Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
16504 papers
The application value of the convolutional neural network (CNN) algorithm in the diagnosis of sports knee osteoarthropathy was investigated in this study. A network model was constructed in this experiment for image analysis of magnetic resonance imaging (MRI) technology. Then, 100 cases of sports knee osteoarthropathy patients and 50 healthy volunteers were selected. Digital radiography (DR) images and MRI images of all the research objects were collected after the inclusion of the two groups. Besides, the important physiological representations were extracted from their image data graphs, and the hidden complex relationships were learned. The state without input results was judged through convolutional network calculation, and the result prediction was given. On this basis, there was an analysis of the diagnostic efficiency of traditional DR images and MRI images based on CNN for patients with sports knee osteoarthropathy. The results showed that the MRI images analyzed by the CNN model showed a more obvious display rate than DR images for some nonbone changes of osteoarthritis. The correlation coefficient between MRI image rating and visual analog scale (VAS) was 0.865, which was higher than 0.713 of DR image rating, with a statistical meaning ( <math xmlns="http://www.w3.org/1998/Math/MathML" id="M1"> <mi>P</mi> <mo><</mo> <mn>0.01</mn> </math> ). For cases with mild lesions, the number of cases detected by MRI based on CNN algorithm in 0–4 image rating was 15, 18, 10, 6, and 7, respectively, which was markedly better than that of DR images. In short, the MRI examination based on the CNN image analysis model could extract important physiological representations from the image data and learn the hidden complex relationships. The convolutional network was calculated to determine the state of the uninput results and give the result predictions. Moreover, MRI examination based on the CNN image analysis model had high overall diagnostic efficiency and grading diagnostic efficiency for patients with motor knee osteoarthropathy, which was of great significance in clinical practice.
Current accounts of neural plasticity emphasize the role of connectivity and conserved function in determining a neural tissue’s functional role even after atypical early experiences. However, in apparent conflict with this view, studies have suggested that in congenitally blind individuals, language activates primary visual cortex, with no evidence of major changes in anatomical connectivity that could explain this apparent drastic functional change in what is typically a low-level visual area. To reconcile what appears to be unprecedented functional reorganization in V1 with known accounts of plasticity limitations, we used functional magnetic resonance imaging (fMRI) to test whether primary visual cortex also responds to spoken language in sighted individuals. We found that primary visual cortex was activated by comprehensible speech as compared to a reversed speech control task, in a left-lateralized and focal manner, in sighted individuals. Importantly, left V1 activation was also significant and comparable for abstract and concrete words, precluding a visual imagery account of such activation, and activation was also not correlated with attentional arousal ratings. Together these findings suggest that primary visual cortex responds to verbal information even in the typically developed brain, potentially to predict visual input. This capability might be the basis for the strong V1 language activation observed in people born blind, re-affirming the notion that plasticity is guided by pre-existing connectivity and abilities in the typically developed brain.
Objective: Somatization symptoms are commonly comorbid with depression. Furthermore, people with depression and somatization have a negative memory bias. We investigated the differences in emotional memory among adolescent patients with depressive disorders, with and without functional somatization symptoms (FSS). Methods: We recruited 30 adolescents with depression and FSS, 38 adolescents with depression but without FSS, and 38 healthy participants. Emotional memory tasks were conducted to evaluate the emotional memory of the participants in the three groups. The clinical symptoms were evaluated using the Hamilton Depression Rating Scale (HDRS) and the Children's Somatization Inventory (CSI). Results: The valence ratings and recognition accuracy rates for positive and neutral images of adolescent patients were significantly lower than those of the control group ( F = 12.208, P &lt; 0.001; F = 6.801, P &lt; 0.05; F = 14.536, P &lt; 0.001; F = 6.306, P &lt; 0.05, respectively); however, the recognition accuracy rate for negative images of adolescent patients of depression without FSS was significantly lower than that of patients with FSS and control group participants ( F = 10.316, P &lt; 0.001). These differences persisted after controlling for HDRS scores. The within-group analysis revealed that patients of depression with FSS showed significantly higher recognition accuracy rates for negative images than the other types ( F = 5.446, P &lt; 0.05). The recognition accuracy rate for negative images was positively correlated with CSI scores ( r = 0.352, P &lt; 0.05). Conclusion: Therefore, emotional memory impairment exists in adolescent patients of depression and FSS are associated with negative emotional memory retention.
Automated machine learning (AutoML) is a technique which helps to determine the optimal or near-optimal model for a specific dataset and has been a focused research area during the last years. The automation of model design opens doors for non-machine learning experts to utilize machine learning models in several scenarios, which is both appealing for a wide range of researchers and for cloud services as well. Neural Architecture Search is a subfield of AutoML where the optimal artificial neural network model's architecture is generally searched with adaptive algorithms. This paper proposes a method to apply Efficient Neural Architecture Search (ENAS) to LSTM-like recurrent architecture, which uses a gating mechanism an inner memory. Using this method, the paper investigates if the handcrafted Long Short-Term Memory (LSTM) cell is an optimal or near-optimal solution of sequence modelling for a given dataset, or other, automatically defined recurrent structures outperform. The performance of vanilla LSTM, and advanced recurrent architectures designed by random search, and reinforcement learning-based ENAS are examined and compared. The proposed methods are evaluated in a text generation task on the Penn TreeBank dataset.
Abstract In this chapter we describe a multilingual extension of Swedish FrameNet++, intended to address research questions of a broad comparative nature, in genealogical, areal and typological linguistics, focusing on the integration into Swedish FrameNet++ of so-called core vocabularies, used in several linguistic subfields in order to conduct massive comparative studies involving large numbers of languages. Specifically, we describe the inclusion of two such lexical databases covering several hundred South Asian languages, with the aim of investigating areal and genealogical connections among these languages.
OBJECTIVE: Research into echocardiography (echo) during cardiac arrest has suffered from methodological flaws that limit aggregation of findings. We developed and validated a novel image rating scale for qualitative analysis of echo images obtained during resuscitation. METHODS: A novel 5-point ordinal rating scale was developed and validated using recorded echo images from 145 consecutive cardiac arrest patients. Recorded echo images were reviewed in a blinded fashion by investigators experienced in cardiac arrest echo, and image quality was rated using this scale. Cardiac activity was subsequently classified as no activity, disorganized activity and organized activity. The primary outcome was inter-rater agreement using the image quality rating scale. Secondary outcome was the qualitative evaluation of the type of cardiac activity. RESULTS: A total of 235 ultrasounds were analyzed by study investigators using the image quality rating scale. The overall image quality agreement between reviewers using the scale was good with a weighted kappa of 0.65. Agreement for image quality in subxyphoid images was greater than in parasternal images (0.65-0.52). Echo analysis of cardiac activity showed no activity (33%), disorganized activity (18%), and organized activity (49%). Agreement was great for presence or absence of "cardiac activity" and "organized cardiac activity" with a kappa of 0.84 and 0.78. CONCLUSIONS: A novel image quality rating scale for echo during cardiac arrest demonstrates substantial agreement between reviewers. Agreement regarding the presence or absence, as well as the organization of cardiac activity was substantial.
Different linearizations have been proposed to cast dependency parsing as sequence labeling and solve the task as: (i) a head selection problem, (ii) finding a representation of the token arcs as bracket strings, or (iii) associating partial transition sequences of a transition-based parser to words. Yet, there is little understanding about how these linearizations behave in low-resource setups. Here, we first study their data efficiency, simulating data-restricted setups from a diverse set of rich-resource treebanks. Second, we test whether such differences manifest in truly low-resource setups. The results show that head selection encodings are more data-efficient and perform better in an ideal (gold) framework, but that such advantage greatly vanishes in favour of bracketing formats when the running setup resembles a real-world low-resource configuration.
Abstract This paper introduces, a new semantic role labeling method that transforms a text into a frame-oriented knowledge graph. It performs dependency parsing, identifies the words that evoke lexical frames, locates the roles and fillers for each frame, runs coercion techniques, and formalizes the results as a knowledge graph. This formal representation complies with the frame semantics used in Framester, a factual-linguistic linked data resource. We tested our method on the WSJ section of the Peen Treebank annotated with VerbNet and PropBank labels and on the Brown corpus. The evaluation has been performed according to the CoNLL Shared Task on Joint Parsing of Syntactic and Semantic Dependencies. The obtained precision, recall, and F1 values indicate that TakeFive is competitive with other existing methods such as SEMAFOR, Pikes, PathLSTM, and FRED. We finally discuss how to combine TakeFive and FRED, obtaining higher values of precision, recall, and F1 measure.
There is empirical evidence in different languages on how the computation of gender morphology during psycholinguistic processing affects the conformation of sex-generic representations. However, there is no empirical evidence on the processing of non-binary morphological variants in Spanish (-x or -e) in contrast to the generic masculine variant (-o). To analyze this phenomenon, we conducted two experiments: an acceptability judgment task and a sentence comprehension task. The results show differences depending on the task. So, the underlying processes that are put into play in each one generate different effects. In acceptability judgments, which involve strategic processes mediated by beliefs and the linguistic norm, the generic masculine is more acceptable to refer to mixed groups. In the sentence comprehension task, which inquires about automatic processes and implicit representations, the non-binary forms consistently elicited a reference to mixed groups. Furthermore, the response times indicated that these morphological variants do not entail a higher processing cost than the generic masculine.
OBJECTIVES: To determine the diagnostic accuracy of dual-energy CT (DECT) virtual noncalcium (VNCa) reconstructions for assessing thoracic disk herniation compared to standard grayscale CT. METHODS: In this retrospective study, 87 patients (1131 intervertebral disks; mean age, 66 years; 47 women) who underwent third-generation dual-source DECT and 3.0-T MRI within 3 weeks between November 2016 and April 2020 were included. Five blinded radiologists analyzed standard DECT and color-coded VNCa images after a time interval of 8 weeks for the presence and degree of thoracic disk herniation and spinal nerve root impingement. Consensus reading of independently evaluated MRI series served as the reference standard, assessed by two separate experienced readers. Additionally, image ratings were carried out by using 5-point Likert scales. RESULTS: MRI revealed a total of 133 herniated thoracic disks. Color-coded VNCa images yielded higher overall sensitivity (624/665 [94%; 95% CI, 0.89-0.96] vs 485/665 [73%; 95% CI, 0.67-0.80]), specificity (4775/4990 [96%; 95% CI, 0.90-0.98] vs 4066/4990 [82%; 95% CI, 0.79-0.84]), and accuracy (5399/5655 [96%; 95% CI, 0.93-0.98] vs 4551/5655 [81%; 95% CI, 0.74-0.86]) for the assessment of thoracic disk herniation compared to standard CT (all p <.001). Interrater agreement was excellent for VNCa and fair for standard CT (ϰ = 0.82 vs 0.37; p <.001). In addition, VNCa imaging achieved higher scores regarding diagnostic confidence, image quality, and noise compared to standard CT (all p <.001). CONCLUSIONS: Color-coded VNCa imaging yielded substantially higher diagnostic accuracy and confidence for assessing thoracic disk herniation compared to standard CT. KEY POINTS: • Color-coded VNCa reconstructions derived from third-generation dual-source dual-energy CT yielded significantly higher diagnostic accuracy for the assessment of thoracic disk herniation and spinal nerve root impingement compared to standard grayscale CT. • VNCa imaging provided higher diagnostic confidence and image quality at lower noise levels compared to standard grayscale CT. • Color-coded VNCa images may potentially serve as a viable imaging alternative to MRI under circumstances where MRI is unavailable or contraindicated.
Theimplicit discourse relation classification is of great importance to discourse analysis. It aims to identify the logical relation between sentence pair. Compared with the linear network model, the graph neural network has a more complex structure to capture cross-sentence interactions. Therefore, this article proposes a semantic graph neural network for implicit discourse relation classification. Specifically, we design a semantic graph to describe the syntactic structure of sentences and semantic interactions between sentence pair. Then, convolutional neural network (CNN) with different convolutional kernels to extract the multi-granularity semantic features. The experimental results on Penn Discourse TreeBank 2.0 (PDTB 2.0) prove that our work performed well.
Although there are numerous studies on collocation in English writing by L2 university students, little is known about the problems encountered by mature researchers writing authentic L2 English texts in their fields. This study investigates collocation issues in L2 English research papers in Brazil. Its starting point was the compilation of the Brazilian Academic Corpus of English (BrACE), a 906,035-word multidisciplinary corpus of journal articles written in English that have been published in Brazilian journals. The most frequent noun collocations in this corpus were contrasted with the expert writing lexical database underlying the ColloCaid academic writing assistant. No evidence of systematic miscollocation was found in the published papers represented in BrACE. However, many general academic English collocations were conspicuous by their absence from BrACE, including collocations with L1 Portuguese cognates. We also observed that the collocations in BrACE were less diverse and tended to score higher in terms of strength of association than their equivalents in the reference data. In addition to feedback on miscollocations which might arise in unedited manuscripts, our findings to conclude that Brazilian (and other English L2) research writers can benefit from suggestions to expand their collocation repertoire, enhance their perceptions of collocation strength, and offset collocation avoidance.
Abstract Graphs have become an increasingly important means of representing data, for instance, when communicating data on climate change. However, graph characteristics might significantly affect graph comprehension. The goal of the present work was to test whether the marking forms usually depicted on line-graphs, can have an impact on graph evaluation. As past work suggests that triangular forms might be related to threat, we compared the effect of triangular marking forms with other symbols (triangles, circles, squares, rhombi, and asterisks) on subjective assessments. Participants in Study 1 ( N = 314) received 5 different line-graphs about climate change, each of them using one out of 5 marking forms. In Study 1, the threat and arousal ratings of the graphs with triangular marking shapes were not higher than those with the other marking symbols. Participants in Study 2 ( N = 279) received the same graphs, yet without labels and indeed rated the graphs with triangle point markers as more threatening. Testing whether local rather than global spatial attention would lead to an impact of marker shape in climate graphs, Study 3 ( N = 307) documented that a task demanding to process a specific data-point on the graph (rather than just the line graph as a whole) did not lead to an effect either. These results suggest that marking symbols can principally affect threat and arousal ratings but not in the context of climate change. Hence, in graphs on climate change, choice of point markers does not have to take potential side-effects on threat and arousal into account. These seem to be restricted to the processing of graphs where form aspects face less competition from the content domain on judgments.
Research on and the development of technology that promotes human subjective well-being is crucial for the current global information society. Focusing on positive psychological intervention in social communication through mobile devices, this study proposes an input method to promote subjective well-being by recommending positive words and phrases for their negative counterparts. Accordingly, a design workshop was conducted to develop a reframing dictionary in Japanese. This dictionary constitutes a collection of negative words and their corresponding positive words and phrases that convey the same meaning. We also developed an input method that encourages users to select positive words and phrases when they enter a negative word. Preliminary evaluation results indicate a significant difference in positive affect ratings before and after communicating on social networking sites using the proposed input method. Thus, this input method contributes to promoting psychological well-being during daily information activities.
Introduction: The addition of graphic health warnings on cigarette packets can facilitate smoking cessation, primarily through their ability to elicit a negative affective response. Smoking is linked to COVID-19 mortality, thus making it likely to elicit a strong affective response in smokers. COVID-19-related health warnings (C19HW) may therefore enhance graphic health warnings, when compared to traditional health warnings (THW). Further, because impulsivity influences smoking behaviours, we also examined whether these affective responses were associated with delay discounting.Methods: In a between-subjects design, 240 smokers rated the valence and arousal elicited by tobacco packaging that contained either a C19HW or THW (both referring to death). Participants also completed questionnaires to quantify delay discounting and attitudes towards COVID-19 and smoking (eg, health risks, motivation to quit).Results: There were no differences between the two health warning types on either valence or arousal, nor any secondary outcome variables. There was, however, a significant interaction between health warning type and delay discounting on arousal ratings. Specifically, in smokers who exhibit low delay discounting, C19HWs elicited significantly greater subjective arousal rating than did THWs, whereas there was no significant effect of health warning type on arousal in smokers who exhibited high delay discounting.Conclusion: The results suggest that in smokers who exhibit low impulsivity (but not high impulsivity), C19HWs may be more arousing than THWs. Future work is required to explore the long-term utility of C19HWs, and to identify the specific mechanism by which delay discounting moderates the impact of tobacco health warnings.
Manually annotating a treebank is time consuming and laborintensive. We conduct delexicalized crosslingual dependency pars ing experiments, where we train the parser on one language and test on our target language. As our test case, we use Xibe, a severely underresourced Tungusic language. We as sume that choosing a closely related language as the source language will provide better re sults than more distant relatives. However, it is not clear how to determine those closely re lated languages. We investigate three differ ent methods: choosing the typologically clos est language, using LangRank, and choosing the most similar language based on perplexity.
Civil identity is one of the most significant factors in modern political practice. Today’s identity formation and development of large national groups is less based on a cultural and historical foundation and increasingly depends on political technologies. Among them, the construction of new languages plays an important role. The article studies the Bosnian language policy, which, contrary to forming a common civil identity, as a result of the politicization of linguistic norms becomes a factor in creating a “forge of hatred”. Drawing on constructivist social theories, the author summarizes Bosnian linguistic practices and examines them through the prism of symbolic interactionism and negative feedback systems. Particular attention is paid to situations when the desire for effective communication motivates speakers to abandon ethnically colored linguistic markers and situations in which the language acts as a defense against the internal “other.” Applying the criteria for distinguishing between language and dialects, the author concludes that the phonetic principle of the Serbo-Croatian language formation made it possible, after the destruction of Yugoslavia, to turn this linguistic continuum into an identification weapon to delimit the citizens of one country. This experience helps analyze the politicization of literary interpretations and linguistic norms in other regions of the world, where there are also examples of the growth of xenophobia, nationalism, and intolerance resulting from a differentiating language policy.
Compound probabilistic context-free grammars (C-PCFGs) have recently established a new state of the art for unsupervised phrase-structure grammar induction. However, due to the high space and time complexities of chart-based representation and inference, it is difficult to investigate C-PCFGs comprehensively. In this work, we rely on a fast implementation of C-PCFGs to conduct an evaluation complementary to that of~\citet{kim-etal-2019-compound}. We start by analyzing and ablating C-PCFGs on English treebanks. Our findings suggest that (1) C-PCFGs are data-efficient and can generalize to unseen sentence/constituent lengths; and (2) C-PCFGs make the best use of sentence-level information in generating preterminal rule probabilities. We further conduct a multilingual evaluation of C-PCFGs. The experimental results show that the best configurations of C-PCFGs, which are tuned on English, do not always generalize to morphology-rich languages.
Abstract Background The Montessori Method underpinned by the principle of person-centered care has been widely adopted to design activities for people with dementia. However, the methodological quality of the existing evidence is fair. The objectives of this study are to examine the feasibility and effects of a culturally adapted group-based Montessori Method for Dementia program in Chinese community on engagement and affect in community-dwelling people with dementia. Methods This was a two-arm randomized controlled trial. People who were aged 60 years or over and with mild to moderate dementia were recruited and randomly assigned to the intervention group to receive Montessori-based activities or the comparison group to receive conventional group activities over eight weeks. The attendance rates were recorded for evaluating the feasibility. The Menorah Park Engagement Scale and the Apparent Affect Rating Scale were used to assess the engagement and affect during the activities based on observations. Generalized Estimating Equation model was used to examine the intervention effect on the outcomes across the sessions. Results A total of 108 people with dementia were recruited. The average attendance rate of the intervention group (81.5%) was higher than that of the comparison group (76.3%). There was a significant time-by-group intervention effect on constructive engagement in the first 10 minutes of the sessions (Wald χ 2 = 15.21–19.93, ps = 0.006–0.033), as well as on pleasure (Wald χ 2 = 25.37–25.73, ps ≤ 0.001) and interest (Wald χ2 = 19.14–21.11, p s = 0.004–0.008) in the first and the middle 10 minutes of the sessions, adjusted for cognitive functioning. Conclusions This study provide evidence that Montessori-based group activities adapted to the local cultural context could effectively engage community-dwelling Chinese older people with mild to moderate dementia in social interactions and meaningful activities and significantly increase their positive affect. Trial registration ClinicalTrials.gov, NCT04352387. Registered 20 April 2020. Retrospectively registered.
There are only a few previous EEG studies that were conducted while the audience is listening to live music. However, in laboratory settings using music recordings, EEG frequency bands theta and alpha are connected to music improvisation and creativity. Here, we measured EEG of the audience in a concert-like setting outside the laboratory and compared the theta and alpha power evoked by partly improvised versus regularly performed familiar versus unfamiliar live classical music. To this end, partly improvised and regular versions of pieces by Bach (familiar) and Melartin (unfamiliar) were performed live by a chamber trio. EEG data from left and right frontal and central regions of interest were analysed to define theta and alpha power during each performance. After the performances, the participants rated how improvised and attractive each of the performances were. They also gave their affective ratings before and after each performance. We found that theta power was enhanced during the familiar improvised Bach piece and the unfamiliar improvised Melartin piece when compared with the performance of the same piece performed in a regular manner. Alpha power was not modulated by manner of performance or by familiarity of the piece. Listeners rated partly improvised performances of a familiar Bach and unfamiliar Melartin piece as more improvisatory and innovative than the regular performances. They also indicated more joy and less sadness after listening to the unfamiliar improvised piece of Melartin and less fearful and more enthusiastic after listening to the regular version of Melartin than before listening. Thus, according to our results, it is possible to study listeners' brain functions with EEG during live music performances outside the laboratory, with theta activity reflecting the presence of improvisation in the performances.
Though machine learning algorithms are able to achieve pattern recognition from the correlation between data and labels, the presence of spurious features in the data decreases the robustness of these learned relationships with respect to varied testing environments. This is known as out-of-distribution (OoD) generalization problem. Recently, invariant risk minimization (IRM) attempts to tackle this issue by penalizing predictions based on the unstable spurious features in the data collected from different environments. However, similar to domain adaptation or domain generalization, a prevalent non-trivial limitation in these works is that the environment information is assigned by human specialists, i.e. a priori, or determined heuristically. However, an inappropriate group partitioning can dramatically deteriorate the OoD generalization and this process is expensive and time-consuming. To deal with this issue, we propose a novel theoretically principled min-max framework to iteratively construct a worst-case splitting, i.e. creating the most challenging environment splittings for the backbone learning paradigm (e.g. IRM) to learn the robust feature representation. We also design a differentiable training strategy to facilitate the feasible gradient- based computation. Numerical experiments show that our algorithmic framework has achieved superior and stable performance in various datasets, such as Colored MNIST and Punctuated Stanford sentiment treebank (SST). Furthermore, we also find our algorithm to be robust even to a strong data poisoning attack. To the best of our knowledge, this is one of the first to adopt differentiable environment splitting method to enable stable predictions across environments without environment index information, which achieves the state-of-the-art performance on datasets with strong spurious correlation, such as Colored MNIST.
Multisensory integration influences emotional perception, as the McGurk effect demonstrates for the communication between humans. Human physiology implicitly links the production of visual features with other modes like the audio channel: Face muscles responsible for a smiling face also stretch the vocal cords that results in a characteristic smiling voice. For artificial agents capable of multimodal expression, this linkage is modeled explicitly. In our study, we observe the influence of visual and audio channel on the perception of the agent’s emotional state. We created two virtual characters to control for anthropomorphic appearance. We record videos of these agents either with matching or mismatching emotional expression in the audio and visual channel. In an online study we measured the agent’s perceived valence and arousal. Our results show that a matched smiling voice and smiling face increase both dimensions of the Circumplex model of emotions: ratings of valence and arousal grow. When the channels present conflicting information, any type of smiling results in higher arousal rating, but only the visual channel increases the perceived valence. When engineers are constrained in their design choices, we suggest they should give precedence to convey the artificial agent’s emotional state through the visual channel.
We present a recurrent neural network memory that uses sparse coding to create a combinatoric encoding of sequential inputs. The network is trained using only local and immediate credit assignment. Despite this constraint, results are comparable to networks trained using deep backpropagation or BackProp Through Time (BPTT). With several examples, we show that the network can associate distant cause and effect in a discrete stochastic process, predict partially-observable higherorder sequences, and learn to generate many time-steps of video simulations. Typical memory consumption is 10-30x less than conventional RNNs, such as LSTM, trained by BPTT. One limitation of the memory is generalization to unseen input sequences. We additionally explore this limitation by measuring next-word prediction perplexity on the Penn Treebank dataset.
Abstract Quantitative 23 Na magnetic resonance imaging (MRI) provides tissue sodium concentration (TSC), which is connected to cell viability and vitality. Long acquisition times are one of the most challenging aspects for its clinical establishment. K‐space undersampling is an approach for acquisition time reduction, but generates noise and artifacts. The use of convolutional neural networks (CNNs) is increasing in medical imaging and they are a useful tool for MRI postprocessing. The aim of this study is 23 Na MRI acquisition time reduction by k‐space undersampling. CNNs were applied to reduce the resulting noise and artifacts. A retrospective analysis from a prospective study was conducted including image datasets from 46 patients (aged 72 ± 13 years; 25 women, 21 men) with ischemic stroke; the 23 Na MRI acquisition time was 10 min. The reconstructions were performed with full dataset (FI) and with a simulated dataset an image that was acquired in 2.5 min (RI). Eight different CNNs with either U‐Net–based or ResNet‐based architectures were implemented with RI as input and FI as label, using batch normalization and the number of filters as varying parameters. Training was performed with 9500 samples and testing included 400 samples. CNN outputs were evaluated based on signal‐to‐noise ratio (SNR) and structural similarity (SSIM). After quantification, TSC error was calculated. The image quality was subjectively rated by three neuroradiologists. Statistical significance was evaluated by Student’s t‐test. The average SNR was 21.72 ± 2.75 (FI) and 10.16 ± 0.96 (RI). U‐Nets increased the SNR of RI to 43.99 and therefore performed better than ResNet. SSIM of RI to FI was improved by three CNNs to 0.91 ± 0.03. CNNs reduced TSC error by up to 15%. The subjective rating of CNN‐generated images showed significantly better results than the subjective image rating of RI. The acquisition time of 23 Na MRI can be reduced by 75% due to postprocessing with a CNN on highly undersampled data.
Techniques that detect sentence similarity have been a very important domain of research and lately many such techniques have been successfully implemented. With the use of Natural Language Processing (NLP) these techniques have been implemented more efficiently. The concept of semantic analysis is very significant in determining sentence similarity. The model proposed in this paper, deploys a NLP based methodology that works on the Sentence Involving Compositional Knowledge (SICK) dataset. The proposed methodology considers the set of sentencesto be a subset of words and it is split based on the semantic and syntactic structure. A lexical database is used by this model, unlike methods deployed by other models. This is followed by the computation of the word order vector. When this NLP based method is tested on the dataset, the accuracy obtained is 82.7% on the basis of mean absolute error. The obtained results are better than the previously used methods. Also, the proposed method is computationally faster than the existing methods.
Disyllabic verb-noun (V-N) items in Shanghai Wu have variable surface tone patterns: They can undergo either a rightward extension tone sandhi, which extends the lexical tone of the first syllable over the entire word, or tonal reduction on the first syllable. The current study investigates how the phonological properties of these alternation processes as well as variation influence how Shanghai speakers represent and access such words. We conducted an auditory-auditory priming lexical decision experiment on Shanghai V-N items that can undergo either tonal extension or tonal reduction with native Shanghai speakers. Each disyllabic target was preceded by monosyllabic primes with the canonical tone, the tonal-extension tone, the surface tone, or a tone unrelated to the tone of the first syllable of the targets. Results showed both canonical and tonal-extension priming effects, but no surface priming effect. Moreover, although more familiar V-Ns were recognized with shorter reaction time, the priming effect did not interact with speakers’ familiarity ratings or sandhi preference ratings of the targets. These data are consistent with the interpretation that both the canonical and tonal-extension forms are represented in Shanghai speakers’ mental lexicon due to tone sandhi variation, but the representation does not seem to be modulated by the frequencies of the variants. Also, together with findings from auditory priming studies of other tone sandhi patterns, the current study suggests that certain phonological properties of an alternation, such as its locality and transparency, influence the representation of words undergoing the alternation; but whether the alternation is structure-preserving does not seem to impact the representation.
An extensive epidemiological literature indicates that increased exposure to tobacco retail outlets (TROs) places never smokers at greater risk for smoking uptake and current smokers at greater risk for increased consumption and smoking relapse. Yet research into the mechanisms underlying this effect has been limited. This preliminary study represents the first effort to examine the neurobiological consequences of exposure to personally relevant TROs among both smokers (n = 17) and nonsmokers (n = 17). Individuals carried a global positioning system (GPS) tracker for 2 weeks. Traces were used to identify TROs and control outlets that fell inside and outside their ideographically defined activity space. Participants underwent functional MRI (fMRI) scanning during which they were presented with images of these storefronts, along with similar store images from a different county and rated their familiarity with these stores. The main effect of activity space was additive with a Smoking status × Store type interaction, resulting in smokers exhibiting greater neural activation to TROs falling inside activity space within the parahippocampus, precuneus, medial prefrontal cortex, and dorsal anterior insula. A similar pattern was observed for familiarity ratings. Together, these preliminary findings suggest that the otherwise distinct neural systems involved in self-orientation/self-relevance and smoking motivation may act in concert and underlie TRO influence on smoking behavior. This study also offers a novel methodological framework for evaluating the influence of community features on neural activity that can be readily adapted to study other health behaviors.
Child-directed speech, as a specialized form of speech directed toward young children, has been found across numerous languages around the world and has been suggested as a universal feature of human experience. However, variation in its implementation and the extent to which it is culturally supported has called its universality into question. Child-directed speech has also been posited to be associated with expression of positive affect or "happy talk." Here, we examined Canadian English-speaking adults' ability to discriminate child-directed from adult-directed speech samples from two dissimilar language/cultural communities; an urban Farsi-speaking population, and a rural, horticulturalist Tseltal Mayan speaking community. We also examined the relationship between participants' addressee classification and ratings of positive affect. Naive raters could successfully classify CDS in Farsi, but only trained raters were successful with the Tseltal Mayan sample. Associations with some affective ratings were found for the Farsi samples, but not reliably for happy speech. These findings point to a complex relationship between perception of affect and CDS, and context-specific effects on the ability to classify CDS across languages.
Coordination is a phenomenon of language that conjoins two or more terms or phrases using a coordinating conjunction. Although coordination has been explored extensively in the linguistics literature, the rules and constraints that govern its structure are still largely elusive and widely debated amongst linguists. This paper presents a study of two-termed unlike coordinations in particular, where the two conjuncts of the coordination phrase form valid constituents but have distinct categories. We conducted a syntactic analysis of the phrasal categories that can be conjoined in such unlike coordinations through a computational corpusbased approach, utilizing the Corpus of Contemporary American English (COCA) as the main data source, as well as the Penn Treebank (PTB). The results show that the two conjuncts within unlike coordinations display different properties based on their position, supporting an antisymmetric view of the structure of coordination. This research provides new data and perspectives through the use of statistical techniques that can help shape future theories and models of coordination.
Online reviews are the newest method for patients to evaluate their providers. However, insufficient studies focus on the role of inherent physician characteristics, such as gender and years of experience, on patient satisfaction. We analyzed both quantitative and qualitative online reviews of 350 general dermatology providers at 121 Accreditation Council for Graduate Medical Education–accredited dermatology programs across the country to determine the effect of gender and years of experience. There were 38,008 online reviews of general dermatology providers. There was no significant difference in male and female overall ratings. Ratings were overall equally positive for both genders. Female providers were more likely to have positive written comments regarding time spent with patients (P = 0.027). New providers received highest overall, promptness, and time spent with patient ratings (P < 0.001). Medium experience providers received highest scores in bedside manner (P < 0.001), accurate diagnosis (P = 0.018), and ability to answer questions (P = 0.005). Advanced providers scored the lowest across all categories. In conclusion, gender did not significantly affect ratings, although females received more positive written comments on time spent with patients. Years of experience, however, is a significant factor in patient ratings, with new or medium experience providers scoring higher than advanced providers in every category. Online reviews are the newest method for patients to evaluate their providers. However, insufficient studies focus on the role of inherent physician characteristics, such as gender and years of experience, on patient satisfaction. We analyzed both quantitative and qualitative online reviews of 350 general dermatology providers at 121 Accreditation Council for Graduate Medical Education–accredited dermatology programs across the country to determine the effect of gender and years of experience. There were 38,008 online reviews of general dermatology providers. There was no significant difference in male and female overall ratings. Ratings were overall equally positive for both genders. Female providers were more likely to have positive written comments regarding time spent with patients (P = 0.027). New providers received highest overall, promptness, and time spent with patient ratings (P < 0.001). Medium experience providers received highest scores in bedside manner (P < 0.001), accurate diagnosis (P = 0.018), and ability to answer questions (P = 0.005). Advanced providers scored the lowest across all categories. In conclusion, gender did not significantly affect ratings, although females received more positive written comments on time spent with patients. Years of experience, however, is a significant factor in patient ratings, with new or medium experience providers scoring higher than advanced providers in every category.
In a 21st century dominated by VUCA environments (Volatile, Uncertain, Complex and Ambiguous) and in an increasingly diverse and global society, education should rethink how to meet the real needs of the citizens of the present and the future. Educational methods for language instruction have received assiduous attention from researchers, that may have overlooked educational ends, and that is to serve real life purposes. Learning a language is more than just acquiring knowledge about a new linguistic norm and its rules: it is above all, a vehicle for communication, an open channel to the world and a new scope with which new cultures are explored and different views and perspectives are discovered and shared. This paper aims at exploring task-based learning approach for language instruction and presenting a study on the benefits attributed to this approach, relating them to existing trends in current educational innovation. In doing so, a comparison between meaning-based learning and instruction-based learning is needed. Here we will review some of the most relevant theories and approaches to better understand task-based learning and explore its potential.
Abstract Listening to pleasurable music is known to engage the brain’s reward system. This has motivated many cognitive-behavioral interventions for healthy aging, but little is known about the effects of music-based intervention (MBI) on plasticity of the cognitive and reward systems. Here we show preliminary evidence that brain network connectivity can change after receptive MBI in cognitively unimpaired older adults. Using a combination of whole-brain regression, seed-based connectivity analysis, and representational similarity analysis (RSA), we examined fMRI responses during music listening in older adults before and after an eight-week personalized MBI. Participants rated self-selected and researcher-selected musical excerpts on liking and familiarity. Parametric effects of liking, familiarity, and selection showed simultaneous activation in auditory, reward, and default mode network (DMN) areas. Seed-based connectivity comparing pre- and post-intervention showed significant increase in functional connectivity between auditory regions and medial prefrontal cortex (mPFC); this auditory-mPFC connectivity was modulated by participant liking and familiarity ratings. RSA showed significant representations of selection and novelty at both time-points, and an increase in striatal representation of musical stimuli following intervention. Taken together, results show how regular music listening can provide an auditory channel towards the mPFC, thus offering a potential neural mechanism for MBI supporting healthy aging.
Facial expressions are a rich information source from which observers infer the emotional states of others. Despite much understanding about the brain regions that represent facial expressions, we do not yet know how representations of these facial movements transform into judgments of emotions in the brain. We addressed this question in 5 participants who judged the emotion of individual face movements called Action Units (AUs) while we concurrently measured brain activity using magnetoencephalography (MEG). Stimuli were animations of 5 facial movements--Outer Brower Raiser (AU2), Nose Wrinkler (AU9), Lip Corner Puller (AU12), Chin Raiser (AU17), Lip Stretcher (AU20), each at 4 levels of intensity (%25 - %100). We instructed participants to rate each animation according to either its perceived valence (‘negative’, ‘neutral’ or ‘positive’) or arousal (‘low,’ ‘neutral’ or ‘high’). Tasks alternated between blocks of 40 trials (5 AUs X 4 intensity levels X 2 repetitions) and participants completed 4,000 ~ 6,000 trials in total. We averaged all ratings of each AU and intensity level per task for each participant. We show that the arousal ratings increased along AU intensity levels while valence ratings are consistent for each AU (e.g., Nose Wrinkler (AU9) as negative and Lip Corner Puller (AU12) as positive). Then, we calculated Mutual Information (MI, permutation test) between MEG recording and task ratings. The results revealed the spatial and temporal distribution of brain activities related to the specific valence and arousal. We found that the valence and arousal evoked similar representational peaks ~270ms and ~750 ms in the temporal lobes while a special peak from parietal lobes at 387ms for valence task that differentiated between the two inferences. Our results show where (in temporal lobes and parietal lobes) and when (at ~270ms, 380ms and 750 ms post stimulus) the brain processes dynamic AUs as meaningful affective signals.
The complete semantic representation of a Tibetan sentence is mainly determined by the addition of a specific functional word. The choice of Tibetan functional words is mainly influenced (both explicitly and implicitly) by the sequence of Tibetan suffixes. In this article, we propose an RNN-based Tibetan radical suffix unit (TRSU) to consider this relationship. Specifically, for the Tibetan radical suffix unit-explicit (TRSU-E) method, the fixed suffix in Tibetan is used to determine the virtual functional words. For the Tibetan radical suffix unit-implicit (TRSU-I) method, the decision is assisted by adding a specific suffix. To test the method, we design a standard Tibetan corpus, which consists of different genres. Our experimental results show that the complexity of our method is reduced by up to 22.2% relative to the best baseline. Furthermore, with the hidden semantic information and implicit suffix, TRSU-I outperforms TRSU-E by reducing the perplexity (PPL) by 3%. Moreover, good results are achieved on the English Penn Treebank data set.
While Out-of-distribution (OOD) detection has been well explored in computer\nvision, there have been relatively few prior attempts in OOD detection for NLP\nclassification. In this paper we argue that these prior attempts do not fully\naddress the OOD problem and may suffer from data leakage and poor calibration\nof the resulting models. We present PnPOOD, a data augmentation technique to\nperform OOD detection via out-of-domain sample generation using the recently\nproposed Plug and Play Language Model (Dathathri et al., 2020). Our method\ngenerates high quality discriminative samples close to the class boundaries,\nresulting in accurate OOD detection at test time. We demonstrate that our model\noutperforms prior models on OOD sample detection, and exhibits lower\ncalibration error on the 20 newsgroup text and Stanford Sentiment Treebank\ndataset (Lang, 1995; Socheret al., 2013). We further highlight an important\ndata leakage issue with datasets used in prior attempts at OOD detection, and\nshare results on a new dataset for OOD detection that does not suffer from the\nsame problem.\n
Data augmentation techniques have been increasingly explored in natural language processing to create more textual data for training. However, the performance gain of existing techniques is often marginal. This paper explores the performance of combining two EDA (Easy Data Augmentation) methods, random swap and random delete for the performance in text classification. The classification tasks were conducted using CNN as a text classifier model on a portion of the SST-2: Stanford Sentiment Treebank dataset. The results show that the performance gain of this hybrid model performs worse than the benchmark accuracy. The research can be continued with a different combination of methods and experimented on larger datasets.
While educators may be well positioned to support unaccompanied immigrant youth, there is limited interdisciplinary research focused on understanding the complexity of youth’s experiences in US schools. The purpose of this qualitative, interview-based study was to better understand how youth’s transnational experiences pre-, during, and post-migration affected their school-based experiences, and to explore how schools supported them. Participants included ten unaccompanied immigrant youths from Central America and six key informants who worked with youth in a professional capacity. Findings indicate that youth experienced multiple challenges including stressful and traumatic events, barriers to mental health and legal services, and unfamiliar cultural and linguistic norms that sometimes were not recognized or understood by their teachers and schools. The youth also brought important resources, such as high expectations and aspirations and strong connections to family and community. School-based experiences that built from youth’s resources and motivations (e.g., through school-community partnerships and responsive classroom practices) had the potential to enhance belonging, community connections, and wellness. More interdisciplinary research is needed to develop and support school-based practices and partnerships in consultation with youth that build from knowledge of their particular resources and challenges.
Dictionary-based methods in sentiment analysis have received scholarly attention recently, the most comprehensive examples of which can be found in English.However, many other languages lack polarity dictionaries, or the existing ones are small in size as in the case of Senti-TurkNet, the first and only polarity dictionary in Turkish.Thus, this study aims to extend the content of SentiTurkNet by comparing the two available WordNets in Turkish, namely KeNet and TR-wordnet of BalkaNet.To this end, a current Turkish polarity dictionary has been created relying on 76,825 synsets matching KeNet, where each synset has been annotated with three polarity labels, which are positive, negative and neutral.Meanwhile, the comparison of KeNet and TR-wordnet of BalkaNet has revealed their weaknesses such as the repetition of the same senses, lack of necessary merges of the items belonging to the same synset and the presence of redundant narrower versions of synsets, which are discussed in light of their potential to the improvement of the current lexical databases of Turkish.
OBJECTIVE: Nonsuicidal self-injury (NSSI) is often cited as a key risk factor for future suicidal behavior. Capability for suicide has been repeatedly cited as an important mechanism that can account for this association. Despite this, direct tests of this hypothesis have been rare and methodologically constrained. In the present study, we conducted a direct test of this hypothesis while addressing several constraints of prior literature. METHOD: In a large sample of suicidal and self-injuring adults (n = 1,020), we tested whether changes in fearlessness about death (FAD), a core facet of the capability for suicide, accounted for the relationship between NSSI and future suicide attempts at 28-day and 2-year follow-up. FAD was assessed using the gold-standard self-report form (ACSS-FAD), an implicit test of suicide-related affect (affect misattribution paradigm-Suicide), and explicit affective ratings of suicide-relevant images. Mediation with bootstrapping was implemented to test our main hypotheses. RESULTS: As anticipated, lifetime NSSI frequency was significantly associated with suicide attempt frequency at follow-up; however, FAD failed to consistently mediate this association. Results were largely consistent across all three measures of FAD. Post hoc power analyses indicated sufficient power to detect small effects. CONCLUSIONS: Taken together, these results fail to support the hypothesis that capability for suicide explains the link between NSSI and future suicidal behavior. We discuss the implications of our results for research and theory, situating our findings in the context of recent advances in the understanding of suicide risk more broadly. (PsycInfo Database Record (c) 2021 APA, all rights reserved).
Morphological tagging of code-switching (CS) data becomes more challenging especially when language pairs composing the CS data have different morphological representations. In this paper, we explore a number of ways of implementing a language-aware morphological tagging method and present our approach for integrating language IDs into a transformerbased framework for CS morphological tagging. We perform our set of experiments on the Turkish-German SAGT Treebank. Experimental results show that including language IDs to the learning model significantly improves accuracy over other approaches.
OBJECTIVE: To evaluate remote testing as a tool for measuring emotional responses to non-speech sounds. DESIGN: Participants self-reported their hearing status and rated valence and arousal in response to non-speech sounds on an Internet crowdsourcing platform. These ratings were compared to data obtained in a laboratory setting with participants who had confirmed normal or impaired hearing. STUDY SAMPLE: Adults with normal and impaired hearing. RESULTS: In both settings, participants with hearing loss rated pleasant sounds as less pleasant than did their peers with normal hearing. The difference in valence ratings between groups was generally smaller when measured in the remote setting than in the laboratory setting. This difference was the result of participants with normal hearing rating sounds as less extreme (less pleasant, less unpleasant) in the remote setting than did their peers in the laboratory setting, whereas no such difference was noted for participants with hearing loss. Ratings of arousal were similar from participants with normal and impaired hearing; the similarity persisted in both settings. CONCLUSIONS: In both test settings, participants with hearing loss rated pleasant sounds as less pleasant than did their normal hearing counterparts. Future work is warranted to explain the ratings of participants with normal hearing.
OBJECTIVES: Age differences in affective experience across adulthood are widely documented. According to the circumplex model of affect consists of 2 aspects-valence (positive vs negative) and arousal (low activation vs high activation). Prior research on age differences has primarily focused on the valence aspect. However, little is known about age differences in daily affect of high and low arousal. METHOD: The present study examined age differences in daily dynamics (i.e., mean levels, variability, and inertia) of negative affect (NA) and positive affect (PA) of high and low arousal in a sample of 492 adults aged 21-91. Participants completed daily affect ratings for 21 consecutive days. RESULTS: Age was negatively and linearly related to mean levels of both high-arousal and low-arousal NA. Both high-arousal and low-arousal PA mean levels showed increases after middle age. Further, age was related to lower variability in both NA and PA regardless of arousal. Additionally, high-arousal NA inertia showed a linear decrease with age, whereas low-arousal PA inertia showed an inverted-U pattern with age. After controlling for mean levels of affect, the associations between age and affect variability remained significant, whereas the associations between age and affect inertia did not. DISCUSSION: The affective profile of older age is characterized by lower mean levels of NA, higher mean levels of PA, lower affect variability, and less persistence in high-arousal NA and low-arousal PA in daily life. Our results contribute to a nuanced understanding of which affective processes improve with age and which do not.
We propose two fast neural combinatory models for constituency parsing: binary and multibranching. Our models decompose the bottomup parsing process into 1) classification of tags, labels, and binary orientations or chunks and 2) vector composition based on the computed orientations or chunks. These models have theoretical sub-quadratic complexity and empirical linear complexity. The binary model achieves an F1 score of 92.54 on Penn Treebank, speeding at 1327.2 sents/sec. Both the models with XLNet provide near state-of-theart accuracies for English. Syntactic branching tendency and headedness of a language are observed during the training and inference processes for Penn Treebank, Chinese Treebank, and Keyaki Treebank (Japanese).
In this study, the affective explicit and implicit attitudes toward electric and gasoline cars are investigated. One hundred sixty-five participants (103 cisgender women, 62 cisgender men) completed an explicit and implicit affective rating task toward pictures of electric and gasoline cars, measurements of sustainability, future and past behaviors, and mindfulness. The results showed a positive emotional attitude for the electric cars compared with the gasoline cars only for the explicit rating but not for the implicit one. Furthermore, factors that correlated to the attitudes were investigated: explicit ratings in car owners correlated with age, degree, sustainability in general, and the expressed intention to purchase an electric car in the future. Implicit attitudes in car owners correlated with the overall score of mindfulness and the dimension of "non-reactivity." For the non-car owners, explicit attitudes correlated with the expressed intention to purchase an electric car in the future and the mindfulness dimension of "describing". In this group, the implicit attitude correlated negatively with the mindfulness intention of acting with awareness. This indicates that several different factors should be considered in the development of promotion campaigns for the advantage of sustainable mobility behavior.
People use their previous experience to predict present affective events. Since we live in ever-changing environments, affective predictions must generalize from past contexts (from which they are implicitly learned) to new, potentially ambiguous contexts. This study investigated how past (un)certain relationships influence subjective experience following new ambiguous cues, and whether past relationships can be learned implicitly. Two S1-S2 paradigms were employed as learning and test phases in two experiments. S1s were colored circles, S2s negative or neutral affective pictures. Participants (N = 121, 116) were assigned to the certain (CG) or uncertain group (UG), and they were presented with 100% (CG) or 50% (UG) S1-S2 congruency during an uninstructed (Experiment 1) or implicit (Experiment 2) learning phase. During the test phase both groups were presented with a new 75% S1-S2 paradigm, and ambiguous (Experiment 1) or unambiguous (Experiment 2) S1s. Participants were asked to rate the expected valence of upcoming S2s (expectancy ratings), or their experienced valence and arousal (valence and arousal ratings). In Experiment 1 ambiguous cues elicited less negative expectancy ratings, and less unpleasant valence ratings, independently from prior experience. In Experiment 2, participants in the CG reported more negative expectancy ratings after the S1s previously paired with negative stimuli. Overall, we found that in the presence of ambiguous cues subjective affective experience is dampened, and we confirmed that people are able to infer probabilistic relationships from the environment (and to use them later) at an implicit level.
STUDY OBJECTIVES: Sleep plays a pivotal role in the off-line processing of emotional memory. However, much remains unknown for its immediate vs. long-term influences. We employed behavioral and electrophysiological measures to investigate the short- and long-term impacts of sleep vs. sleep deprivation on emotional memory. METHODS: Fifty-nine participants incidentally learned 60 negative and 60 neutral pictures in the evening and were randomly assigned to either sleep or sleep deprivation conditions. We measured memory recognition and subjective affective ratings in 12- and 60-h post-encoding tests, with EEGs in the delayed test. RESULTS: In a 12-h post-encoding test, compared to sleep deprivation, sleep equally preserved both negative and neutral memory, and their affective tones. In the 60-h post-encoding test, negative and neutral memories declined significantly in the sleep group, with attenuated emotional responses to negative memories over time. Furthermore, two groups showed spatial-temporally distinguishable ERPs at the delayed test: while both groups showed the old-new frontal negativity (300-500 ms, FN400), sleep-deprived participants additionally showed an old-new parietal, Late Positive Component effect (600-1000 ms, LPC). Multivariate whole-brain ERPs analyses further suggested that sleep prioritized neural representation of emotion over memory processing, while they were less distinguishable in the sleep deprivation group. CONCLUSIONS: These data suggested that sleep's impact on emotional memory and affective responses is time-dependent: sleep preserved memories and affective tones in the short term, while ameliorating affective tones in the long term. Univariate and multivariate EEG analyses revealed different neurocognitive processing of remote, emotional memories between sleep and sleep deprivation groups.
Recurrent neural networks are efficient ways of training language models, and various RNN networks have been proposed to improve performance. However, with the increase of network scales, the overfitting problem becomes more urgent. In this paper, we propose a framework-G2Basy-to speed up the training process and ease the overfitting problem. Instead of using predefined hyperparameters, we devise a gradient increasing and decreasing technique that changes the parameters training batch size and input dropout simultaneously by a user-defined step size. Together with a pretrained word embedding initialization procedure and the introduction of different optimizers at different learning rates, our framework speeds up the training process dramatically and improves performance compared with a benchmark model of the same scale. For the word embedding initialization, we propose the concept of "artificial features" to describe the characteristics of the obtained word embeddings. We experiment on two of the most often used corpora-the Penn Treebank and WikiText-2 datasets-and both outperform the benchmark results and show potential towards further improvement. Furthermore, our framework shows better results with the larger and more complicated WikiText-2 corpus than with the Penn Treebank. Compared with other state-of-the-art results, we achieve comparable results with network scales hundreds of times smaller and within fewer training epochs.
BACKGROUND: Youth with anxiety disorders struggle with managing emotions relative to peers, but the neural basis of this difference has not been examined. METHODS: = 13.6; range = 8-17) with (n = 37) and without (n = 24) anxiety disorders completed a cognitive reappraisal task while undergoing functional magnetic resonance imaging. Emotional reactivity and regulation, functional activation, and beta-series connectivity were compared across groups. RESULTS: Groups did not differ on emotional reactivity or regulation. However, fronto-limbic activation after viewing aversive imagery with and without regulation, as well as affect ratings without regulation, were higher for anxious youth. Neither group demonstrated age-related changes in regulation, though anxious youth became less reactive with age. Stronger amygdala-ventromedial prefrontal cortex connectivity related to greater anxiety in control youth, but less anxiety in anxious youth. CONCLUSION: Anxious youth regulated when instructed, but regulation ability did not relate to age. Viewing aversive imagery related to heightened fronto-limbic activation even after reappraisal. Emotion dysregulation in youth anxiety disorders may stem from heightened emotionality and potent bottom-up neurobiological responses to aversive stimuli. Findings suggest the importance of treatments focused on both reducing initial emotional reactivity and bolstering regulatory capacity.
State-of-the-art neural language models represented by Transformers are becoming increasingly complex and expensive for practical applications. Low-bit deep neural network quantization techniques provides a powerful solution to dramatically reduce their model size. Current low-bit quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of the system to quantization errors. To this end, novel mixed precision DNN quantization methods are proposed in this paper. The optimal local precision settings are automatically learned using two techniques. The first is based on a quantization sensitivity metric in the form of Hessian trace weighted quantization perturbation. The second is based on mixed precision Transformer architecture search. Alternating direction methods of multipliers (ADMM) are used to efficiently train mixed precision quantized DNN systems. Experiments conducted on Penn Treebank (PTB) and a Switchboard corpus trained LF-MMI TDNN system suggest the proposed mixed precision Transformer quantization techniques achieved model size compression ratios of up to 16 times over the full precision baseline with no recognition performance degradation. When being used to compress a larger full precision Transformer LM with more layers, overall word error rate (WER) reductions up to 1.7% absolute (18% relative) were obtained.
A growing body of research analyzing musical scores suggests mode’s relationship with other expressive cues has changed over time. However, to the best of our knowledge, the perceptual implications of these changes have not been formally assessed. Here, we explore how compositional choices of 17th- and 19th-century composers (J. S. Bach and F. Chopin, respectively) differentially affect emotional communication. This novel exploration builds on our team’s previous techniques using commonality analysis to decompose intercorrelated cues in unaltered excerpts of influential compositions. In doing so, we offer an important naturalistic complement to traditional experimental work—often involving tightly controlled stimuli constructed to avoid the intercorrelations inherent to naturalistic music. Our data indicate intriguing changes in cues’ effects between Bach and Chopin, consistent with score-based research suggesting mode’s “meaning” changed across historical eras. For example, mode’s unique effect accounts for the most variance in valence ratings of Chopin’s preludes, whereas its shared use with attack rate plays a more prominent role in Bach’s. We discuss the implications of these findings as part of our field’s ongoing effort to understand the complexity of musical communication—addressing issues only visible when moving beyond stimuli created for scientific, rather than artistic, goals.