Dictionary Model Research Articles

Domain-Specific Aspect-Sentiment Pair Extraction Using Rules and Compound Noun Lexicon for Customer Reviews

Online reviews are an important source of opinion to measure products’ quality. Hence, automated opinion mining is used to extract important features (aspect) and related comments (sentiment). Extraction of correct aspect-sentiment pairs is critical for overall outcome of opinion mining; however, current works still have limitations in terms of identifying special compound noun and parent-child relationship aspects in the extraction process. To address these problems, an aspect-sentiment pair extraction using the rules and compound noun lexicon (ASPERC) model is proposed. The model consists of three main phases, such as compound noun lexicon generation, aspect-sentiment pair rule generation, and aspect-sentiment pair extraction. The combined approach of rules generated from training sentences and domain specific compound noun lexicon enable extraction of more aspects by firstly identifying special compound noun and parent-child aspects, which eventually contribute to more aspect-sentiment pair extraction. The experiment is conducted with the SemEval 2014 dataset to compare proposed and baseline models. Both ASPERC and its variant, ASPER, result higher in recall (28.58% and 22.55% each) compared to baseline and satisfactorily extract more aspect sentiment pairs. Lastly, the reasonable outcome of ASPER indicates applicability of rules to various domains.

Class-specific synthesized dictionary model for Zero-Shot Learning

Zero-Shot Learning (ZSL) aims at recognizing unseen classes that are absent during the training stage. Unlike the existing approaches that learn a visual-semantic embedding model to bridge the low-level visual space and the high-level class prototype space, we propose a novel synthesized approach for addressing ZSL within a dictionary learning framework. Specifically, it learns both a dictionary matrix and a class-specific encoding matrix for each seen class to synthesize pseudo instances for unseen classes with auxiliary of seen class prototypes. This allows us to train the classifiers for the unseen classes with these pseudo instances. In this way, ZSL can be treated as a traditional classification task, which makes it applicable for traditional and generalized ZSL settings simultaneously. Extensive experimental results on four benchmark datasets (AwA, CUB, aPY, and SUN) demonstrate that our method yields competitive performances compared to state-of-the-art methods on both settings.

Large vocabulary recognition for online Turkish handwriting with sublexical units

We present a system for large vocabulary recognition of online Turkish handwriting, using hidden Markov models. While using a traditional approach for the recognizer, we have identified and developed solutions for the main problems specific to Turkish handwriting recognition. % First, since large amounts of Turkish handwriting samples are not available, the system is trained and optimized using the large UNIPEN dataset of English handwriting, before extending it to Turkish using a small Turkish dataset. % The delayed strokes, which pose a significant source of variation in writing order due to the large number of diacritical marks in Turkish, are removed during preprocessing. % Finally, as a solution to the high out-of-vocabulary rates encountered when using a fixed size lexicon in general purpose recognition, a lexicon is constructed from sublexical units (stems and endings) learned from a large Turkish corpus. A statistical bigram language model learned from the same corpus is also applied during the decoding process. The system obtains a 91.7 % word recognition rate when tested on a small Turkish handwritten word dataset using a medium sized (1950 words) lexicon corresponding to the vocabulary of the test set and 63.8 % using a large, general purpose lexicon (130,000 words). However, with the proposed stem+ending lexicon (12,500 words) and bigram language model with lattice expansion, a 67.9 % word recognition accuracy is obtained, surpassing the results obtained with the general purpose lexicon while using a much smaller one.% that still has the same text corpus coverage.

The bilingual mental lexicon beyond Dutch–English written words

The contribution that Ton Dijkstra has made to the field of bilingualism, with his colleagues over the years, is beyond measure. He has advanced our field with the thoughtful and thought-provoking models of the bilingual lexicon he has put forward, and with the vast empirical data he and his colleagues have collected from numerous bilinguals, using a variety of experimental methods. This paper by Dijkstra, Wahl, Buytenhuijs, van Halem, Al-jibouri, de Korte, and Rekké (2018) is no exception. It comprises a thoughtful and detailed description of a new model, Multilink, and provides relevant information regarding the context in which the model was developed, its assumptions, its successes and challenges. The model is the first to consider aspects of both word production and word recognition, as well as of word translation, and to explicitly address how translation equivalents that share both form and meaning – cognates – may be processed. Also addressed are word-frequency concerns, including differences in word frequency that are dependent on language proficiency and exposure. The assumptions that are made within the model are clearly laid out, and less-than-ideal decisions that needed to be made are acknowledged. The authors conclude that a model like Multilink could be the basis for a general model of the mental lexicon. In that, they promote the view taken in Libben, Goral, and Libben (2017) that the bilingual – not the monolingual – lexicon should be considered as the default.

Adaptive Super-Resolution for Remote Sensing Images Based on Sparse Representation With Global Joint Dictionary Model

Sparse representation has been widely used in the field of remote sensing image super-resolution (SR) to restore a high-quality image from a low-resolution (LR) image, e.g., from the blurred and downsampled version of an LR image’s high-resolution (HR) counterpart. It is well known that each image patch can be represented by a linear combination of the atoms of an overcomplete dictionary, and we can obtain an expression of sparse coefficients by $l_{1}$ norm regularization. Owing to the lack of an inner relationship between image patches and an image’s global information, the traditional methods of jointly training two overcomplete dictionaries cannot obtain good SR results. Therefore, we propose an effective approach for remote sensing image SR based on sparse representation. More specifically, a novel global joint dictionary model (GJDM) is used to explore the prior knowledge of images, including local and global characteristics. First, we train two dictionaries for detail image patches and HR patches. Second, in order to enhance the inner relationship between image patches, we introduce a global self-compatibility model for global regularization. Finally, the sparse representation and the local and nonlocal constraints are integrated to improve the performance of the model, and the fast adaptive shrinkage-thresholding algorithm is employed to solve the convex optimization problem in the GJDM. Compared with other methods, the results of the proposed method show good SR performance in preserving details and texture information and significant improvement in a peak signal-to-noise ratio.

The Effect of Word Type on the L1 Support for L2 Vocabulary Learning: The Case of Arab EFL Learners

There is strong agreement in the literature that the first language (L1) supports second language (L2) vocabulary learning particularly at early stages of learning. The present study examines the influence of word type on the use of L1 language to support L2 vocabulary learning. To this end, 130 Arabic-speaking tertiary elementary female learners of English were taught 24 unfamiliar English words in 4 sessions over 4 weeks. The participants were divided into three groups; (1) a group where target words were taught with the use of translation equivalents, (2) a group taught with L2-only-definitions and (3) a control group who received no special treatment. Comparing the participants’ performance in pre-, immediate post- and delayed post-tests showed that the two experimental groups outperformed the control group, reflecting the benefit of the treatment. The results also showed greater vocabulary gains in terms of long-term retrieval for L1 use, but did not show consistent patterns for the special benefit of L1 in terms of specific word types. The results are interpreted in light of the existing literature and relevant bilingual lexicon models. Pedagogical implications and suggested research directions are also proposed.

Droplet Image Super Resolution Based on Sparse Representation and Kernel Regression

Microgravity and containerless conditions, which are produced via electrostatic levitation combined with a drop tube, are important when studying the intrinsic properties of new metastable materials. Generally, temperature and image sensors can be used to measure the changes of sample temperature, morphology and volume. Then, the specific heat, surface tension, viscosity changes and sample density can be obtained. Considering that the falling speed of the material sample droplet is approximately 31.3 m/s when it reaches the bottom of a 50-meter-high drop tube, a high-speed camera with a collection rate of up to 106 frames/s is required to image the falling droplet. However, at the high-speed mode, very few pixels, approximately 48-120, will be obtained in each exposure time, which results in low image quality. Super-resolution image reconstruction is an algorithm that provides finer details than the sampling grid of a given imaging device by increasing the number of pixels per unit area in the image. In this work, we demonstrate the application of single image-resolution reconstruction in the microgravity and electrostatic levitation for the first time. Here, using the image super-resolution method based on sparse representation, a low-resolution droplet image can be reconstructed. Employed Yang’s related dictionary model, high- and low-resolution image patches were combined with dictionary training, and high- and low-resolution-related dictionaries were obtained. The online double-sparse dictionary training algorithm was used in the study of related dictionaries and overcome the shortcomings of the traditional training algorithm with small image patch. During the stage of image reconstruction, the algorithm of kernel regression is added, which effectively overcomes the shortcomings of the Yang image’s edge blurs.

Multiplex model of mental lexicon reveals explosive learning in humans

Word similarities affect language acquisition and use in a multi-relational way barely accounted for in the literature. We propose a multiplex network representation of this mental lexicon of word similarities as a natural framework for investigating large-scale cognitive patterns. Our representation accounts for semantic, taxonomic, and phonological interactions and it identifies a cluster of words which are used with greater frequency, are identified, memorised, and learned more easily, and have more meanings than expected at random. This cluster emerges around age 7 through an explosive transition not reproduced by null models. We relate this explosive emergence to polysemy – redundancy in word meanings. Results indicate that the word cluster acts as a core for the lexicon, increasing both lexical navigability and robustness to linguistic degradation. Our findings provide quantitative confirmation of existing conjectures about core structure in the mental lexicon and the importance of integrating multi-relational word-word interactions in psycholinguistic frameworks.

Corpus Based Twitter Sentiment Analysis

Social media is the main advertising and analysis tool in today’s world. Twitter is a microblogging service usually used as an instant communication platform. It contains a rich amount of data in semi-structured format. The capacity to provide information in real time has stimulated many companies to use this service to understand their consumers. In this work, a method had been adopted to fetch the data in real time environment directly from twitter and sentiment analysis was performed on the streamed data. To provide scalability and reduced cost of analytics, the work was implemented using the Apache Open Source Platform, Hadoop. Further, the sentiment analysis was conducted based on the dictionary model.

The multiplex interplay between phonological and semantic networks impacts word production across different types of aphasia

The multiplex interplay between phonological and semantic networks impacts word production across different types of aphasia

Los controvertidos límites de la diferencia específica en el segundo enunciado lexicográfico

Elaborar una definición lexicográfica no es una tarea fácil, sobre todo porque, al margen de la técnica que se desarrolla para configurar y acotar una realidad concreta, se deben tener en cuenta las expectativas del destinatario del repertorio léxico, aunque sin incurrir en excesos informativos que resulten superfluos o desorienten en la correcta interpretación de lo que se dice. Habitualmente, se considera que no solo basta con elegir el hiperónimo acertado, sino que las diferencias específicas no deben contravenir el modelo del diccionario de lengua, en el que se espera que se aporte a través del segundo enunciado la información justa de las entradas para que queden claramente delimitadas. En todo caso, resulta complejo establecer compartimentos estancos entre lo lingüístico y lo extralingüístico, pues, a veces, puede ser contraproducente eludir la información referente a las cosas, ya que la finalidad de la definición no es otra que la de ayudar al usuario a captar el contenido del vocablo que se parafrasea.

Hybrid CTC/Attention Architecture for End-to-End Speech Recognition

Conventional automatic speech recognition (ASR) based on a hidden Markov model (HMM)/deep neural network (DNN) is a very complicated system consisting of various modules such as acoustic, lexicon, and language models. It also requires linguistic resources, such as a pronunciation dictionary, tokenization, and phonetic context-dependency trees. On the other hand, end-to-end ASR has become a popular alternative to greatly simplify the model-building process of conventional ASR systems by representing complicated modules with a single deep network architecture, and by replacing the use of linguistic resources with a data-driven learning method. There are two major types of end-to-end architectures for ASR; attention-based methods use an attention mechanism to perform alignment between acoustic frames and recognized symbols, and connectionist temporal classification (CTC) uses Markov assumptions to efficiently solve sequential problems by dynamic programming. This paper proposes hybrid CTC/attention end-to-end ASR, which effectively utilizes the advantages of both architectures in training and decoding. During training, we employ the multiobjective learning framework to improve robustness and achieve fast convergence. During decoding, we perform joint decoding by combining both attention-based and CTC scores in a one-pass beam search algorithm to further eliminate irregular alignments. Experiments with English (WSJ and CHiME-4) tasks demonstrate the effectiveness of the proposed multiobjective learning over both the CTC and attention-based encoder–decoder baselines. Moreover, the proposed method is applied to two large-scale ASR benchmarks (spontaneous Japanese and Mandarin Chinese), and exhibits performance that is comparable to conventional DNN/HMM ASR systems based on the advantages of both multiobjective learning and joint decoding without linguistic resources.

Na bezrybiu i rak ryba. Językowo-kulturowy obraz raka w polszczyźnie

<p>Artykuł jest próbą rekonstrukcji językowo-kulturowego obrazu raka (Astacus astacus) w polszczyźnie. Wzorem Słownika stereotypów i symboli ludowych pod red. Jerzego Bartmińskiego wykorzystano informacje etymologiczne, a także dane zawarte w słownikach współczesnego języka polskiego, dawnej polszczyzny i gwarowych (w tym zasób frazematyki), teksty folkloru oraz przekazy etnograficzne. Na podstawie tak dobranego materiału udało się sformułować następującą definicję raka „Rak to czerwone stworzenie z ostrymi, służącymi do obrony szczypcami, a także długimi wąsami, małe i małowartościowe, pokryte pancerzem, nieowłosione i nieopierzone, o wielu odnóżach i oczach zlokalizowanych w tylnej części ciała, żyjące w wodzie (rzekach i stawach), w wykopanych przez siebie norach. Nie wiadomo, gdzie spędza zimę. Szczypie, porusza się wspak, niezgrabnie i powoli, nie wydaje głosu, wgryza się w ciało swojej ofiary, powodując jej śmierć. Najbliższa jest mu ryba. Raki dla mięsa, którego dostarczają w niewielkiej ilości, są poławiane przez ludzi, a następnie pieczone lub gotowane. Uchodzą za wykwintne pożywienie. Części ciała raka są lekarstwem na wiele chorób. Rakowi przypisuje się związki ze sferą podziemną”.</p>

Semantic action recognition by learning a pose lexicon

This paper proposes a semantic representation, pose lexicon, for action recognition. The lexicon is composed of a set of semantic poses, a set of visual poses and a probabilistic mapping between the visual and semantic poses. Specially, an action can be represented by a sequence of semantic poses extracted from an associated textual instruction. Visual frames of the action are considered to be generated from a sequence of hidden visual poses. To learn the lexicon, a visual pose model is learned from training samples by a Gaussian Mixture model to characterize the likelihood of an observed visual frame being generated by a visual pose. A pose lexicon model is also learned by an extended hidden Markov alignment model to encode the probabilistic mapping between hidden visual poses and semantic poses sequences. With the lexicon, action classification is formulated as a problem of finding the maximum posterior probability of a given sequence of visual frames that fits to a given sequence of semantic poses through the most likely visual pose and alignment sequences. The efficacy of the proposed method was evaluated on MSRC-12, WorkoutSU-10, WorkoutUOW-18, Combined-15 and Combined-17 action datasets using cross-subject, cross-dataset and zero-shot protocols.

A Case Study on Development of Fraud Dictionary about Course Reviews for Quality Control of Vocational Training

본 연구는 정부가 지원하는 직업훈련의 품질을 관리하기 위해 훈련생이 훈련프로그램을 이수한 이후 작성한 수강후기에 대해 부정내역이 담긴 수강후기를 자동으로 검출해 주는 부정사전을 개발하는데 목적이 있다. 이를 위해 고용노동부가 지원하는 2015년 훈련프로그램의 수강후기 자료(2015.1～2015.12, 118,879건)를 토대로 부정사전을 개발하고, 가장 최근 1년간의 수강후기 자료(2015.7～2016.6, 88,816건)에 부정사전을 적용하여 부정 수강후기를 추출하였다. 본 연구의 주요 결과는 첫째, 비정형화된 텍스트 데이터인 훈련생의 수강후기에 대해 의미분석과 오피니언 마이닝 기법을 적용하여 6단계의 개발절차에 따라 수강후기 부정사전을 개발하였다. 부정사전은 1,278개의 부정단어로 구성 되어있고, 부정단어를 형태와 의미를 고려하여 68개 소분류, 28개 중분류, 3개 대분류로 구성되었으며, 분류 단위에 따라 의미별 가중치를 부여하는 모델을 선정하였다. 전체 수강후기(118,879건)에 대해 사람이 직접 부정 수강후기를 추출할 때는 0.7%(788건)가 추출되었고(8주 소요), 부정사전을 적용할 때는 0.6%(668건)가 추출되어, 부정사전이 사람이 직접 수행할 때와 유사하게 부정 수강후기를 추출하였다. 둘째, 가장 최근에 입력된 1년간의 수강후기 자료에 부정사전 모델을 적용한 결과, 1차적으로 전체 수강후기의 7.2%(6,413건)를 부정 수강후기로 검출하였고, 이를 의미 분석한 결과 최종적으로 0.6%(499건)가 부정내역이 담긴 수강후기로 추출되었다(1주 소요). 이 연구의 의의는 첫째, 직업훈련의 품질을 관리하는데 있어 훈련프로그램에 참여한 훈련생의 주관적인 의견인 빅데이터로 구성된 방대한 수강후기를 오피니언 마이닝 기법을 적용하여 부정의견을 분류하는 부정사전을 개발하여 행정조치 등으로 연계하였다는 점에 큰 의의가 있다. 둘째, 기존에 사람이 직접 부정 수강후기를 추출할 때는 8주의 시간이 소요되었지만, 부정사전을 적용할 때는 1주 밖에 걸리지 않아, 업무수행의 효율성이 크게 증대되었다. 셋째, 오피니언 마이닝 기법을 적용하기 이전에 의미분석 과정을 거침으로써 부정단어 추출의 오차를 최소화하였다.The purpose of this study is to develop a fraud dictionary about course reviews that automatically detects the lecture statements pointing out fraudful training after the trainees completed the training program to manage the quality of the vocational training supported by the government. In order to do this, the fraud dictionary was developed based on the course reviews data of the training program for 2015(2015.1～2015.12, 118,879 cases), and the fraud dictionary was applied to extract lecture statements pointing out fraudful training among course reviews for the latest one year(2015.7～2016.6, 88,816 cases). The finding of the study were as follows. First, we developed the fraud dictionary about course reviews according to the six-step development procedure by applying the semantic analysis and opinion mining techniques to the informative text data. The fraud dictionary consists of 1,278 fraud words. The fraud word is composed of 68 subdivisions, 28 subdivisions, and 3 subdivisions in consideration of form and meaning. We selected a model that assigns semantic weighting according to classification unit. Among the whole course reviews(118,879 cases), 788 cases(0.7%) were extracted as the lecture statements pointing out fraudful training by the person within eight weeks, and 668 cases(0.6%) were extracted as one by the fraud dictionary. The fraud dictionary was extracted in a similar way to the one performed directly by the person. Second, as a result of applying the fraud dictionary model to the most recent one-year course review data, 6,413 cases(7.2%) of the whole course reviews were detected as lecture statements pointing out fraudful training. And as a result of applying the semantic analysis of them, 499 cases(0.6%) were finally extracted as lecture statements pointing out fraudful training within only one week.

FRIST—flipping and rotation invariant sparsifying transform learning and applications

Features based on sparse representation, especially using the synthesis dictionary model, have been heavily exploited in signal processing and computer vision. However, synthesis dictionary learning typically involves NP-hard sparse coding and expensive learning steps. Recently, sparsifying transform learning received interest for its cheap computation and its optimal updates in the alternating algorithms. In this work, we develop a methodology for learning flipping and rotation invariant sparsifying transforms, dubbed FRIST, to better represent natural images that contain textures with various geometrical directions. The proposed alternating FRIST learning algorithm involves efficient optimal updates. We provide a convergence guarantee, and demonstrate the empirical convergence behavior of the proposed FRIST learning approach. Preliminary experiments show the promising performance of FRIST learning for sparse image representation, segmentation, denoising, robust inpainting, and compressed sensing-based magnetic resonance image reconstruction.

What is in the neighborhood of a tonal syllable? Evidence from auditory lexical decision in Mandarin Chinese

Phonological neighborhood effects have been found in spoken word recognition, word production and phonetic variation (Gahl, Yao, & Johnson, 2012; Luce & Pisoni, 1998; Vitevitch, 2002). Overall, words from dense neighborhoods are harder to recognize but easier to produce. However, most previous studies have focused on English, while evidence suggests that these effects may not generalize cross-linguistically due to language-specific configurations of the lexicon (Michael S Vitevitch & Stamer, 2006, 2009). In the current study, we investigate the effects of phonological neighborhoods in Mandarin Chinese, which has a vastly different lexicon structure from that of English. Results from an auditory lexical decision experiment showed that phonological neighborhood density and neighbor frequency (defined by the one-phoneme/tone difference rule) are predictive of the speed and accuracy of lexical decision. Homophone density also has a facilitative effect on the accuracy of lexical decision. The implications of the current findings are discussed in the framework of the lexicon model proposed by Zhou & Marslen-Wilson (1994, 2009).

Efficient Sum of Outer Products Dictionary Learning (SOUP-DIL) and Its Application to Inverse Problems.

The sparsity of signals in a transform domain or dictionary has been exploited in applications such as compression, denoising and inverse problems. More recently, data-driven adaptation of synthesis dictionaries has shown promise compared to analytical dictionary models. However, dictionary learning problems are typically non-convex and NP-hard, and the usual alternating minimization approaches for these problems are often computationally expensive, with the computations dominated by the NP-hard synthesis sparse coding step. This paper exploits the ideas that drive algorithms such as K-SVD, and investigates in detail efficient methods for aggregate sparsity penalized dictionary learning by first approximating the data with a sum of sparse rank-one matrices (outer products) and then using a block coordinate descent approach to estimate the unknowns. The resulting block coordinate descent algorithms involve efficient closed-form solutions. Furthermore, we consider the problem of dictionary-blind image reconstruction, and propose novel and efficient algorithms for adaptive image reconstruction using block coordinate descent and sum of outer products methodologies. We provide a convergence study of the algorithms for dictionary learning and dictionary-blind image reconstruction. Our numerical experiments show the promising performance and speedups provided by the proposed methods over previous schemes in sparse data representation and compressed sensing-based image reconstruction.

Two-Stream Dictionary Learning Architecture for Action Recognition

In this paper, a novel method based on the two-stream dictionary learning architecture for human action recognition is proposed. The architecture consists of interest patch (IP) detector and descriptor, two-stream dictionary models, and support vector machine (SVM) for classification. The novel IP detector combines a human detector and a contour detector to extract patches of interest on human contours. Then the IP descriptors are calculated in spatial stream and temporal stream separately. In each stream, a dictionary is trained for each action with the IP descriptors as an action model. In this way, measuring the similarity between an action sequence and an action model is transformed to reconstructing the IPs in this sequence with the model and computing the reconstruction error. For each action, an IP distribution histogram is constructed and the histogram is further used to train an SVM classifier in each stream. A score fusion method is applied to fuse the spatial and temporal SVM classification results to make a final decision. The proposed architecture is examined on four public data sets with different background complexities and camera motion conditions: Weizmann data set, KTH data set, Olympic sports data set, and HMDB51 data set. The results are further compared with state-of-the-art approaches in the experiment section to confirm the effectiveness of this architecture.

Word frequency predicts translation asymmetry

Bilingualism studies report asymmetries in word processing across languages. Access to L2 words is slower and sensitive to semantic blocking. These observations inform influential models of bilingual processing, which propose autonomous lexicons with different processing routes. In a series of experiments, we explored an alternative hypothesis that the asymmetries are due to frequency of use. Using a within-language ‘translation’ task, involving high/low frequency (HF/LF) synonyms, we obtained parallel results to bilingual studies. Experiment 1 revealed that HF synonyms were accessed faster than LF ones. Experiment 2 showed that semantic blocking slowed retrieval only of LF synonyms, while form blocking produced powerful interference of both HF and LF words. Experiment 3 examined translation speed and sensitivity to blocking in two groups of Russian-English bilinguals who differed in frequency of use of their languages. Translation asymmetries were modulated by frequency of use. The results support an integrated lexicon model of bilingual processing.