Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18625 papers
We evaluate two cross-lingual techniques for adding enhanced dependencies to existing treebanks in Universal Dependencies. We apply a rule-based system developed for English and a data-driven system trained on Finnish to Swedish and Italian. We find that both systems are accurate enough to bootstrap enhanced dependencies in existing UD treebanks. In the case of Italian, results are even on par with those of a prototype language-specific system.
Abstract Contrary to what non-practitioners might expect, the systems of phonetic notation used by linguists are highly idiosyncratic. Not only do various linguistic subfields disagree on the specific symbols they use to denote the speech sounds of languages, but also in large databases of sound inventories considerable variation can be found. Inspired by recent efforts to link cross-linguistic data with help of reference catalogues (Glottolog, Concepticon) across different resources, we present initial efforts to link different phonetic notation systems to a catalogue of speech sounds. This is achieved with the help of a database accompanied by a software framework that uses a limited but easily extendable set of non-binary feature values to allow for quick and convenient registration of different transcription systems, while at the same time linking to additional datasets with restricted inventories. Linking different transcription systems enables us to conveniently translate between different phonetic transcription systems, while linking sounds to databases allows users quick access to various kinds of metadata, including feature values, statistics on phoneme inventories, and information on prosody and sound classes. In order to prove the feasibility of this enterprise, we supplement an initial version of our cross-linguistic database of phonetic transcription systems (CLTS), which currently registers five transcription systems and links to fifteen datasets, as well as a web application, which permits users to conveniently test the power of the automatic translation across transcription systems.
We conducted a semantic similarity study of semantic concepts in the context of the Holy Book Quran. Semantic similarity examines the degree of likeness and shared common properties of two concepts. For example, the Quranic concept of Allah and God will result in a high score of semantic similarity, whereas hell and paradise will yield in a low score because of its extremely different attributes and semantic features. Apart from that, we also delivered the Quranic concept semantic similarity standard dataset which consists of some pairs of Quranic concept along with its similarity score, which was manually annotated by human raters. This dataset resulted in the score of inter-annotator agreement 0.63, not far from the the ones yielded by some well-known datasets such as WordSim and Simlex. Furthermore, to measure the semantic similarity score, we chose the knowledge-based approach by utilizing lexical database properties such as the length and depth of a synonym set (synset). We then applied it to Yuhua Li equation, which has been considered to be the baseline among researchers within the problem of semantic similarity. In terms of the result, our system gained Pearson's correlation 0.33 and Spearman's 0.19. By considering inter-annotator agreement 0.63 that our Quranic standard dataset has as the upper bound score, there are still quite large room for improvement to better mimicking Muslim's intuition to measure the degree of similarity of concepts within the domain of Quran.
Many recent studies, academic and non-academic alike, have argued that the use of Urdu in Bollywood has started to decline. These studies, important as they are, however, suffer from some limitations. They are either impressionistic or based on non-representative data. Furthermore, they do not specify the object of the study or the site of the assumed decline of Urdu. Therefore, it remains vague which element of Urdu, for example sounds, words, syntax or script is under investigation. Similarly, it is not clear which component of film for example titles, dialogues or songs are experiencing the decline. Fulfilling this research gap this paper makes two contributions. Analyzing songs from 1959 to 2010’s, it empirically demonstrates the decline by documenting the shift in the pronunciation of the sounds /kh̲/, /gh̲/, and /q/ from the Urdu to Hindi phonetic norms. Singers from the 1990’s, unlike those from the previous generations, merge them with the sounds /kh/, /gh/, and /q/. The paper also makes a methodological contribution in that it shows how language in cinema can be studied empirically using a corpus.
Cilin is one of the most popular semantic knowledge bases in Chinese information processing. Due to its coding scheme and taxonomical arrangement, some semantic relations among words are not explicitly shown. Our work aims to characterize its semantic relations by adding tags and compound codes to optimize the taxonomy and hierarchy of Cilin. Experiment results show that using the tag-augmented Cilin as a knowledge base improves the performance in the task of semantic similarity measurement.
This lexical cognate data was exported from the Indo-European Lexical Database (IELex) supporting the publication: Verkerk, Annemarie. (2018). Detecting non-tree-like signal using multiple tree topologies. Journal of Historical Linguistics. If you are looking to use the IELex data in your own research, we recommend you use the latest version, which is available along with documentation, phylogenetic tree samples and example BEAST control scripts.
This paper addresses the scalability challenge of architecture search by formulating the task in a differentiable manner. Unlike conventional approaches of applying evolution or reinforcement learning over a discrete and non-differentiable search space, our method is based on the continuous relaxation of the architecture representation, allowing efficient search of the architecture using gradient descent. Extensive experiments on CIFAR-10, ImageNet, Penn Treebank and WikiText-2 show that our algorithm excels in discovering high-performance convolutional architectures for image classification and recurrent architectures for language modeling, while being orders of magnitude faster than state-of-the-art non-differentiable techniques. Our implementation has been made publicly available to facilitate further research on efficient architecture search algorithms.
In the central nervous system the neuropeptide oxytocin mediates a range of behaviors related primarily to emotionality. One factor that influences oxytocinergic communication in the human brain and correlates with emotional behaviors is the single nucleotide polymorphism rs53576 on the oxytocin receptor gene (OXTR). For example, variations in this OXTR genotype are related to parental, altruistic, and other prosocial behaviors. Electroencephalographic waveforms of visually evoked response potentials recorded at the midline parietal electrode site display a prominent component putatively involved with attention allocation called the late positive potential. The magnitude of the late positive potential was found to be significantly higher in homozygous G allele individuals compared with A allele carriers when viewing negative emotionally charged images. Inversely, A allele carriers rated these negative images as more arousing, when measured by the Self-Assessment Manikin rating scale. These data suggest that OXTR functioning contributes to visual processing and subjective experience of negative stimuli.
Developed by the CLLD project with support from the Department of Linguistic and Cultural Evolution of the Max Planck Institute for the Science of Human History.
The significance of tourism within the ASEAN region is recognised by multiple stakeholders. Presenting and promoting a distinctive image of tourism is a common agenda across the ten ASEAN nations. The aim of this study is to document and interpret the emotional connotations of ASEAN tourism slogans. Arguably, such messages provide an initial guide to the appeal and competitive advantages of each individual country. The study is underpinned by considering key ideas on destination positioning and the lexical analysis of emotions. By mining archival resources about word frequencies, synonyms and meanings, the positions of the slogans in an emotion space originally developed by Plutchik were compared and plotted. Joy, admiration and ecstasy were the dominant emotional connotations of most slogans. Thailand and Malaysia have the most distinctive tourism slogans, followed by Vietnam and Laos. Expressions used in the slogans for these four nations overlapped less with other countries across the families of emotion words.
We present the Uppsala system for the CoNLL 2018 Shared Task on universal dependency parsing. Our system is a pipeline consisting of three components: the first performs joint word and sentence segmentation; the second predicts part-ofspeech tags and morphological features; the third predicts dependency trees from words and tags. Instead of training a single parsing model for each treebank, we trained models with multiple treebanks for one language or closely related languages, greatly reducing the number of models. On the official test run, we ranked 7th of 27 teams for the LAS and MLAS metrics. Our system obtained the best scores overall for word segmentation, universal POS tagging, and morphological features.
We know very little about how neural language models (LM) use prior linguistic context.In this paper, we investigate the role of context in an LSTM LM, through ablation studies.Specifically, we analyze the increase in perplexity when prior context words are shuffled, replaced, or dropped.On two standard datasets, Penn Treebank and WikiText-2, we find that the model is capable of using about 200 tokens of context on average, but sharply distinguishes nearby context (recent 50 tokens) from the distant history.The model is highly sensitive to the order of words within the most recent sentence, but ignores word order in the long-range context (beyond 50 tokens), suggesting the distant past is modeled only as a rough semantic field or topic.We further find that the neural caching model (Grave et al., 2017b) especially helps the LSTM to copy words from within this distant context.Overall, our analysis not only provides a better understanding of how neural LMs use their context, but also sheds light on recent success from cache-based models.
This paper describes our system (HIT-SCIR) submitted to the CoNLL 2018 shared task on Multilingual Parsing from Raw Text to Universal Dependencies. We base our submission on Stanford's winning system for the CoNLL 2017 shared task and make two effective extensions: 1) incorporating deep contextualized word embeddings into both the part of speech tagger and dependency parser; 2) ensembling parsers trained with different initialization. We also explore different ways of concatenating treebanks for further improvements. Experimental results on the development data show the effectiveness of our methods. In the final evaluation, our system was ranked first according to LAS (75.84%) and outperformed the other systems by a large margin.
In this paper, we propose RNN-Capsule, a capsule model based on Recurrent Neural Network (RNN) for sentiment analysis. For a given problem, one capsule is built for each sentiment category e.g., 'positive' and 'negative'. Each capsule has an attribute, a state, and three modules: representation module, probability module, and reconstruction module. The attribute of a capsule is the assigned sentiment category. Given an instance encoded in hidden vectors by a typical RNN, the representation module builds capsule representation by the attention mechanism. Based on capsule representation, the probability module computes the capsule's state probability. A capsule's state is active if its state probability is the largest among all capsules for the given instance, and inactive otherwise. On two benchmark datasets (i.e., Movie Review and Stanford Sentiment Treebank) and one proprietary dataset (i.e., Hospital Feedback), we show that RNN-Capsule achieves state-of-the-art performance on sentiment classification. More importantly, without using any linguistic knowledge, RNN-Capsule is capable of outputting words with sentiment tendencies reflecting capsules' attributes. The words well reflect the domain specificity of the dataset.
We introduce a novel architecture for dependency parsing: stack-pointer networks (STACKPTR). Combining pointer networks The stack tracks the status of the depthfirst search and the pointer networks select one child for the word at the top of the stack at each step. The STACKPTR parser benefits from the information of the whole sentence and all previously derived subtree structures, and removes the leftto-right restriction in classical transitionbased parsers. Yet, the number of steps for building any (including non-projective) parse tree is linear in the length of the sentence just as other transition-based parsers, yielding an efficient decoding algorithm with O(n 2 ) time complexity. We evaluate our model on 29 treebanks spanning 20 languages and different dependency annotation schemas, and achieve state-of-theart performance on 21 of them.
Recently, researchers have developed black-box approaches to mine design and interaction data from mobile apps. Although the data captured during this interaction mining is descriptive, it does not expose the design semantics of UIs: what elements on the screen mean and how they are used. This paper introduces an automatic approach for generating semantic annotations for mobile app UIs. Through an iterative open coding of 73k UI elements and 720 screens, we contribute a lexical database of 25 types of UI components, 197 text button concepts, and 135 icon classes shared across apps. We use this labeled data to learn code-based patterns to detect UI components and to train a convolutional neural network that distinguishes between icon classes with 94% accuracy. To demonstrate the efficacy of our approach at scale, we compute semantic annotations for the 72k unique UIs in the Rico dataset, assigning labels for 78% of the total visible, non-redundant elements.
As the amount of unstructured text data that humanity produces overall and on the Internet grows, so does the need to intelligently to process it and extract different types of knowledge from it. Convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have been applied to natural language processing systems with comparative, remarkable results. The CNN is a noble approach to extract higher level features that are invariant to local translation. However, it requires stacking multiple convolutional layers in order to capture long-term dependencies, due to the locality of the convolutional and pooling layers. In this paper, we describe a joint CNN and RNN framework to overcome this problem. Briefly, we use an unsupervised neural language model to train initial word embeddings that are further tuned by our deep learning network, then, the pre-trained parameters of the network are used to initialize the model. At a final stage, the proposed framework combines former information with a set of feature maps learned by a convolutional layer with long-term dependencies learned via long-short-term memory. Empirically, we show that our approach, with slight hyperparameter tuning and static vectors, achieves outstanding results on multiple sentiment analysis benchmarks. Our approach outperforms several existing approaches in term of accuracy; our results are also competitive with the state-of-the-art results on the Stanford Large Movie Review data set with 93.3% accuracy, and the Stanford Sentiment Treebank data set with 48.8% fine-grained and 89.2% binary accuracy, respectively. Our approach has a significant role in reducing the number of parameters and constructing the convolutional layer followed by the recurrent layer as a substitute for the pooling layer. Our results show that we were able to reduce the loss of detailed, local information and capture long-term dependencies with an efficient framework that has fewer parameters and a high level of performance.
To approximately parse an unfamiliar language, it helps to have a treebank of a similar language. But what if the closest available treebank still has the wrong word order? We show how to (stochastically) permute the constituents of an existing dependency treebank so that its surface part-of-speech statistics approximately match those of the target language. The parameters of the permutation model can be evaluated for quality by dynamic programming and tuned by gradient descent (up to a local optimum). This optimization procedure yields trees for a new artificial language that resembles the target language. We show that delexicalized parsers for the target language can be successfully trained using such "made to order" artificial languages.
This paper presents a truly full character-level neural dependency parser together with a newly released character-level dependency treebank for Chinese, which has suffered a lot from the dilemma of defining word or not to model character interactions. Integrating full character-level dependencies with character embedding and human annotated character-level part-of-speech and dependency labels for the first time, we show an extra performance enhancement from the evaluation on Chinese Penn Treebank and SJTU (Shanghai Jiao Tong University) Chinese Character Dependency Treebank and the potential of better understanding deeper structure of Chinese sentences.
In literature, writers have the liberty to deviate from linguistic norms under a principle known as poetic license. Poetic license allows deviation in favour of making language inspiring. Deviation from linguistic norms often implies that writers can take liberties with word formation, thus neology in literary contexts should be addressed specifically. This article analyses the status of literary coinages in the scope of neology and describes the specific context of children’s literature. The article also offers a typology of nonce formation processes for occasionalisms, with textual analysis, from a corpus of children’s books, using J. Tournier’s matrices of lexicogenesis [2007: 51].
This paper presents results from the first statistical dependency parser for Turkish. Turkish is a free-constituent order language with complex agglutinative inflectional and derivational morphology and presents interesting challenges for statistical parsing, as in general, dependency relations are between “portions” of words – called inflectional groups. We have explored statistical models that use different representational units for parsing. We have used the Turkish Dependency Treebank to train and test our parser but have limited this initial exploration to that subset of the treebank sentences with only left-to-right non-crossing dependency links. Our results indicate that the best accuracy in terms of the dependency relations between inflectional groups is obtained when we use inflectional groups as units in parsing, and when contexts around the dependent are employed.
Released only a year ago as the outputs of a research project (``Parsing Web 2.0 Sentences'', supported in part by a TÜBİTAK 1001 grant (No. 112E276) and a part of the ICT COST Action PARSEME (IC1207)), IMST and IWT are currently the most comprehensive Turkish dependency treebanks in the literature. This article introduces the final states of our treebanks, as well as a newly integrated hierarchical categorization of the multiheaded dependencies and their organization in an exclusive deep dependency layer in the treebanks. It also presents the adaptation of recent studies on standardizing multiword expression and named entity annotation schemes for the Turkish language and integration of benchmark annotations into the dependency layers of our treebanks and the mapping of the treebanks to the latest Universal Dependencies (v2.0) standard, ensuring further compliance with rising universal annotation trends. In addition to significantly boosting the universal recognition of Turkish treebanks, our recent efforts have shown an improvement in their syntactic parsing performance (up to 77.8{\%}/82.8{\%} LAS and 84.0{\%}/87.9{\%} UAS for IMST/IWT, respectively). The final states of the treebanks are expected to be more suited to different natural language processing tasks, such as named entity recognition, multiword expression detection, transfer-based machine translation, semantic parsing, and semantic role labeling.
Recent work on the problem of latent tree learning has made it possible to train neural networks that learn to both parse a sentence and use the resulting parse to interpret the sentence, all without exposure to ground-truth parse trees at training time. Surprisingly, these models often perform better at sentence understanding tasks than models that use parse trees from conventional parsers. This paper aims to investigate what these latent tree learning models learn. We replicate two such models in a shared codebase and find that (i) only one of these models outperforms conventional tree-structured models on sentence classification, (ii) its parsing strategies are not especially consistent across random restarts, (iii) the parses it produces tend to be shallower than standard Penn Treebank (PTB) parses, and (iv) they do not resemble those of PTB or any other semantic or syntactic formalism that the authors are aware of.
Previous studies have demonstrated differential perception of body expressions between males and females. However, only two recent studies (Kret et al., 2011; Krüger et al., 2013) explored the interaction effect between observer gender and subject gender, and it remains unclear whether this interaction between the two gender factors is gender-congruent (i.e., better recognition of emotions expressed by subjects of the same gender) or gender-incongruent (i.e., better recognition of emotions expressed by subjects of the opposite gender). Here, we used event-related potentials (ERPs) to investigate the recognition of fearful and angry body expressions posed by males and females. Male and female observers also completed an affective rating task (including valence, intensity, and arousal ratings). Behavioral results showed that male observers reported higher arousal rating scores for angry body expressions posed by females than males. ERP data showed that when recognizing angry body expressions, female observers had larger P1 for male than female bodies, while male observers had larger P3 for female than male bodies. These results indicate gender-incongruent effects in early and later stages of body expression processing, which fits well with the evolutionary theory that females mainly play a role in care of offspring while males mainly play a role in family guarding and protection. Furthermore, it is found that in both angry and fearful conditions male observers exhibited a larger N170 for male than female bodies, and female observers showed a larger N170 for female than male bodies. This gender-incongruent effect in the structural encoding stage of processing may be due to the familiarity of the body configural features of the same gender. The current results provide insights into the significant role of gender in body expression processing, helping us understand the issue of gender vulnerability associated with psychiatric disorders characterized by deficits of body language reading.
We present an architecture based on neural networks to generate natural language from unordered dependency trees. The task is split into the two subproblems of word order prediction and morphology inflection. We test our model gold corpus (the Italian portion of the Universal Dependency treebanks) and an automatically parsed corpus from the Web.
Chronic pain may alter both affect- and value-related behaviors, which represents a potentially treatable aspect of chronic pain experience. Current understanding of how chronic pain influences the function of brain reward systems, however, is limited. Using a monetary incentive delay task and functional magnetic resonance imaging (fMRI), we measured neural correlates of reward anticipation and outcomes in female participants with the chronic pain condition of fibromyalgia (N = 17) and age-matched, pain-free, female controls (N = 15). We hypothesized that patients would demonstrate lower positive arousal, as well as altered reward anticipation and outcome activity within corticostriatal circuits implicated in reward processing. Patients demonstrated lower arousal ratings as compared with controls, but no group differences were observed for valence, positive arousal, or negative arousal ratings. Group fMRI analyses were conducted to determine predetermined region of interest, nucleus accumbens (NAcc) and medial prefrontal cortex (mPFC), responses to potential gains, potential losses, reward outcomes, and punishment outcomes. Compared with controls, patients demonstrated similar, although slightly reduced, NAcc activity during gain anticipation. Conversely, patients demonstrated dramatically reduced mPFC activity during gain anticipation-possibly related to lower estimated reward probabilities. Further, patients demonstrated normal mPFC activity to reward outcomes, but dramatically heightened mPFC activity to no-loss (nonpunishment) outcomes. In parallel to NAcc and mPFC responses, patients demonstrated slightly reduced activity during reward anticipation in other brain regions, which included the ventral tegmental area, anterior cingulate cortex, and anterior insular cortex. Together, these results implicate altered corticostriatal processing of monetary rewards in chronic pain.
We explore dynamic evaluation, where sequence models are adapted to the recent sequence history using gradient descent, assigning higher probabilities to re-occurring sequential patterns. We develop a dynamic evaluation approach that outperforms existing adaptation approaches in our comparisons. We apply dynamic evaluation to outperform all previous word-level perplexities on the Penn Treebank and WikiText-2 datasets (achieving 51.1 and 44.3 respectively) and all previous character-level cross-entropies on the text8 and Hutter Prize datasets (achieving 1.19 bits/char and 1.08 bits/char respectively).
We present a novel abstractive summarization framework that draws on the recent development of a treebank for the Abstract Meaning Representation (AMR). In this framework, the source text is parsed to a set of AMR graphs, the graphs are transformed into a summary graph, and then text is generated from the summary graph. We focus on the graph-to-graph transformation that reduces the source semantic graph into a summary graph, making use of an existing AMR parser and assuming the eventual availability of an AMR-to-text generator. The framework is data-driven, trainable, and not specifically designed for a particular domain. Experiments on gold-standard AMR annotations and system parses show promising results. Code is available at: https://github.com/summarization
We evaluate corpus-based measures of linguistic complexity obtained using Universal Dependencies (UD) treebanks. We propose a method of estimating robustness of the complexity values obtained using a given measure and a given treebank. The results indicate that measures of syntactic complexity might be on average less robust than those of morphological complexity. We also estimate the validity of complexity measures by comparing the results for very similar languages and checking for unexpected differences. We show that some of those differences that arise can be diminished by using parallel treebanks and, more importantly from the practical point of view, by harmonizing the languagespecific solutions in the UD annotation.
In this paper, we describe our project of building a phrase structure treebank for Persian. The treebank consists of approximately 30000 sentences. With the help of this treebank, the researcher can investigate syntactic phenomena, extract grammars, train and test parsers, etc. In addition to these motivations, as another advantage of it we can refer to the fact that the sentences of this treebank are selected from an available dependency treebank. So the final treebank has two syntactic representations: phrase structure and dependency structure. The treebank is built using a bootstrapping approach, which converts a dependency structure tree to a phrase structure tree and the annotations are corrected manually. Using the new phrase structure treebank, we train models for constituency parsers. The treebank is freely available for educational purposes <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup>.
The paper describes the conversion of an LFG treebank of Polish into enhanced Universal Dependencies, and—more generally—identifies the kinds of information lost in translation from LFG to UD. The paper also presents the resulting UD treebank of Polish and compares it to the previous UD treebank of Polish.
В статье представлен обзор исследований, посвященных проблеме территориальной диф- ференциации языка в ортологическом аспекте. Установлено, что в связи с языковой нормой лин- гвисты выделяют три типа регионального варьирования: 1) сосуществование отдельных локаль- но маркированных единиц в одной нормативной системе; 2) дивергенцию диатопических орто- логических комплексов; 3) взаимодействие нормативных реализаций с их ненормативными диалектными аналогами в рамках определенного национального языка. Особое внимание уделя- ется ортологическому подходу к изучению регионального варьирования в отечественной лин- гвистике, а также особенностям использования локализмов в современной российской массовой коммуникации.
Past studies examining how people judge faces for trustworthiness and dominance have suggested that they use particular facial features (e.g. mouth features for trustworthiness, eyebrow and cheek features for dominance ratings) to complete the task. Here, we examine whether eye movements during the task reflect the importance of these features. We here compared eye movements for trustworthiness and dominance ratings of face images under three stimulus configurations: Small images (mimicking large viewing distances), large images (mimicking face to face viewing), and a moving window condition (removing extrafoveal information). Whereas first area fixated, dwell times, and number of fixations depended on the size of the stimuli and the availability of extrafoveal vision, and varied substantially across participants, no clear task differences were found. These results indicate that gaze patterns for face stimuli are highly individual, do not vary between trustworthiness and dominance ratings, but are influenced by the size of the stimuli and the availability of extrafoveal vision.
In this paper, we propose Dynamic Self-Attention (DSA), a new self-attention mechanism for sentence embedding. We design DSA by modifying dynamic routing in capsule network (Sabouretal.,2017) for natural language processing. DSA attends to informative words with a dynamic weight vector. We achieve new state-of-the-art results among sentence encoding methods in Stanford Natural Language Inference (SNLI) dataset with the least number of parameters, while showing comparative results in Stanford Sentiment Treebank (SST) dataset.
Deep neural networks (DNNs) have achieved impressive predictive performance due to their ability to learn complex, non-linear relationships between variables. However, the inability to effectively visualize these relationships has led to DNNs being characterized as black boxes and consequently limited their applications. To ameliorate this problem, we introduce the use of hierarchical interpretations to explain DNN predictions through our proposed method, agglomerative contextual decomposition (ACD). Given a prediction from a trained DNN, ACD produces a hierarchical clustering of the input features, along with the contribution of each cluster to the final prediction. This hierarchy is optimized to identify clusters of features that the DNN learned are predictive. Using examples from Stanford Sentiment Treebank and ImageNet, we show that ACD is effective at diagnosing incorrect predictions and identifying dataset bias. Through human experiments, we demonstrate that ACD enables users both to identify the more accurate of two DNNs and to better trust a DNN's outputs. We also find that ACD's hierarchy is largely robust to adversarial perturbations, implying that it captures fundamental aspects of the input and ignores spurious noise.
OBJECTIVE: Previous research has consistently shown that the ability to recognize emotions from facial expressions is impaired in Huntington's disease (HD). The aim of this study was to examine whether people with the gene expansion for HD visually scan the most emotionally informative features of human faces less than unaffected individuals, and whether altered visual scanning predicts emotion recognition in HD beyond general disease-related decline. METHOD: We recorded eye movements of 25 participants either in the late premanifest or early stage of HD and 25 age-matched healthy control participants during a face-viewing task. The task involved the viewing of pictures depicting human faces with angry, disgusted, fearful, happy, and neutral expressions, and evaluating each face on a valence rating scale. For data analysis, we defined 2 regions of interest (ROIs) on each picture, including an eye-ROI and a nose/mouth-ROI. Emotion recognition abilities were measured using an established emotion-recognition task and general disease-related decline was measured using the UHDRS motor score. RESULTS: Compared to the control participants, the HD participants spent less time looking at the ROIs relative to the total time spent looking at the pictures (partial η2 = 0.10), and made fewer fixations on the ROIs (partial η2 = 0.16). Furthermore, visual scanning of the eye-ROI, but not the nose/mouth-ROI, predicted emotion recognition performance in the HD group, over and beyond general disease-related decline. CONCLUSION: The emotion recognition deficit in HD may partly be explained by general disease-related decline in cognition and motor functioning and partly by a social-emotional deficit, which is reflected in reduced eye-viewing. (PsycINFO Database Record
This paper demonstrates an end-to-end Chinese discourse parser. We propose a unified framework based on recursive neural network (RvNN) to jointly model the subtasks including elementary discourse unit (EDU) segmentation, tree structure construction, center labeling, and sense labeling. Experimental results show our parser achieves the state-of-the-art performance in the Chinese Discourse Treebank (CDTB) dataset. We release the source code with a pre-trained model for the NLP community. To the best of our knowledge, this is the first open source toolkit for Chinese discourse parsing. The standalone toolkit can be integrated into subsequent applications without the need of external resources such as syntactic parser.
Failing to recognize one's mirror image can signal an abnormality in one's sense of self. In dissociative identity disorder (DID), individuals often report that their mirror image can feel unfamiliar or distorted. They also experience some of their own thoughts, emotions, and bodily sensations as if they are nonautobiographical and sometimes as if instead, they belong to someone else. To assess these experiences, we designed a novel backwards masking paradigm in which participants were covertly shown their own face, masked by a stranger's face. Participants rated feelings of familiarity associated with the strangers' faces. 21 control participants without trauma-generated dissociation rated masks, which were covertly preceded by their own face, as more familiar compared to masks preceded by a stranger's face. In contrast, across two samples, 28 individuals with DID and similar clinical presentations (DSM-IV Dissociative Disorder Not Otherwise Specified type 1) did not show increased familiarity ratings to their own masked face. However, their familiarity ratings interacted with self-reported identity state integration. Individuals with higher levels of identity state integration had response patterns similar to control participants. These data provide empirical evidence of aberrant self-referential processing in DID/DDNOS and suggest this is restored with identity state integration.
We propose a novel neural network model for joint part-of-speech (POS) tagging and dependency parsing. Our model extends the well-known BIST graph-based dependency parser (Kiperwasser and Goldberg, 2016) by incorporating a BiLSTM-based tagging component to produce automatically predicted POS tags for the parser. On the benchmark English Penn treebank, our model obtains strong UAS and LAS scores at 94.51% and 92.87%, respectively, producing 1.5+% absolute improvements to the BIST graph-based parser, and also obtaining a state-of-the-art POS tagging accuracy at 97.97%. Furthermore, experimental results on parsing 61 "big" Universal Dependencies treebanks from raw texts show that our model outperforms the baseline UDPipe (Straka and Straková, 2017) with 0.8% higher average POS tagging score and 3.6% higher average LAS score. In addition, with our model, we also obtain state-of-the-art downstream task scores for biomedical event extraction and opinion analysis applications. Our code is available together with all pre-trained models at: https://github.com/datquocnguyen/jPTDP
We unify recent neural approaches to one-shot learning with older ideas of associative memory in a model for metalearning. Our model learns jointly to represent data and to bind class labels to representations in a single shot. It builds representations via slow weights, learned across tasks through SGD, while fast weights constructed by a Hebbian learning rule implement one-shot binding for each new task. On the Omniglot, Mini-ImageNet, and Penn Treebank one-shot learning benchmarks, our model achieves state-of-the-art results.
بنك المشجّرات محلّل حاسوبيّ للظّواهر التّركيبيّة في اللّغة العربيّة، استثمر مبادئ نظريّة التّحكّم والرّبط التّوليديّة، وحوسباتها وتصوّراتها للنّحو الكلّيّ، غايته في ذلك بناء نظام حوسبيّ آليّ، يحاكي في اشتغاله النّظام الحوسبيّ اللّغويّ الطّبيعيّ. وقد حقّق بنك المشجّرات نتائج مهمّة في هذا الشّأن، تتمثّل في بلوغه الانتظام والتّناسق في معالجة الأبنية الإعرابيّة، لكنّ العمل لم يخل من هنات، أهمّها عدم اتّسام السّيرورة الاشتقاقيّة بالخاصّيّة التّكراريّة المميّزة للّغة البشريّة، وخرق حوسبة النّقل للقيود الجزبريّة التي أقرّتها النّظريّة اللّسانيّة، وهو ما يجعلنا نشكّك في كفايته الوصفيّة لسانيّا.
The contrast between the contextual and general meaning of a word serves as an important clue for detecting its metaphoricity. In this paper, we present a deep neural architecture for metaphor detection which exploits this contrast. Additionally, we also use cost-sensitive learning by re-weighting examples, and baseline features like concreteness ratings, POS and WordNet-based features. The best performing system of ours achieves an overall F1 score of 0.570 on All POS category and 0.605 on the Verbs category at the Metaphor Shared Task 2018.
Cross-domain sentiment classifiers aim to predict the polarity, namely the sentiment orientation of target text documents, by reusing a knowledge model learned from a different source domain. Distinct domains are typically heterogeneous in language, so that transfer learning techniques are advisable to support knowledge transfer from source to target. Distributed word representations are able to capture hidden word relationships without supervision, even across domains. Deep neural networks with memory (MemDNN) have recently achieved the state-of-the-art performance in several NLP tasks, including cross-domain sentiment classifica- tion of large-scale data. The contribution of this work is the massive experimentations of novel outstanding MemDNN architectures, such as Gated Recurrent Unit (GRU) and Differentiable Neural Computer (DNC) both in cross-domain and in-domain sentiment classification by using the GloVe word embeddings. As far as we know, only GRU neural networks have been applied in cross-domain sentiment classification. Senti- ment classifiers based on these deep learning architectures are also assessed from the viewpoint of scalability and accuracy by gradually increasing the training set size, and showing also the effect of fine-tuning, an ex- plicit transfer learning mechanism, on cross-domain tasks. This work shows that MemDNN based classifiers improve the state-of-the-art on Amazon Reviews corpus with reference to document-level cross-domain sen- timent classification. On the same corpus, DNC outperforms previous approaches in the analysis of a very large in-domain configuration in both binary and fine-grained document sentiment classification. Finally, DNC achieves accuracy comparable with the state-of-the-art approaches on the Stanford Sentiment Treebank dataset in both binary and fine-grained single-sentence sentiment classification.
Anger is considered a unique high-arousal and approach-related negative emotion. The influence of individual differences in trait anger on the processing of visual stimuli is relevant to questions about emotional processing and remains to be explored. Using functional magnetic resonance imaging (fMRI), we explored the neural responses to standardized images, selected based on valence and arousal ratings in a group of men with high trait anger compared to those with normative to low anger scores (controls). Results show increased activation in the left-lateralized ventral fronto-parietal attention network to unpleasant images by individuals with high trait anger. There was also a group by arousal interaction in the left thalamus/pulvinar such that individuals with high trait anger had increased pulvinar activation to the high-arousal (versus low arousal) unpleasant images as compared to controls. Thus, individual differences in trait anger in men are associated with brain regions subserving executive attentional and sensory integration during the processing of unpleasant emotional stimuli, particularly to high arousal images.