Word Sequences Research Articles

Video as Conditional Graph Hierarchy for Multi-Granular Question Answering

Video question answering requires the models to understand and reason about both the complex video and language data to correctly derive the answers. Existing efforts have been focused on designing sophisticated cross-modal interactions to fuse the information from two modalities, while encoding the video and question holistically as frame and word sequences. Despite their success, these methods are essentially revolving around the sequential nature of video- and question-contents, providing little insight to the problem of question-answering and lacking interpretability as well. In this work, we argue that while video is presented in frame sequence, the visual elements (e.g., objects, actions, activities and events) are not sequential but rather hierarchical in semantic space. To align with the multi-granular essence of linguistic concepts in language queries, we propose to model video as a conditional graph hierarchy which weaves together visual facts of different granularity in a level-wise manner, with the guidance of corresponding textual cues. Despite the simplicity, our extensive experiments demonstrate the superiority of such conditional hierarchical graph architecture, with clear performance improvements over prior methods and also better generalization across different type of questions. Further analyses also demonstrate the model's reliability as it shows meaningful visual-textual evidences for the predicted answers.

Novel multi‐domain attention for abstractive summarisation

AbstractThe existing abstractive text summarisation models only consider the word sequence correlations between the source document and the reference summary, and the summary generated by models lacks the cover of the subject of source document due to models' small perspective. In order to make up these disadvantages, a multi‐domain attention pointer (MDA‐Pointer) abstractive summarisation model is proposed in this work. First, the model uses bidirectional long short‐term memory to encode, respectively, the word and sentence sequence of source document for obtaining the semantic representations at word and sentence level. Furthermore, the multi‐domain attention mechanism between the semantic representations and the summary word is established, and the proposed model can generate summary words under the proposed attention mechanism based on the words and sentences. Then, the words are extracted from the vocabulary or the original word sequences through the pointer network to form the summary, and the coverage mechanism is introduced, respectively, into word and sentence level to reduce the redundancy of summary content. Finally, experiment validation is conducted on CNN/Daily Mail dataset. ROUGE evaluation indexes of the model without and with the coverage mechanism are improved respectively, and the results verify the validation of model proposed by this paper.

Using phrase-frames to trace the language development of L1 Chinese learners of English

A recent trend in corpus-based research emphasizes the importance to examine repeated, discontinuous word sequences in addition to continuous ones. The present study extends this trend into research on the use of phrase-frames (repeated word sequences with variable slots, e.g., it is * to) and their variants (e.g., important, easy) in a large corpus of L1 Chinese learner writing in English across five proficiency levels. Phrase-frames were automatically identified based on established frequency and range thresholds and manually filtered for meaningfulness. The analysis focuses on differences in the variability, predictability, and functions of the 100 most frequent phrase-frames (of spans three and four) across levels. The results revealed that, as learners move to more advanced proficiency levels, they tend to employ phrase-frames that are more variable and less predictable. Results also indicate that the functions expressed by high-frequency phrase-frames vary between proficiency levels. This study contributes to a better understanding of the process of L2 learners’ development of phraseological competence.

Learning cricket strokes from spatial and motion visual word sequences

Learning cricket strokes from spatial and motion visual word sequences

Semantic ghost imaging based on recurrent-neural-network.

Ghost imaging (GI) illuminates an object with a sequence of light patterns and obtains the corresponding total echo intensities with a bucket detector. The correlation between the patterns and the bucket signals results in the image. Due to such a mechanism different from the traditional imaging methods, GI has received extensive attention during the past two decades. However, this mechanism also makes GI suffer from slow imaging speed and poor imaging quality. In previous work, each sample, including an illumination pattern and its detected bucket signal, was treated independently with each other. The correlation is therefore a linear superposition of the sequential data. Inspired by human's speech, where sequential words are linked with each other by a certain semantic logic and an incomplete sentence could still convey a correct meaning, we here propose a different perspective that there is potentially a non-linear connection between the sequential samples in GI. We therefore built a system based on a recurrent neural network (RNN), called GI-RNN, which enables recovering high-quality images at low sampling rates. The test with MNIST's handwriting numbers shows that, under a sampling rate of 1.28%, GI-RNN have a 12.58 dB higher than the traditional basic correlation algorithm and a 6.61 dB higher than compressed sensing algorithm in image quality. After trained with natural images, GI-RNN exhibits a strong generalization ability. Not only does GI-RNN work well with the standard images such as "cameraman", but also it can recover the natural scenes in reality at the 3% sampling rate while the SSIMs are greater than 0.7.

Examining performance indicators for a written expression test based on the curriculum

This study aimed to investigate the relationships between students' word-level reading competency, verbal working memory, and performance on a curriculum-based narrative writing exam (N = 42; 23 males). Composition quality and CMIWS, an abbreviation that stands for correct minus incorrect word sequences, were the outcomes evaluated in this study. CMIWS is an abbreviation for "Creation of suitable text grammar and spelling." This is what the acronym means. The CMIWS scores and the total quality score did not strongly correlate. As a predictor of CMIWS, word reading ability fared significantly better than gender, grade level, the automaticity of handwriting, and working memory. This is because having a strong reading ability lowered the amount of variety in the data. Reading fluency remained a strong predictor of composition quality even when other factors, such as the writer's gender and the degree to which their handwriting was automatic, were considered. An updating task was used to evaluate working memory; however, the results did not indicate that it was a unique or significant predictor of CMIWS or composition quality. In addition, it was found that scores on the CMIWS varied depending on grade (the findings for grade 5 were higher than those for grade 4) and gender. Grade 5 scores were higher than grade 4 (girls had higher scores than males). Even though there is a correlation between automaticity in handwriting and both CMIWS scores and writing quality, automaticity in handwriting alone cannot be used to predict any of these measures. The findings support the hypothesis that the CMIWS indicator is sensitive to variations in the environment in which it is located. When evaluating children's writing abilities and developing instructional strategies for children who struggle with writing, it is essential to consider not only the children's reading fluency but also the amount of automaticity in their handwriting. As highlighted by these outcomes, the importance of taking this step cannot be overstated.

Analysis of sentiments on the onset of Covid-19 using Machine Learning Techniques

The novel coronavirus (Covid-19) pandemic has struck the whole world and is one of the most striking topics on social media platforms. Sentiment outbreak on social media enduring various thoughts, opinions, and emotions about the Covid-19 disease, expressing views they are feeling presently. Analyzing sentiments helps to yield better results. Gathering data from different blogging sites like Facebook, Twitter, Weibo, YouTube, Instagram, etc., and Twitter is the largest repository. Videos, text, and audio were also collected from repositories. Sentiment analysis uses opinion mining to acquire the sentiments of its users and categorizes them accordingly as positive, negative, and neutral. Analytical and machine learning classification is implemented to 3586 tweets collected in different time frames. In this paper, sentiment analysis was performed on tweets accumulated during the Covid-19 pandemic, Coronavirus disease. Tweets are collected from the Twitter database using Hydrator a web-based application. Data-preprocessing removes all the noise, outliers from the raw data. With Natural Language Toolkit (NLTK), text classification for sentiment analysis and calculate the score subjective polarity, counts, and sentiment distribution. N-gram is used in textual mining -and Natural Language Processing for a continuous sequence of words in a text or document applying uni-gram, bi-gram, and tri-gram for statistical computation. Term frequency and Inverse document frequency (TF-IDF) is a feature extraction technique that converts textual data into numeric form. Vectorize data feed to our model to obtain insights from linguistic data. Linear SVC, MultinomialNB, GBM, and Random Forest classifier with Tfidf classification model applied to our proposed model. Linear Support Vector classification performs better than the other two classifiers. Results depict that RF performs better.

DINOS: Data INspired Oligo Synthesis for DNA Data Storage

As interest in DNA-based information storage grows, the costs of synthesis have been identified as a key bottleneck. A potential direction is to tune synthesis for data. Data strands tend to be composed of a small set of recurring code word sequences, and they contain longer sequences of repeated data. To exploit these properties, we propose a new framework called DINOS. DINOS consists of three key parts: (i) The first is a hierarchical strand assembly algorithm, inspired by gene assembly techniques that can assemble arbitrary data strands from a small set of primitive blocks. (ii) The assembly algorithm relies on our novel formulation for how to construct primitive blocks, spanning a variety of useful configurations from a set of code words and overhangs. Each primitive block is a code word flanked by a pair of overhangs that are created by a cyclic pairing process that keeps the number of primitive blocks small. Using these primitive blocks, any data strand of arbitrary length can be assembled, theoretically. We show a minimal system for a binary code with as few as six primitive blocks, and we generalize our processes to support an arbitrary set of overhangs and code words. (iii) We exploit our hierarchical assembly approach to identify redundant sequences and coalesce the reactions that create them to make assembly more efficient. We evaluate DINOS and describe its key characteristics. For example, the number of reactions needed to make a strand can be reduced by increasing the number of overhangs or the number of code words, but increasing the number of overhangs offers a small advantage over increasing code words while requiring substantially fewer primitive blocks. However, density is improved more by increasing the number of code words. We also find that a simple redundancy coalescing technique is able to reduce reactions by 90.6% and 41.2% on average for decompressed and compressed data, respectively, even when the smallest data fragments being assembled are 16 bits. With a simple padding heuristic that finds even more redundancy, we can further decrease reactions for the same operating point up to 91.1% and 59% for decompressed and compressed data, respectively, on average. Our approach offers greater density by up to 80% over a prior general purpose gene assembly technique. Finally, in an analysis of synthesis costs in which we make 1 GB volume using de novo synthesis versus making only the primitive blocks with de novo synthesis and otherwise assembling using DINOS, we estimate DINOS as 10 5 × cheaper than de novo synthesis.

A Comparative Study of Text Genres in English-Chinese Translation Effects Based on Deep Learning LSTM.

In recent years, neural network-based English-Chinese translation models have gradually supplanted traditional translation methods. The neural translation model primarily models the entire translation process using the “encoder-attention-decoder” structure. Simultaneously, grammar knowledge is essential for translation, as it aids in the grammatical representation of word sequences and reduces grammatical errors. The focus of this article is on two major studies on attention mechanisms and grammatical knowledge, which will be used to carry out the following two studies. Firstly, in view of the existing neural network structure to build translation model caused by long distance dependent on long-distance information lost in the delivery, leading to problems in terms of the translation effect which is not ideal, put forward a kind of embedded attention long short-term memory (LSTM) network translation model. Secondly, in view of the lack of grammatical prior knowledge in translation models, a method is proposed to integrate grammatical information into translation models as prior knowledge. Finally, the proposed model is simulated on the IWSLT2019 dataset. The results show that the proposed model has a better representation of source language context information than the existing translation model based on the standard LSTM model.

T-BERTSum: Topic-Aware Text Summarization Based on BERT

In the era of social networks, the rapid growth of data mining in information retrieval and natural language processing makes automatic text summarization necessary. Currently, pretrained word embedding and sequence to sequence models can be effectively adapted in social network summarization to extract significant information with strong encoding capability. However, how to tackle the long text dependence and utilize the latent topic mapping has become an increasingly crucial challenge for these models. In this article, we propose a topic-aware extractive and abstractive summarization model named T-BERTSum, based on Bidirectional Encoder Representations from Transformers (BERTs). This is an improvement over previous models, in which the proposed approach can simultaneously infer topics and generate summarization from social texts. First, the encoded latent topic representation, through the neural topic model (NTM), is matched with the embedded representation of BERT, to guide the generation with the topic. Second, the long-term dependencies are learned through the transformer network to jointly explore topic inference and text summarization in an end-to-end manner. Third, the long short-term memory (LSTM) network layers are stacked on the extractive model to capture sequence timing information, and the effective information is further filtered on the abstractive model through a gated network. In addition, a two-stage extractive–abstractive model is constructed to share the information. Compared with the previous work, the proposed model T-BERTSum focuses on pretrained external knowledge and topic mining to capture more accurate contextual representations. Experimental results on the CNN/Daily mail and XSum datasets demonstrate that our proposed model achieves new state-of-the-art results while generating consistent topics compared with the most advanced method.

STYLISTIC DEVICES IN POLITICAL DISCOURSE

The article considers stylistic means in political discourse on the examples of modern and past Ukrainian and foreign politicians’ speeches. The topicality of the research is defined by social significance of political discourse in the life of the community. The term “political discourse” is determined as coherent oral or written text, expressed through both verbal and nonverbal means, which directly depends on the situation of political communication in combination with pragmatic, socio-cultural, psychological and other factors. A “figure of speech” (also called “stylistic device” or “rhetorical device”) is commented on as a modification of the usual or expected sequence of words, grammatical structure of the text, the use of any of a variety of techniques to produce an auxiliary meaning, idea, or feeling. The definitions and examples of “metaphors” (especially those related to war, sports, family, and nature), “personification”, “repetition”, “simile”, “allusion”, “synechdoche”, “hyperbole”, “emotive language”, “balance”, “inverted phrases”, “rhetorical figures” (rhetorical appeals, questions, assumptions, exclamations, statements), “gradation”, “ invective” and “poetic devices” (irony as one of the ways of comic perception of reality, which contains a hidden mockery of facts or people; satire as sharp, sensitive ridicule of vices, errors, negative phenomena of reality, a sharp, scathing mockery; sarcasm as malicious, scathing mockery, caustic irony; humour as benevolent laughter, aimed at exposing certain defects of human character or inconsistencies in people’s lives, in their behaviour; word game as a means of artistic expression that functions in language as a joke, created on the basis of polysemy, homonymy and similar sounding words) are given in the context of political discourse. “Sustained (prolonged) metaphor” is enlarged on with relation to the problem researched. Factors influencing the choice of political discourse strategies and tactics are commented on, namely – a personality of a politician and their function in the political system and the addressee of communication.

The Use of N-Gram Language Model in Predicting Nepali Words

This paper aims to study the problems of automated generation and understanding of natural human languages. The word prediction and word completion from a tab-complete in typing is particularly useful to minimized keystrokes for the users with specific necessaries, and to reduce mistakes, and typographic errors. The word prediction techniques are well-established methods that are frequently used as communication aids for people with disabilities to accelerate the writing, to reduce the effort needed to type and to suggest the correct words. It is something that is skillful at doing prediction according to the previous context. Projection can either be established on word figures or verbal rules. The N-gram model is about predicting nth word from N-1 words. It assigns the probabilities to sentences and sequences of words of all possible combination of n words. To meet the objective, this research uses statistics amount of Nepali language of diverse word kinds to expect right word with as much precision as possible. Under the statistical method, this research will deal with the N-gram method to predict the next word for the Nepali language using Viterbi as decoding algorithm.

A deep learning framework for enhancer prediction using word embedding and sequence generation

Enhancers are non-coding DAN fragments that play key roles in gene regulations and can promote the transcription of structural genes, thereby affecting the expression of structural protein catalytic enzymes and regulatory proteins. Accurate identification of enhancers helps to understand the transcription of structural genes and the development of human tumorigenesis, diagnosis and treatment. The enhancer sequences have high position variations and dispersions, and the identification of enhancers is more challenging than other genetic factors. Based on word embedding and sequence generative adversarial networks, a deep learning framework for enhancer identification is proposed. Firstly, considering the small number of sequences in the benchmark dataset, RankGAN is used to amplify the dataset size while maintaining the data characteristics. Then, in view of the similarity between DNA sequence and natural language, DNA sequence is regarded as a sentence composed of multiple “words”, and the word embedding technology FastText is applied to transform it into a numerical matrix. To extract the dependencies and highly abstract features of nucleotides in DNA sequences, a Long Short-Term Memory Convolutional Neural network (LSTM-CNN) is constructed to perform the identification task. On the independent test set, the accuracy and Matthew's correlation coefficient (MCC) for enhancer prediction are 0.7525 and 0.5051, respectively. For the enhancer type prediction, the accuracy and MCC of this method are 0.6972 and 0.3954, respectively. Compared with existing methods, this method achieves more satisfactory results for the prediction of enhancers and their types. This study will further enrich the application of natural language processing in bioinformatics.

Pay attention to what you read: Non-recurrent handwritten text-Line recognition

The advent of recurrent neural networks for handwriting recognition marked an important milestone reaching impressive recognition accuracies despite the great variability that we observe across different writing styles. Sequential architectures are a perfect fit to model text lines, not only because of the inherent temporal aspect of text, but also to learn probability distributions over sequences of characters and words. However, using such recurrent paradigms comes at a cost at training stage, since their sequential pipelines prevent parallelization. In this work, we introduce a novel method that bypasses any recurrence during the training process with the use of transformer models. By using multi-head self-attention layers both at the visual and textual stages, we are able to tackle character recognition as well as to learn language-related dependencies of the character sequences to be decoded. Our model is unconstrained to any predefined vocabulary, being able to recognize out-of-vocabulary words, i.e. words that do not appear in the training vocabulary. We significantly advance over prior art and demonstrate that satisfactory recognition accuracies are yielded even in few-shot learning scenarios.

Building a Korean morphological analyzer using two Korean BERT models.

A morphological analyzer plays an essential role in identifying functional suffixes of Korean words. The analyzer input and output differ from each other in their length and strings, which can be dealt with by an encoder-decoder architecture. We adopt a Transformer architecture, which is an encoder-decoder architecture with self-attention rather than a recurrent connection, to implement a Korean morphological analyzer. Bidirectional Encoder Representations from Transformers (BERT) is one of the most popular pretrained representation models; it can present an encoded sequence of input words, considering contextual information. We initialize both the Transformer encoder and decoder with two types of Korean BERT, one of which is pretrained with a raw corpus, and the other is pretrained with a morphologically analyzed dataset. Therefore, implementing a Korean morphological analyzer based on Transformer is a fine-tuning process with a relatively small corpus. A series of experiments proved that parameter initialization using pretrained models can alleviate the chronic problem of a lack of training data and reduce the time required for training. In addition, we can determine the number of layers required for the encoder and decoder to optimize the performance of a Korean morphological analyzer.

Instagram as a Da’wah Medium for Al-Hasany Foundation Islamic Boarding School

Social media expands the volume of information for each individual so that it holds power over control. Human interaction through online media is not limited by time and space, as long as the user is connected to an internet connection. Unlimited interaction is a reason for an Islamic boarding school manager to create a @alhasanymedia Instagram account. This study explores the use of Instagram by Al-Hasany Islamic boarding school for da’wah. The research uses a qualitative field method, the research procedure assembles descriptive data of interview or written word sequences from the observed people, and this approach is directed at the individual setting in a holistic (whole) manner. In qualitative research, the author becomes an essential instrument (human instrument) to establish the focus of the research, find and select informants as the data source, collect data, assess data quality, analyze data, interpret data and make conclusions on the findings. The results of this study show that the Instagram account @alhasanymedia has contributed to Islamic boarding school as an educational institution which includes the media of da’wah, information, blessing, promotion, nostalgically, and documentation for pesantren and graduates of Al-Hasany Sumber Islamic boarding school, parents and Abituren.

P125 The advancement of digital health technology in the outpatient setting, is speech recognition software useful?

Abstract Background/Aims Over the last 20 years, innovation in digital health technology have been propelling health systems forwards. With the advent of the electronic health record, digital prescribing and artificial intelligence and machine learning comes new opportunities. The use of voice recognition systems in the transcription of rheumatology clinics has the potential for cost and time effectiveness which is ever important within the NHS. Here we review the use, user-interface acceptability and potential cost savings. Methods In speech recognition software the speech is converted to a sequence of words in written text. This is not a new technology and in particular has been widely used within radiology departments across the UK. Rheumatology is largely a clinic-based specialty, with high output of clinic letters. Speech recognition software removes the need for transcription-based services. As a new technology we have reviewed its use within the rheumatology, dermatology, endocrinology, renal and paediatric departments at a large district general hospital. Results 143 staff members have been provided with a license and on average 64% are regularly using the available software. Over 3 months an average of 14,843 minutes per month of dictation has been completed. It has been estimated that over 12 months’ use there has been £100,000 savings through administerial time saved. The average turnaround time for clinic letters from dictation to delivery (electronically) to GPs has improved from approximately 1-2 weeks (but delays up to 4 weeks seen) to an approximately 24-36 hours. Preliminary data from users of this software suggest that compared to traditional dictation devices: 33% found it ‘much better’, 33% ‘better’ and 33% ‘the same’. 67% of participants found it ‘easy, or very easy’ to use whilst 16% found it ‘difficult’. 50% of participants found it ‘usually’ saved time, 33% ‘sometimes’ and 16% ‘rarely’ saved time. Conclusion This preliminary data suggest potential time and cost savings with largely positive feedback from users. Further work is needed to assess potential patient safety issues including errors and inaccuracies compared to traditional dictation means. We aim to complete this work prior to presentation of this abstract should it be accepted. Disclosure H. Crawshaw: None. S. Jamal: None. R. Andev: None.

Implementation of quantum stochastic walks for function approximation, two-dimensional data classification, and sequence classification

We study a quantum stochastic neural network (QSNN) based on quantum stochastic walks on a graph, and use gradient descent to update the network parameters. We apply a toy model of QSNN with a few neurons to the problems of function approximation, two-dimensional data classification, and sequence classification. A simple QSNN with five neurons is trained to determine whether a sequence of words is a sentence or not, and we find that a QSNN can reduce the number of training steps. A QSNN with 11 neurons shows a quantum advantage in improving the accuracy of recognizing new types of inputs like verses. Moreover, with our toy model, we find the coherent QSNN is more robust against both label noise and device noise, compared with the decoherent QSNN. These results show that quantum stochastic walks may be a useful resource to implement a quantum neural network.

Intonational Cues to Segmental Contrasts in the Native Language Facilitate the Processing of Intonational Cues to Lexical Stress in the Second Language

This study examines whether second language (L2) learners' processing of an intonationally cued lexical contrast is facilitated when intonational cues signal a segmental contrast in the native language (L1). It does so by investigating Seoul Korean and French listeners' processing of intonationally cued lexical-stress contrasts in English. Neither Seoul Korean nor French has lexical stress; instead, the two languages have similar intonational systems where prominence is realized at the level of the Accentual Phrase. A critical difference between the two systems is that French has only one tonal pattern underlying the realization of the Accentual Phrase, whereas Korean has two underlying tonal patterns that depend on the laryngeal feature of the phrase-initial segment. The L and H tonal cues thus serve to distinguish segments at the lexical level in Korean but not in French; Seoul Korean listeners are thus hypothesized to outperform French listeners when processing English lexical stress realized only with (only) tonal cues (H* on the stressed syllable). Seoul Korean and French listeners completed a sequence-recall task with four-item sequences of English words that differed in intonationally cued lexical stress (experimental condition) or in word-initial segment (control condition). The results showed higher accuracy for Seoul Korean listeners than for French listeners only when processing English lexical stress, suggesting that the processing of an intonationally cued lexical contrast in the L2 is facilitated when intonational cues signal a segmental contrast in the L1. These results are interpreted within the scope of the cue-based transfer approach to L2 prosodic processing.

Natural Language Description Generation Method of Intelligent Image Internet of Things Based on Attention Mechanism

With the rapid development of Internet of Things technology, the image data on the Internet are growing at an amazing speed. How to describe the semantic content of massive image data is facing great challenges. Attentional mechanisms originate from the study of human vision. In cognitive science, due to bottlenecks in information processing, humans selectively attend to a portion of all information while ignoring the rest of the visible information. This study mainly discusses the natural language description generation method of Internet of Things intelligent image based on attention mechanism. In this study, a CMOS sensor based on Internet of Things technology is used for image data acquisition and display. FPGA samples cis16bit parallel port data, writes FIFO, stores image data, and then transmits it to host computer for display through network interface. In order to minimize the value of cross-entropy loss function, maximum-likelihood estimation is used to maximize the joint probability of word sequences in the language model when sentence descriptions are generated using the encoder-decoder framework. At each moment, in addition to image features, additional text features are input. Image feature vector and text feature vector are weighted and summed by attention mechanism at each time. In decoding, the attention mechanism gives each image region feature weight, and the long-term and short-term memory network decodes in turn, but the long-term and short-term memory network has limited decoding ability. We use bidirectional long-term and short-term memory network instead of long-term memory network, and dynamically focus on context information through forward LSTM and reverse LSTM. The specificity of the proposed network is 5% higher than that of the 3D convolution residual link network. The results show that the performance of image description model is improved by inputting image context and text context into long-term memory network decoder.