Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18625 papers
We develop an approach to growing deep network architectures over the course of training, driven by a principled combination of accuracy and sparsity objectives. Unlike existing pruning or architecture search techniques that operate on full-sized models or supernet architectures, our method can start from a small, simple seed architecture and dynamically grow and prune both layers and filters. By combining a continuous relaxation of discrete network structure optimization with a scheme for sampling sparse subnetworks, we produce compact, pruned networks, while also drastically reducing the computational expense of training. For example, we achieve $49.7\%$ inference FLOPs and $47.4\%$ training FLOPs savings compared to a baseline ResNet-50 on ImageNet, while maintaining $75.2\%$ top-1 accuracy -- all without any dedicated fine-tuning stage. Experiments across CIFAR, ImageNet, PASCAL VOC, and Penn Treebank, with convolutional networks for image classification and semantic segmentation, and recurrent networks for language modeling, demonstrate that we both train faster and produce more efficient networks than competing architecture pruning or search methods.
Grammatical models which represent the hierarchical structure of chord sequences have proven very useful in recent analyses of Jazz harmony. A critical resource for building and evaluating such models is a ground-truth database of syntax trees that encode hierarchical analyses of chord sequences. In this paper, we introduce the Jazz Harmony Treebank (JHT), a dataset of hierarchical analyses of complete Jazz standards. The analyses were created and checked by experts, based on lead sheets from the open iRealPro collection. The JHT is publicly available in JavaScript Object Notation (JSON), a human-understandable and machine-readable format for structured data. We additionally discuss statistical properties of the corpus and present a simple open-source web application for the graphical creation and editing of trees which was developed during the creation of the dataset.
The Neural Machine-Parsed IcePaHC is a machine-parsed treebank which consists of Icelandic texts from the 13th to 20th century, mostly Icelandic sagas. The texts were parsed using the IceNeuralParsingPipeline, a parsing pipeline which includes an Icelandic model of the Berkeley Neural Parser along with pre- and postprocessing steps. The parser was trained on IcePaHC and the parsing scheme of the treebank is therefore the same, although the treebank does not include empty phrases or lemmas. The treebank includes 52 texts. The total word count is 1,716,429 and the total number of clauses is 167,815.
Keywords play a major role in representing the gist of a document. Therefore, a lot of Natural Language processing tools have been implemented to identify keywords in both structured and unstructured texts. Text that appears in social media platforms such as twitter is mostly unstructured because of the character limitation. Consequently, a lot of short terms and symbols such as emoticons and URLs are included in tweets. Keyword extraction from grammatically ambiguous text is not easy compared to structured text since it is hard to rely on the linguistic features in unstructured texts. But when it comes to news on twitter, it may contain somewhat structured text than informal text does but it depends on the tweeter, the person who posts the tweet. In this paper, a methodology is proposed to extract keywords from a given tweet to retrieve relevant news that has been posted on twitter, for fake news detection. The intention of extracting keywords is to find more related news efficiently and effectively. For this approach, a corpus that contains tweet texts from different domains is built in order to make this approach more generic instead of making it a domain-specific approach. In fact, the Stanford Core NLP tool kit, Wordnet linguistic database and statistical method are used for extracting keywords from a tweet. For the system evaluation, the Turing test which has human intervention is used. The system was able to acquire an accuracy of 67.6% according to the evaluation conducted.
Affect fluctuates in a moment-to-moment fashion, reflecting the continuous relationship between the individual and the environment. Despite substantial research, there remain important open questions regarding how a stream of sensory input is dynamically represented in experienced affect. Here, approaching affect as a temporally dependent process, we show that momentary affect is shaped by a combination of the affective impact of stimuli (i.e., visual images for the current studies) and previously experienced affect. We also found that this temporal dependency is influenced by uncertainty of the affective context. Participants in each trial viewed sequentially presented images and subsequently reported their affective experience, which was modeled based on images' normative affect ratings and participants' previously reported affect. Study 1 showed that self-reported valence and arousal in a given trial is partly shaped by the affective impact of the given images and previously experienced affect. In Study 2, we manipulated context uncertainty by controlling occurrence probabilities for normatively pleasant and unpleasant images in separate blocks. Increasing context uncertainty (i.e., random occurrence of pleasant and unpleasant images) was associated with increased negative affect. In addition, the relative contribution of the most recent image to experienced pleasantness increased with increasing context uncertainty. Taken together, these findings provide clear behavioral evidence that momentary affect is a temporally dependent and continuous process, which reflects the affective impact of recent input variables and the previous internal state, and that this process is sensitive to the affective context and its uncertainty. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
This paper demonstrates how multilingual adolescents initiate language-directed teasing in family interaction and thus contribute to reinforcing or challenging social hierarchies and norms in the family. It investigates the case of a multilingual family living in Northern Norway (two parents and five children ranging from 3 to 18 years old). To a varying extent, and with varying degrees of competence, all family members use three languages in their daily lives: English, Spanish, and Norwegian. The data consists of self-recorded material of family interactions (9 h; 549 min) that were collected over the course of one year. A close interactional analysis shows how the siblings target linguistic production in teasing attacks, and use language-directed teasing as an interactional resource to position themselves and their family members. Drawing on Billig's (2005) theory of disciplinary humor, the article argues that playful corrections of perceived norm transgressions may be understood as situated (re)production and negotiation of social and linguistic norms, through which the young family members participate in the construction of the family as a community of practice.
In this paper, we introduce the first Universal Dependencies (UD) treebank for standard Albanian, consisting of 60 sentences collected from the Albanian Wikipedia, annotated with lemmas, universal part-of-speech tags, morphological features and syntactic dependencies. In addition to presenting the treebank itself, we discuss a selection of linguistic constructions in Albanian whose analysis in UD is not self-evident, including core arguments and the status of indirect objects, pronominal clitics, genitive constructions, prearticulated adjectives, and modal verbs.
Phonological segment borrowing is a process through which languages acquire new contrastive speech sounds as the result of borrowing new words from other languages. Despite the fact that phonological segment borrowing is documented in many of the world’s languages, to date there has been no large-scale quantitative study of the phenomenon. In this paper, we present SegBo, a novel cross-linguistic database of borrowed phonological segments. We describe our data aggregation pipeline and the resulting language sample. We also present two short case studies based on the database. The first deals with the impact of large colonial languages on the sound systems of the world’s languages; the second deals with universals of borrowing in the domain of rhotic consonants.
Olfactory research in humans has largely focused on odors perceived via sniffing, orthonasal olfaction, while ignoring odors perceived from the mouth, retronasal olfaction. Prior work on retronasally presented odors use animal models and focus mainly on odor sensitivity, but little is known about retronasal olfactory perception and cognition in humans. In this study, we compared orthonasal and retronasal odor presentation routes to investigate differences in odor descriptions and evaluations. Thirty-six individuals participated in a within-subjects study using twelve odors (varying in pleasantness and edibility) in perceptual and semantic tasks. As expected, we found route-dependent differences in perceptual odor quality (pleasantness, edibility, and familiarity ratings), and a better ability to identify odors during orthonasal presentation. Additionally, more concrete (and source-based) language was used when describing odors presented orthonasally. Exploratory analyses revealed that while orthonasal odors were described with words that had visual associations, retronasal odors were described with words that had interoceptive associations. Interestingly, these route-dependent differences in descriptor usage were not explained by sensitivity and intensity differences, suggesting cognitive and linguistic processes differed between odors presented orthonasally and retronasally. Our results indicate that olfaction is in fact a dual sense, in which the routes change the perception of an odor.
Previous studies have repeatedly demonstrated the attentional prioritization of emotional information over neutral information. However, the parsing of interference from negative and positive stimuli has not received the same attention. In the study reported here, we examined the effect of real-world visual scenes of neutral, positive, and negative valence, as well as the effect of both high-and low-arousal (differentially categorized based on their arousal and valence ratings) on scene gist identification. Using a partial-report paradigm, participants were asked to report the gist of a post-cued scene from a briefly-presented array of four scenes. Scene gist identification performance was significantly higher for positive scenes, regardless of arousal, than for negative scenes. All emotional scenes, regardless of valence and arousal, interfered with reporting the gist of neutral scenes. The findings support the hypothesis that emotional scenes more often interfere with processing of neutral scenes and are selectively attended to during briefly-presented scene arrays. Moreover, the results suggest that the identification and the interference of positive, higharousal scenes are prioritized in visual information processing.
Linguistics cannot obviate the fact that, according to a vast majority of the population not well versed in linguistic issues, normativity constitutes the key aspect both of languages and of language. The high number of daily consultations regarding language registered by the Espanol al dia Service of the Real Academia Espanola serves as an example of the great importance that popular metalinguistics confers upon speaking well. No wonder. The linguistic norm is a changing one and linguistic tradition provides valuable evidence that today’s mistakes can become tomorrow’s norm. It is no easy task though to identify these good usage models and the way in which they are established. Furthermore, it is difficult to establish how the so-called normative or prescriptive grammars - inevitably linked to the tasks performed by the Academias de la Lengua Espanola - reconcile the variation that is inherent to every linguistic system with the codification of the standard variety or varieties mainly used in the formal, administrative, institutional and academic contexts of communicative distance.
This paper focuses on parsing processes and principles, which are essential tasks for machines to understand syntactic and semantic structures of a sentence. Machine analysis procedures for Chinese sentences, including word segmentation, and part-of-speech tagging and parsing, were visually represented using Cparser, a rule-based constituency parser developed by Peking University. Then we explained how linguistic knowledge was embodied in Cparser lexical, syntactic and semantic components, and discussed their complex interplay that allows automatic parsing. As a practical example, a Chinese textbook treebank is also constructed using Cparser. According to the theoretical and practical discussion in this paper, Peking University Cparser can easily to reflect and modify undated linguistic knowledge and is expected to be widely used as an analysis and verification tool for Chinese grammar research.
This article discusses the modes of defining cultural content in teaching Polish as a foreign language. It proposes strategies for introducing cultural knowledge and knowledge about culture in the practice of teaching Polish as a non-native language. It seems appropriate to offer a division into:
 
 The cultural semantics of lexemes, set phrases and expressions;
 The culture (meaning the correct use) of expression and the linguistic norm; and
 Culture content and areas conveyed to foreigners through the language they are learning.
 
 This discussion fits the debate regarding the language which we teach to foreigners. In it, researchers have offered examples of semantic distortions (mostly shifts), and the dangers of private or rather extra-curricular linguistic contact with a language’s native users. It also carries reflections on linguistic regionalisation and its presence in the practice of teaching Polish as a non-native language.
RST-based discourse parsing is an important NLP task with numerous downstream\napplications, such as summarization, machine translation and opinion mining. In\nthis paper, we demonstrate a simple, yet highly accurate discourse parser,\nincorporating recent contextual language models. Our parser establishes the new\nstate-of-the-art (SOTA) performance for predicting structure and nuclearity on\ntwo key RST datasets, RST-DT and Instr-DT. We further demonstrate that\npretraining our parser on the recently available large-scale "silver-standard"\ndiscourse treebank MEGA-DT provides even larger performance benefits,\nsuggesting a novel and promising research direction in the field of discourse\nanalysis.\n
Current methods of cross-lingual parser transfer focus on predicting the best\nparser for a low-resource target language globally, that is, "at treebank\nlevel". In this work, we propose and argue for a novel cross-lingual transfer\nparadigm: instance-level parser selection (ILPS), and present a\nproof-of-concept study focused on instance-level selection in the framework of\ndelexicalized parser transfer. We start from an empirical observation that\ndifferent source parsers are the best choice for different Universal POS\nsequences in the target language. We then propose to predict the best parser at\nthe instance level. To this end, we train a supervised regression model, based\non the Transformer architecture, to predict parser accuracies for individual\nPOS-sequences. We compare ILPS against two strong single-best parser selection\nbaselines (SBPS): (1) a model that compares POS n-gram distributions between\nthe source and target languages (KL) and (2) a model that selects the source\nbased on the similarity between manually created language vectors encoding\nsyntactic properties of languages (L2V). The results from our extensive\nevaluation, coupling 42 source parsers and 20 diverse low-resource test\nlanguages, show that ILPS outperforms KL and L2V on 13/20 and 14/20 test\nlanguages, respectively. Further, we show that by predicting the best parser\n"at the treebank level" (SBPS), using the aggregation of predictions from our\ninstance-level model, we outperform the same baselines on 17/20 and 16/20 test\nlanguages.\n
Der Beitrag reflektiert sprachliche Normen und den Umgang mit innersprachlicher Variation im Deutschunterricht. Das Inklusionsparadigma fordert hier von den Lehrkräften eine gesteigerte Sprachreflexionskompetenz, die Exklusionsrisiken erkennt. Am Beispiel von Sprachideologien in Bezug auf Bildungs- und Standardsprache im Deutschunterricht wird die Problemlage illustriert, und es werden Schlussfolgerungen für die Lehrer*innenbildung gezogen. The article reflects linguistic norms and linguistic variation in German teaching and learning at school. The call for inclusion demands from teachers a greater ability to reflect language use and by that, identify risks of exclusion. Based on the analysis of language ideologies with a focus on academic and standard language in school, the article draws conclusions regarding teacher training under the conditions of inclusion.
espanolEn este trabajo, argumentamos que en los actuales treebanks que aplican el formalismo de las Dependencias Universales, la anotacion de los reflexivos espanoles es un problema sin resolver, que afecta claramente a la precision y consistencia de los parsers actuales. Evaluamos diferentes propuestas para afinar las diferentes categorias y discutimos los problemas pendientes. Creemos que la solucion para estos problemas se puede encontrar en una anotacion en multiples niveles, combinando la anotacion de la relacion de dependencia y de las caracteristicas (features) de los tokens, en lugar de ampliar el numero de categorias en un solo nivel de anotacion. Aplicamos la propuesta a la version espanola del treebank UD AnCora (v2.5) y proporcionamos una tabla de conversion categorizada que se puede ejecutar mediante un script Python. EnglishIn this paper, we argue that in current Universal Dependencies treebanks, the annotation of Spanish reflexives is an unsolved problem, which clearly affects the accuracy and consistency of current parsers. We evaluate different proposals for fine-tuning the various categories, and discuss remaining open issues. We believe that the solution for these issues could lie in a multi-layered way of annotating the characteristics, combining annotation of the dependency relation and of the so-called token features, rather than in expanding the number of categories on one layer. We apply this proposal to the v2.5 Spanish UD AnCora treebank and provide a categorized conversion table that can be run with a Python script.
Investigating the neural correlates of arousal in non-offending paedophiles provides an opportunity to focus on the dysfunctional process underlying paedophilic predilections. The study aimed to investigate the neural correlates of sexual arousal related to paedophilic preference. Functional magnetic resonance imaging (fMRI) and phallometric testing were used to determine sexual arousal related brain activity while listening to a series of auditory narratives including neutral, adult sexual, and child sexual content. Participants included 12 non-offending paedophiles and a control sample of 12 males with teleiophilic preferences. Blood oxygen level-dependent fMRI, phallometric response and subjective arousal ratings related to auditory stimuli. Subjective arousal ratings showed expected direction of preference for adult and child sexual content between control group and non-offending paedophiles respectively. Phallometric response during adult sexual content was higher in the control group although there was no difference between samples during the child sexual content. Imaging contrasts of the child relative to adult sexual content revealed functional activity in the cerebellum specific to non-offending paedophiles.
A wide variety of transition-based algorithms are currently used for dependency parsers. Empirical studies have shown that performance varies across different treebanks in such a way that one algorithm outperforms another on one treebank and the reverse is true for a different treebank. There is often no discernible reason for what causes one algorithm to be more suitable for a certain treebank and less so for another. In this paper we shed some light on this by introducing the concept of an algorithm's inherent dependency displacement distribution. This characterises the bias of the algorithm in terms of dependency displacement, which quantify both distance and direction of syntactic relations. We show that the similarity of an algorithm's inherent distribution to a treebank's displacement distribution is clearly correlated to the algorithm's parsing performance on that treebank, specifically with highly significant and substantial correlations for the predominant sentence lengths in Universal Dependency treebanks. We also obtain results which show a more discrete analysis of dependency displacement does not result in any meaningful correlations.
We experience our neighborhoods and cities through our streets. An essential question is what street features contribute to a superior pedestrian experience? Further, can interacting with urban streetscapes improve cognitive functioning, as interacting with nature does? To answer these questions, large urban street perception databases are enormously helpful. Toward this end, we propose a crowdsourcing method for measuring the pedestrian qualities of streets – preference, imageability, complexity, enclosure, human scale, transparency, and order, all of which are important design dimensions for pedestrian experience. We obtained 556 street images from Google Street View by sampling two sidewalk images from 278 geo-coordinates in Chicago. For each dimension, over 58 (SD=2.5) Amazon Mechanical Turk workers (469 in total) completed an image rating task for the 556 images. In each trial, participants were shown 12 images in a 4x3 grid and asked to choose four images that they evaluate highly on that dimension. The probability of selecting each image for a given dimension across participants was used to quantify how much that image represented that dimension. To test the inter-rater reliability, we randomly split participants into two groups 2000 times and calculated rank-correlations between the measures from each group. We found that the split-half correlations were high for walkability (0.86±0.01; M±SD), preference (0.83±0.02), imageability (0.78±0.02), complexity (0.79±0.02), enclosure (0.80±0.02), and transparency (0.88±0.01) and modest for human scale (0.45±0.05) and order (0.57±0.05). To test whether the measures are internally consistent, we randomly split two images from the same geo-coordinate into two bins (of 278 images) 2000 times, calculated rank-correlations, and found modest to low correlation values, range: [0.06, 0.37]. These results suggest that our method can be used to efficiently measure pedestrian street qualities from non-experts, but with some limitations, to build a large database for studying street features affecting urban street pedestrian experiences.
Existing approaches for graph-based deep dependency parsing that force connexity in predicted graphs do no cover the structures observed in French treebanks. We propose a novel algorithm that covers the full set of possible structures. We evaluate our approach on the French corpora FTB and Sequoia and observe a trade-off between the validity of predicted structures and the quality of predictions
Ignoring or withholding a response from a stimulus causes it to become affectively devalued. Leading accounts posit that this is due to negative affect elicited by neurocognitive inhibition when it is applied to resolve conflict from distracting or otherwise inappropriate stimulus/response representations. Other research, however, suggests that stimulus/response conflict may itself elicit negative affect and devalue stimuli, raising questions about whether effects previously attributed to inhibition may instead reflect the emotional impact of conflict, per se. To address this, we measured affective ratings of art-like patterns that previously appeared on critical trials of a task-switching paradigm (ABA vs. CBA task sequences) known for its capacity to distinguish behavioural effects of inhibition and conflict. Stimuli from the ABA-sequence experimental condition showing behavioural evidence of backward inhibition (n-2 repetition costs) received more negative ratings than those from the CBA-sequence control condition. This stimulus-devaluation effect was not impacted by the level of conflict associated with high uncertainty or low uncertainty about upcoming task order. Moreover, the response-time index of inhibition was larger on ABA trials in which the associated stimuli later received negative ratings than on trials preceding relatively positive ratings. Inhibition therefore appears to have negative affective consequences that exceed any emotional impact of conflict, with fluctuations in inhibition linked to subsequent stimulus evaluations.
An automatically derived AMR annotation layer of the FullStack multi-layer text corpus of Latvian. First, Latvian UD Treebank (v2.5) sentences were translated to English using a state-of-the-art Latvian-English neural MT system (Hugo.lv). Second, a state-of-the-art AMR parser for English (AMREager) was applied to the MT-translated sentences. Additionally, alignment information is provided for both Latvian-English translations and English-AMR parses.
In this study, it is our objective to carry out a historical tour of the main antecedents that we can find on the linguistic theories in Ferdinand de Saussure, with special emphasis on the influences he took for the elaboration of his theory of the sign. To do this, given the philosophicalrationalist nature that supports his theoretical conceptions, we are going to study the hypotheses preceding his, which had a logical-speculative nature. In this sense, we will start with Classical Antiquity focusing on the contributions made by the main Greek philosophers (Socrates, Platon and Aristotle) on the language / thought duality and the origin motivated or not of linguistic signs. Next, we will address the medieval theories of scholasticism and its conception of language as a syntactic and paradigmatic system in which agreement and rection were of fundamental importance, as Saussure would explain centuries later, categorizing language as a formal and functional system. Next, we will carry out an overview of the rationalist linguistic thought conceived by El Brocense in the 16th century and made explicit in his Minerva. From him, Saussure would take the conception that reason was above any use or linguistic norm that tried to limit language. Later, already located in the seventeenth century, we will study the general and reasoned Grammar of Port-Royal and its influence on Ferdinand de Saussure, especially with regard to the conception of the two faces of the linguistic sign (meaning and signifier). Finally, we will review some of the late nineteenth century theories that influenced Saussure and that were basically those conceived by the Kazan and Moscow schools and by the thought of the American linguist W. D. Whitney. Finally, we will expose some of the fundamental concepts contained in Ferdinand de Saussure's General Linguistics Course in which he presented his linguistic theories
Background: Obesity is a chronic disease characterized by a reduction in life expectancy. Bariatric surgery has been an alternative to conventional treatments, but lifestyle changes such as increased physical activity are crucial for achieving body image and health outcomes. Objective: Test the hypothesis that physical activity is associated with satisfaction with body image in obese individuals undergoing bariatric surgery. Methods: Cross-sectional study conducted in Salvador, BA, Brazil. All participants at 3 and 6 months postoperatively answered the Stunkard Image Rating Scales as the main outcome, in addition to the International Physical Activity Questionnaire to assess the physical activity profile. Results: Physically active individuals were 80% more likely to report body satisfaction when compared with physically inactive individuals. According to multivariate analysis, the adjusted physical activity for females and advanced age increased by 83% the chance of reporting satisfaction with body image and, when adjusted for marital status, body mass index, and surgery time, the strength of association increased to 89% the chance to refer body satisfaction when compared with physically inactive people. Conclusion: Physical activity was associated with better body image. These results indicate opportunities to improve outcomes in patients after bariatric surgery through counseling and treatment intervention.
With the popularity of the Internet, the public can get news from the recent hottest news, and express their opinion in time in the network of social media, such as microblog, twitter. With the help of sentiment analysis of the comments, the government or the media can inform of the public opinion, and make corresponding decisions, so that they are able to get the positive feedback. Also, the sentiment analysis can help people block the detrimental comments. Since the sentiment analysis is useful in the daily life, the author made an experiment about the sentiment analysis of three models, namely, Naive Byes, Maximum Entropy, and SVM, to compare the results' accuracy of them. And the dataset used in this experiment is Stanford Twitter Sentiment (STS). Besides, the reference is made to Stanford Sentiment Treebank and IMDB. In addition, the determination of the emotional tone is based on emotional dictionaries like GI (General Inquirer) and How Net. By comparing the accuracy and training time of different models, SVM is selected to be the optimal model with the F-scores of 82%.
This paper presents the first treebank for the Laz language, which is also the first Universal Dependencies Treebank for a South Caucasian language. This treebank aims to create a syntactically and morphologically annotated resource for further research. We also aim to document an endangered language in a systematic fashion within an inherently cross-linguistic framework: the Universal Dependencies Project (UD). As of now, our treebank consists of 576 sentences and 2,306 tokens annotated in light with the UD guidelines. We evaluated the treebank on the dependency parsing task using a pretrained multilingual parsing model, and the results are comparable with other low-resourced treebanks with no training set. We aim to expand our treebank in the near future to include 1,500 sentences. The bigger goal for our project is to create a set of treebanks for minority languages in Anatolia.
Electroencephalography (EEG)-based emotion recognition has advanced the field in affective computing and has enabled applications in human-computer interactions. Despite significant progress has been made in decoding emotion using supervised machine-learning methods, few studies applied data-driven, unsupervised approaches to explore the underlying EEG dynamics during an emotion experiment and examine how such dynamics correlate with subjective reports of emotion. This study employs the adaptive mixture independent component analysis (AMICA), an unsupervised approach, to EEG data from the DEAP dataset where 32 subjects watched emotional videos. Empirical results showed that AMICA could learn distinct models that separated EEG date collected in the emotion experiment. The identified changes in EEG patterns were weakly-correlated with the four reported emotion scales, indicating the underlying EEG dynamics partially reflected the emotional activities as well as the emotion-irrelevant brain dynamics. Further, the correlations between EEG dynamics and individuals' subjective emotional ratings were significantly higher than those between the EEG and the average ratings from online raters. Finally, building an emotion-decoding model based on the EEG dynamics revealed a significantly better classification performance for valence ratings compared to arousal. This study demonstrated the use of AMICA in characterizing the EEG dynamics in emotion experiments and provided insight into the relationship between EEG and the reported emotional experiences. The unsupervised learning approach can be applied to studying emotion and other confounding factors such as emotion irrelevant EEG artifacts, thereby improving the performance of emotion decoding for EEG-based affective computing.
Abstract This article presents a novel method to determine particular syntactical attributes of Ancient Greek oratory to quantitatively compare the orators of the Classical to those of the Imperial era based on their style of writing. The study first provides a philological overview of Classical Atticism and its Imperial counterpart and argues that the latter is the product of creative mimēsis and not a mere reproduction of archetypes. Then the article briefly explains a node-based metric method that was developed to quantify the morphology of a syntactically annotated Treebank that led to a more thorough weighting scheme using Haar Wavelets. Wavelets were then used to capture both the linear topology of a sentence and the tree network topology of the corresponding syntactical tree. The results were subsequently processed using principal component analysis to analyze and visualize the data. The method is demonstrated using a database of syntactically annotated sentences from six Attic orators.
The performance of a long short-term memory (LSTM) recurrent neural network (RNN)-based language model has been improved on language model benchmarks. Although a recurrent layer has been widely used, previous studies showed that an LSTM RNN-based language model (LM) cannot overcome the limitation of the context length. To train LMs on longer sequences, attention mechanism-based models have recently been used. In this paper, we propose a LM using a neural Turing machine (NTM) architecture based on localized content-based addressing (LCA). The NTM architecture is one of the attention-based model. However, the NTM encounters a problem with content-based addressing because all memory addresses need to be accessed for calculating cosine similarities. To address this problem, we propose an LCA method. The LCA method searches for the maximum of all cosine similarities generated from all memory addresses. Next, a specific memory area including the selected memory address is normalized with the softmax function. The LCA method is applied to pre-trained NTM-based LM during the test stage. The proposed architecture is evaluated on Penn Treebank and enwik8 LM tasks. The experimental results indicate that the proposed approach outperforms the previous NTM architecture.
This paper investigates whether typical stress patterns in English nouns and verbs are available as a prosodic cue for categorisation and accelerated word learning during first language acquisition. The stress typicality hypothesis states that left-stressed nouns and right-stressed verbs should be acquired earlier than the reverse configurations if stress effectively signals lexical class membership. In this view, class-typical stress patterns are expected to facilitate learning of novel items. A series of generalized additive models (GAMs) based on a comprehensive set of lexical data (CELEX) as well as a large set of age-of-acquisition (AoA) and concreteness ratings reveals that stress typicality plays a minor role in early acquisition, as it is generally superseded by a preference for left-hand (or 'trochaic') patterns in both nouns and verbs. This may be explained by general cognitive constraints (perceptual salience and recency) or exposure to the dominant pattern in the ambient language.
An interesting and frequent type of multiword expression (MWE) is the headless MWE, for which there are no true internal syntactic dominance relations; examples include many named entities ("Wells Fargo") and dates ("July 5, 2020") as well as certain productive constructions ("blow for blow", "day after day").Despite their special status and prevalence, current dependency-annotation schemes require treating such flat structures as if they had internal syntactic heads, and most current parsers handle them in the same fashion as headed constructions.Meanwhile, outside the context of parsing, taggers are typically used for identifying MWEs, but taggers might benefit from structural information.We empirically compare these two common strategies-parsing and tagging-for predicting flat MWEs.Additionally, we propose an efficient joint decoding algorithm that combines scores from both strategies.Experimental results on the MWE-Aware English Dependency Corpus and on six non-English dependency treebanks with frequent flat structures show that: (1) tagging is more accurate than parsing for identifying flat-structure MWEs, (2) our joint decoder reconciles the two different views and, for non-BERT features, leads to higher accuracies, and (3) most of the gains result from feature sharing between the parsers and taggers.
Sentiment analysis, especially for long documents, plausibly requires methods\ncapturing complex linguistics structures. To accommodate this, we propose a\nnovel framework to exploit task-related discourse for the task of sentiment\nanalysis. More specifically, we are combining the large-scale,\nsentiment-dependent MEGA-DT treebank with a novel neural architecture for\nsentiment prediction, based on a hybrid TreeLSTM hierarchical attention model.\nExperiments show that our framework using sentiment-related discourse\naugmentations for sentiment prediction enhances the overall performance for\nlong documents, even beyond previous approaches using well-established\ndiscourse parsers trained on human annotated data. We show that a simple\nensemble approach can further enhance performance by selectively using\ndiscourse, depending on the document length.\n
In neural machine translation (NMT), sequence distillation (SD) through creation of distilled corpora leads to efficient (compact and fast) models.However, its effectiveness in extremely low-resource (ELR) settings has not been well-studied.On the other hand, transfer learning (TL) by leveraging larger helping corpora greatly improves translation quality in general.This paper investigates a combination of SD and TL for training efficient NMT models for ELR settings, where we utilize TL with helping corpora twice: once for distilling the ELR corpora and then during compact model training.We experimented with two ELR settings: Vietnamese-English and Hindi-English from the Asian Language Treebank dataset with 18k training sentence pairs.Using the compact models with 40% smaller parameters trained on the distilled ELR corpora, greedy search achieved 3.6 BLEU points improvement in average while reducing 40% of decoding time.We also confirmed that using both the distilled ELR and helping corpora in the second round of TL further improves translation quality.Our work highlights the importance of stage-wise application of SD and TL for efficient NMT modeling for ELR settings.
BACKGROUND: How dental education influences students' dental and dentofacial esthetic perception has been studied for some time, given the importance of esthetics in dentistry. However, no study before has studied this question in a large sample of students from all grades of dental school. This study sought to fill that gap. The aim was to assess if students' dentofacial esthetic autoperception and heteroperception are associated with their actual stage of studies (grade) and if autoperception has any effect on heteroperception. METHODS: Between October 2018 and August 2019, a questionnaire was distributed to 919 dental students of all 5 grades of dental school at all four dental schools in Hungary. The questionnaire consisted of the following parts (see also the supplementary material): 1. Demographic data (3 items), Self-Esthetics I (11 multiple- choice items regarding the respondents' perception of their own dentofacial esthetics), Self-Esthetics II (6 Likert-type items regarding the respondents' perception of their own dentofacial esthetics), and Image rating (10 items, 5 images each, of which the respondents have to choose the one they find the most attractive). Both the self-esthetics and the photo rating items were aimed at the assessment of mini- and microesthetic features. RESULTS: The response rate was 93.7% (861 students). The self-perception of the respondents was highly favorable, regardless of grade or gender. Grade and heteroperception were significantly associated regarding maxillary midline shift (p < 0.01) and the relative visibility of the arches behind the lips (p < 0.01). Detailed analysis showed a characteristic pattern of preference changes across grades for both esthetic aspects. The third year of studies appeared to be a dividing line in both cases, after which a real preference order was established. Association between autoperception and heteroperception could not be verified for statistical reasons. CONCLUSION: Our findings corroborate the results of most previous studies regarding the effect of dental education on the dentofacial esthetic perception of students. We have shown that the effect can be demonstrated on the grade level, which we attribute to the specific curricular contents. We found no gender effect, which, in the light of the literature, suggests that the gender effect in dentofacial esthetic perception is highly culture dependent. The results allow no conclusion regarding the relation between autoperception and heteroperception.
The article analyzes differences in the description of discourse relations in corpus research, in particular with the reference to the use of discourse markers – expressions that tie together subsequent fragments of the text and provide information about the nature of these relations. The text presents three concepts of the description of explicitness and implicitness of the content: Rhetorical Structure Theory, Penn Discourse Treebank and the author’s original proposal and indicates the consequences of each solution. The analysis of relations with particles as metatexual expressions defined in accordance with The Nest Dictionary of Polish reveals the possibility of expressing explicitness as a representation of elements of informational structure shaped by the use of a given particle, and implicitness as a lack of representation of certain elements of this type.
Neural machine translation (NMT) models are typically trained using a softmax cross-entropy loss where the softmax distribution is compared against smoothed gold labels. In low-resource scenarios, NMT models tend to over-fit because the softmax distribution quickly approaches the gold label distribution. To address this issue, we propose to divide the logits by a temperature coefficient, prior to applying softmax, during training. In our experiments on 11 language pairs in the Asian Language Treebank dataset and the WMT 2019 English-to-German translation task, we observed significant improvements in translation quality by up to 3.9 BLEU points. Furthermore, softmax tempering makes the greedy search to be as good as beam search decoding in terms of translation quality, enabling 1.5 to 3.5 times speed-up. We also study the impact of softmax tempering on multilingual NMT and recurrently stacked NMT, both of which aim to reduce the NMT model size by parameter sharing thereby verifying the utility of temperature in developing compact NMT models. Finally, an analysis of softmax entropies and gradients reveal the impact of our method on the internal behavior of NMT models.
This journal article follows the research line opened on the search for semantic primes’ exponents in Old English within the frame of the Natural Semantic Metalanguage theory (Goddard 1997, 2012; Goddard and Wierzbicka 2002). The aim of this study is to complete the line of research on prime identification opened on the category Actions, events, movement, contact by establishing the Old English exponent of the prime DO. With this purpose, this paper discusses the adequacy of different OE verbs as possible prime exponent on the basis of textual frequency, morphology, semantics and syntactic complementation. Relevant data of analysis have been retrieved mainly from the lexical database of Old English Nerthus, the Dictionary of Old English (Healey et al. 2018) and the Dictionary of Old English Corpus (Healey et al. 2009).
The aim of this paper is to retrieve the most relevant expansion words for expanding the initial query of the user in order to enhance the outcomes of web search results. Query expansion plays a major role in reformulating a user’s initial query to a one more pertinent to the user’s intended meaning. The reformulated query is then used to obtain more appropriate outcomes from a large amount of information on the web. The proposed semantic query expansion technique uses Wikipedia and WordNet as data sources. Wikipedia is taken as a base for all query expansions because it is one of the most diversified and relevant databases available on the web. To further improve the proposed query expansion technique,WordNet—a lexical database—is used as the as another data source because the synonyms (synsets) of the query term provided by it can be quite useful for query expansion. The proposed expansion technique successfully combines the two data sources to retrieve the most relevant expansion terms from the data sources in response to the user’s original query. The proposed work has been divided into four phases: (1) extraction of relevant words from Wikipedia (2) extraction of relevant words from WordNet (3) merging of the expansion terms obtained from Wikipedia and WordNet, and (4) query formulation by combining the expansion terms using Boolean operators. This reformulated query is then fired on the web to find the desired result. The Experimental result shows a significant improvement in information retrieval using query expansion.
Using incorrect worked examples during mathematics instruction can improve student learning. However, teachers worry that students may confuse correct and incorrect examples over time, and memory research supports this fear. To examine if this forgetting occurs, we had undergraduates rate the correctness of correct and incorrect worked examples immediately and one week later (Experiment 1). Previously studied incorrect examples were rated as slightly more correct after the delay, but this did not affect ratings of unstudied examples or problem-solving accuracy. In Experiment 2, we more closely mimicked how incorrect worked examples are used in classroom settings. Again, we found only small changes in students’ memory for studied worked examples after the delay, and no changes for unstudied examples or problem-solving accuracy. Our findings suggest the costs of teaching with incorrect worked examples are limited to the specific studied problems, and do not affect learning of the underlying mathematical rule.
BACKGROUND: Instrumental activities of daily living (IADL) impairment can begin in mild cognitive impairment (MCI), and is the core criteria for diagnosing dementia in both Alzheimer's (AD) and Parkinson's (PD) diseases. The Functional Activities Questionnaire (FAQ) has high discriminative power for dementia and MCI in older age populations, but is influenced by demographic factors. It is currently unclear whether the FAQ is suitable for assessing cognitive-associated IADL in non-demented PD patients, as motor disorders may affect ratings. OBJECTIVE: To compare IADL profiles in MCI patients with PD (PD-MCI) and AD (AD-MCI) and to verify the discriminative ability of the FAQ for MCI in patients with (PD-MCI) and without (AD-MCI) additional motor impairment. METHODS: Data of 42 patients each of PD-MCI, AD-MCI, PD cognitively normal (PD-CN), and healthy controls (HC), matched according to age, gender, education, and global cognitive impairment were analyzed. ANCOVA and binary regressions were used to examine the relationship between the FAQ scores and groups. FAQ cut-offs for PD-MCI (versus PD-NC) and AD-MCI (versus HC) were separately identified using receiver operating characteristic analyses. RESULTS: FAQ total score did not differentiate between MCI groups. PD-MCI subjects had greater difficulties with tax records and traveling while AD-MCI individuals were more impaired in managing finances and remembering appointments. Classification accuracy of the FAQ was good for diagnosing AD-MCI (69%, cut-off ≥1) compared to HC, and sufficient for differentiating PD-MCI (38.1%, cut-off ≥3) from PD-CN. CONCLUSION: The FAQ task profiles and classification accuracy differed between MCI related to PD and AD.
This paper analyses data to address a specific linguistic problem, i.e. the acquisition of the modification potential of the three more or less synonymous Dutch degree modifiers heel, erg and zeer, all meaning 'very', which show syntactic differences in modification potential. It continues the research reported on in The analysis makes crucial use of linguistic applications developed in the CLARIN infrastructure, in particular the treebank search applications PaQu (Parse and Query) and GrETEL Version 4.00. The analysis benefits from the use of parsed corpora (treebanks) in combination with the search and analysis options offered by PaQu and GrETEL. Earlier work showed that despite little data for zeer modifying adpositional phrases adult speakers end up with a generalised modification potential for this word. In this paper, I extend the dataset considered, and find more (but still little) data for this phenomenon. However, I also find a similar amount of data that form counterexamples to the non-generalisation of the modification potential of heel. I argue that the examples with heel concern constructions with idiosyncratic semantics and therefore are not counted as evidence for the general rule of modification. I suggest a simple statistical analysis to account for the fact that children 'learn' that heel cannot modify verbs or adpositions though there is no explicit evidence for this and they are not explicitly taught so.
Scene graph is a graph representation that explicitly represents high-level semantic knowledge of an image such as objects, attributes of objects and relationships between objects. Various tasks have been proposed for the scene graph, but the problem is that they have a limited vocabulary and biased information due to their own hypothesis. Therefore, results of each task are not generalizable and difficult to be applied to other down-stream tasks. In this paper, we propose Entity Synset Alignment(ESA), which is a method to create a general scene graph by aligning various semantic knowledge efficiently to solve this bias problem. The ESA uses a large-scale lexical database, WordNet and Intersection of Union (IoU) to align the object labels in multiple scene graphs/semantic knowledge. In experiment, the integrated scene graph is applied to the image-caption retrieval task as a downstream task. We confirm that integrating multiple scene graphs helps to get better representations of images.
This paper represents the development of the Myanmar Named Entity Recognition (NER) system using Conditional Random Fields (CRFs). In order to develop the system, a manually annotated Named Entities (NEs) corpus - collected from Myanmar news websites and Asia Language Treebank(ALT)-Parallel-Corpus has been used. We compare the performance of the system getting syllable-based input to the one getting character-based input. We observed that training data has more impact on the performance of the system. The experimental results show that the syllable-based system performs better than the character-based system. It achieves that Precision, Recall and F1-score values of 93.62%, 91.64% and 92.62% respectively.
Abstract Syntactic parsing is an important topic in the field of Mongolian language information processing. Compared with English and Chinese dependency parsing, Mongolian dependency parsing is still at the beginning stage. Mongolian syntactic parsing lack of Treebank resources seriously, under such conditions, a high quality syntactic parser cannot be developed by statistical methods simply. Aiming at the characteristics that Mongolian language has rich morphological features, this paper presented a rule and statistics-based dependency parsing model using Mongolian Dependency Treebank as training and evaluation data. The morphological and syntactic rules are represented using complex features and unification operations. The statistical model is represented using lexical dependent probability. This model has now achieved accuracies of 77.18%, 69.42% and 95.44% for the unlabelled annotation score, the labeled annotation score and the head word annotation score respectively.
OBJECTIVES: To investigate whether tethered swimming (TS) performed 8 minutes before a 50-m freestyle swimming sprint could be an effective postactivation potentiation method to improve performance in young swimmers. METHODS: Fourteen regional-level male adolescent swimmers (age 13.0 [2.0] y; height 161.1 [12.4] cm; body mass 52.5 [9.5] kg) underwent 2 trial conditions in a randomized and counterbalanced order (1 experimental [TS], 1 control) on different days. During the experimental session, the participants performed a standard warm-up of 1200 m followed by a TS exercise, which consisted of 3 × 10-second maximal efforts of TS with 1-minute rests between bouts. In the control condition, the warm-up phase was immediately followed by 200 m at a moderate pace (same duration as the TS in the experimental session). Performance (time trial); biomechanical (stroke length), physiological (blood lactate concentrations), and psychophysiological (ratings of perceived exertion) variables; and countermovement-jump (CMJ) flight time were collected. RESULTS: TS warm-up had no significant effect on 50-m swimming performance (P =.27), postexercise ratings of perceived exertion, stroke length, or CMJ flight time (P ≥.05). Blood lactate concentrations significantly increased at the end of the warm-up in the TS condition only (interaction effect: F1.91,29.91 = 4.91, P =.01, η2 =.27) and after the 50-m trial in both conditions (F1.57,20.41 = 62.39, P =.001, η2 =.82). CONCLUSIONS: The present study demonstrated that 3 × 10-second TS exercises performed 8 minutes prior to the event did not affect ratings of perceived exertion, stroke length, or CMJ flight time. In addition, tethered swimming did not affect 50-m freestyle sprint performance in young swimmers.
Pointing devices are the primary media of interac-tion between humans and computers. The three most popular pointing devices used in computers (both portable and non-portable) are mouse, touchpad and nubs (joystick). They have their different advantages and use cases while being targeted to different user groups. The aim of this study was to investigate whether the aforementioned pointing devices have different effects on human valence. A total of 12 participants were recruited for the experiment. Each participant completed a pointing reaction test with every pointing device aforementioned, where they selected as many randomly appearing circles as possible in a given amount of time. Then, subjective ratings of emotional valence and arousal were collected, and the effects of the pointing device used on these ratings were investigated. Our study shows that the valence rating of using the mouse were significantly higher in challenging scenarios, compared to the likes of touchpad and nub.
A cache-inspired approach is proposed for neural language models (LMs) to improve long-range dependency and better predict rare words from long contexts.This approach is a simpler alternative to attention-based pointer mechanism that enables neural LMs to reproduce words from recent history.Without using attention and mixture structure, the method only involves appending extra tokens that represent words in history to the output layer of a neural LM and modifying training supervisions accordingly.A memory-augmentation unit is introduced to learn words that are particularly likely to repeat.We experiment with both recurrent neural network-and Transformer-based LMs.Perplexity evaluation on Penn Treebank and WikiText-2 shows the proposed model outperforms both LSTM and LSTM with attention-based pointer mechanism and is more effective on rare words.N -best rescoring experiments on Switchboard indicate that it benefits both very rare and frequent words.However, it is challenging for the proposed model as well as two other models with attention-based pointer mechanism to obtain good overall WER reductions.
Aims: This study provides new insights into Arabic-English code-switching with particular reference to verb insertion. It aims to identify (1) patterns of English verb insertion into Arabic; (2) factors affecting them. We offer an alternative to previous studies’ conclusions regarding a supposed lack of English verbs integrated morphologically into Arabic, which is claimed to result from incongruence between Arabic and English verb systems. Methodology: We employ the Matrix Language Frame (MLF) model and the 4-M model. Data and analysis: The data comprise 14,414 clauses obtained from interviews with students at the American University in Cairo. Data were analyzed quantitatively. Findings: Most (80.17%) of inserted verbs were inflected with Arabic tense, gender, and number prefixes showing morphological integration into Arabic. We distinguished four recurrent patterns in verb insertion: (1) complete morphological integration in the present tense; (2) incomplete assimilation of forms requiring the use of the plural suffix -u; (3) lack of morphological integration in the past tense; and (4) lack of suffixation of Arabic clitics to English verbs. Originality: This is the first study focusing on verb insertion in Arabic-English code-switching based on empirical data collected in Egypt. It offers different findings on verb patterns and their explanation compared with other quantitative studies based on the MLF model. We propose to look beyond incongruence between Arabic and English as a factor determining verb patterns to include linguistic convention. Thus, we hypothesize that verb insertion might be controlled by linguistic norms accepted and perpetuated in a given speech community. Significance: Contrary to previous claims, our results show that patterns of verb insertion in Arabic-English code-switching are consistent with the MLF model. Hence, the study contributes evidence for the MLF model and its explanatory value.
Generating novel design concepts is a cornerstone for producing innovative products. Although many methods have been proposed for supporting the task, their performance depends on human ability. The ultimate goal of this research is to build a method supporting designers to generate novel design concepts with the knowledge of human creativity. Toward the goal, this research assumes that the more distant two function concepts chosen, the more novel idea would be come up with by the combination of the two concepts. Based on the assumption, this paper introduces a notion of novelty potential of the combination of two function concepts, and builds a method to assess it by the function similarity. Some alternative methods are proposed to calculate it with the integration of a lexical database for natural language called WordNet and a distributional semantics method called word2vec. They are verified with an evaluation experiment which performs correlation analysis between the human’s evaluation of the novelty potential and the proposed method’s assessment of function similarity. This paper discusses which method matches a human sense most, and its possibility for design concept generation based on the results of the experiment.