaling (nMDS) were used to analyze the geographic pattern of Buntonia species based on the presence/absence matrix of 42 species. The Jaccard-based HC shows that the African countries can be subdivided into three main bioregions: North Africa (Algeria, Libya, Tunisia, Morocco, and Sudan), West Africa (Mali, Togo, Benin, Cote d’Ivoire, Senegal, Ghana, Gambia, Nigeria, Cameron, and Mauritania), and South Africa (Namibia and South Africa). ANOSIM indicated a non-significant difference between North and West African countries and between West and Southern Africa. However, a significant difference was found between North and Southern Africa (sequential Bonferroni corrected p < 0.01). These patterns are well-characterized also by both of HC and nMDS. Moreover, analyses of the stratigraphic ranges of the Buntonia species revealed a distinct range shift within the regions of the Guinean Gulf (Togo, Benin, and Cote d’Ivoire) through Senegal towards North Africa. The latter indicates two possible migration routes from West Africa to North Africa through the Trans-Saharan seaway and from the Atlantic to the Tethyan Sea. Migration from region to region played an important role in the distribution of the studied Buntonia species rather than origination of new species and/or extinction of existing species.

We present COSS, an exact method for high-dimensional distance similarity self-joins using the GPU, which finds all points within a search distance e from each point in a dataset. The similarity self-join can take advantage of the massive parallelism afforded by GPUs, as each point can be searched in parallel. Despite high GPU throughput, distance similarity self-joins exhibit irregular memory access patterns which yield branch divergence and other performance limiting factors. Consequently, we propose several GPU optimizations to improve self-join query throughput, including an index designed for GPU architecture. As data dimensionality increases, the search space increases exponentially. Therefore, to find a reasonable number of neighbors for each point in the dataset, e may need to be large. The majority of indexing strategies that are used to prune the ∈-search focus on a spatial partition of data points based on each point's coordinates. As dimensionality increases, this data partitioning and pruning strategy yields exhaustive searches that eventually degrade to a brute force (quadratic) search, which is the well-known curse of dimensionality problem. To enable pruning the search using an indexing scheme in high-dimensional spaces, we depart from previous indexing approaches, and propose an indexing strategy that does not index based on each point's coordinate values. Instead, we index based on the distances to reference points, which are arbitrary points in the coordinate space. We show that our indexing scheme is able to prune the search for nearby points in high-dimensional spaces where other approaches yield high performance degradation. COSS achieves a speedup over CPU and GPU reference implementations up to 17.7X and 11.8X, respectively.

With the rapid development of the Internet and multimedia technologies over the last decade, a huge amount of data has become available, from text corpus, to collections of online images and videos. Cheap storage cost and modern database technologies have made it possible to accumulate large-scale datasets. However, the ever-growing sizes of the datasets make it harder to search useful information from such data. A fundamental computational primitive for dealing with massive multimedia datasets is the similarity search problem. Multimedia similarity search aims to preprocess a database so that given a query object, one can quickly find its similar objects in the database. Searching similar objects from a large dataset in high-dimensional spaces is at the heart of many multimedia applications, such as near-duplicate retrieval, multimedia tagging, recommendation, and so on. Driven by its significance, lots of efforts have been made on this topic. The goal of my research is to design efficient hashing methods for large-scale multimedia search. In this paper, we first present the general framework for multimedia similarity search and discuss the latest improvements and progresses in the field. Then we describe the contributions we have made to effectively and efficiently search similar multimedia objects from large-scale databases. Finally, we discuss the future work and draw a conclusion.

There is a growing interest in the study of the human gut microbiota, as correlationsbetween changes in bacterial profiles and diseases are increasingly discovered. Studies inthis field generally use fecal samples, but it is often easier to obtain colon contentaspirates during colonoscopy. This study used automated ribosomal internal spacer analysis(ARISA) to examine the extent to which the microbiota of colon aspirate samples obtainedafter bowel cleansing can reflect interindividual differences and serve as a proxy forfecal samples. Pre-bowel preparation fecal samples as well as colonoscopy aspirate samplesfrom the cecum and rectum were obtained from 19 subjects. DNA was extracted from allsamples, and comparative analysis was performed, including analysis of similarity (ANOSIM)and nonmetric multidimensional scaling. ANOSIM confirmed that samples from the sameindividual were well separated from samples from different individuals. Significantlylarger differences were found between samples from different individuals than betweensamples of the same individual (R = 0.7605, p < 0.0001). These findings show thatpost-bowel preparation aspirates maintain a strong individual signature. Colonoscopyaspirates can therefore serve as a substitute for fecal samples in studies comparing themicrobiota of different clinical study groups, especially when fecal samples areunavailable.

The paper analyzes the concepts and relations between them in the RuWordNet digital thesaurus of the Russian language. Semantically related words automatically extracted from the corpus of texts are compared with the words of the thesaurus. The characteristic cases of information gap (omission of word meanings and semantic relationships) in the thesaurus are revealed, their causes of a computer and cognitive nature are described. Recommendations on improving the RuWordNet thesaurus are given.

Ecological assessment of highly heterogeneous systems: The importance of taxonomic sufficiency

Statistical language models improve the performance of speech recognition systems by providing estimates of a priori probabilities of word sequences. The commonly used trigram language models obtain the conditional probability estimate of a word given the previous two words, from a large corpus of text. The text corpus is often a collection of several small diverse segments such as newspaper articles, or conversations on different topics. Knowledge of the current topic could be utilized to adapt the general trigram language models to match that topic closely. For example, an interpolation of the general language model with one built on the topic data could be used. The authors first discuss the adaptation of general trigram language models to a known topic using the minimum discrimination information (MDI) method. They then present results on the switchboard corpus which consists of telephone conversations on several topics.

The recent development and refinement of cDNA array and genechip techniques allows for large-scale parallel screening of gene expression and thereby provides a global assessment of molecular events transpiring in cell populations. We hypothesized that such an approach might help illuminate current issues in glioma research. To what degree are glioblastoma multiforme (GBM) cell lines representative of primary GBM tumors? We carried out a gene expression profiling study using cDNA array technology on 10 GBM primary tissues and 3 GBM cell lines. The gene expression levels were quantified, and the within-subject ranks of the gene expression levels were subsequently evaluated by hierarchical clustering analysis, multidimensional scaling analysis, and principal component analysis. Hierarchical cluster analysis shows that the 3 cell lines form one main cluster and the 10 tissue samples form a separate cluster. Multidimensional scaling and principal components analysis provided further graphical demonstration that the cell lines are clearly different from the tissue samples and that the cell lines are more different between themselves than are the tissues.

The health condition of rolling bearings directly impacts the service performance and reliability of the entire rotating system, with degradation indexs serving as critical elements for effective bearing health monitoring. To address the challenges in timely and accurate reflecting the degradation state, this study integrates multi-resolution singular value decomposition (MRSVD), feature weighting and multidimensional scaling (MDS) analysis to construct a comprehensive degradation index for rolling bearings. Time-domain features are initially extracted from bearing vibration signals, followed by correlation analysis using Pearson’s correlation coefficient to select highly correlated features. The MRSVD method is then employed for multi-level decomposition to achieve deep-level feature extraction. The selected features undergo Gini index and Box–Cox sparsity weighting processing to enhance sensitive feature identification capability. Furthermore, MDS is implemented for dimensionality reduction of the weighted features, ultimately enabling the integration of degradation indices into a unified metric, and the comprehensive degradation index is constructed. Experimental results demonstrate that the proposed method achieves superior performance in real-time bearing health monitoring, it significantly improves the accuracy and timeliness of fault detection, thereby providing effective technical support for the maintenance management of rotating machinery.

Objective: Brain-Computer Interfaces (BCI) may help patients with faltering communication abilities due to neurodegenerative diseases produce text or speech by direct neural processing. However, their practical realization has proven difficult due to limitations in speed, accuracy, and generalizability of existing interfaces. The goal of this study is to evaluate the BCI performance of a robust speech decoding system that translates neural signals evoked by speech to a textual output. While previous studies have approached this problem by using neural signals to choose from a limited set of possible words, we employ a more general model that can type any word from a large corpus of English text. Approach: In this study, we create an end-to-end BCI that translates neural signals associated with overt speech into text output. Our decoding system first isolates frequency bands in the input depth-electrode signal encapsulating differential information regarding production of various phonemic classes. These bands form a feature set that then feeds into a Long Short-Term Memory (LSTM) model which discerns at each time point probability distributions across all phonemes uttered by a subject. Finally, a particle filtering algorithm temporally smooths these probabilities by incorporating prior knowledge of the English language to output text corresponding to the decoded word. The generalizability of our decoder is driven by the lack of a vocabulary constraint on this output word. Main result: This method was evaluated using a dataset of 6 neurosurgical patients implanted with intra-cranial depth electrodes to identify seizure foci for potential surgical treatment of epilepsy. We averaged 32% word accuracy and on the phoneme-level obtained 46% precision, 51% recall and 73.32% average phoneme error rate while also achieving significant increases in speed when compared to several other BCI approaches. Significance: Our study employs a more general neural signal-to-text model which could facilitate communication by patients in everyday environments.