Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
16504 papers
Emotional memories are preferentially consolidated during sleep, through the process of memory reactivation. Targeted memory reactivation (TMR) has been shown to boost memory consolidation during sleep, but its neural correlates remain unclear, particularly for emotional memories. Here, we aimed to examine how TMR of emotional material during slow wave sleep (SWS) impacts upon neural processing during a subsequent arousal rating task. Participants were trained on a spatial memory task including negative and neutral pictures paired with semantically matching sounds. The picture-sound pairs were rated for emotional arousal before and after the spatial memory task. Then, half of the sounds from each emotional category (negative and neutral) were cued during SWS. The next day, participants were retested on both the arousal rating and the spatial memory task inside an MRI scanner, followed by another retest session a week later. Memory consolidation and arousal processing did not differ between cued and non-cued items of either emotional category. We found increased responses to emotional stimuli in the amygdala and orbitofrontal cortex (OFC), and a cueing versus emotion interaction in the OFC, whereby cueing neutral stimuli led to an increase in OFC activity, while cueing negative stimuli led to decreased OFC activation. Interestingly, the effect of cueing on amygdala activation was modulated by time spent in REM sleep. We conclude that SWS TMR impacts OFC activity, while REM sleep plays a role in mediating the effect of such cueing on amygdala.
<em>This article discusses the transition of language norms at school (phonetic, grammatical, linguistic norm) and the fact that language is a weapon of communication and exchange of ideas between people.</em>
The presence of English has grown in public and commercial spaces of urban localities across the world, yet the French language is gaining an increasing presence in Japan and is thus characterizing its linguistic landscape of its cities. Many language signs in these cities containing names of shops or products in French, appear on storefronts and packages. These texts seemingly convey shop policies or product images rather than a correct use of the French language. These situations require us to cast questions such as the following: 1) Which categories produce signs in French? 2) Which demographics constitute targeted receivers of these signs? 3) To what extent to these demographics comprehend the language n these signs? 4) What are the formal features specific to these signs? 5) What are the motivations and identities of the stakeholders of signs? 6) What are the motives for opting for French rather than other foreign languages in Japanese society? To respond to these questions, in this article, I will report on my preliminary analysis of signage data collected over several years throughout Japan’s urban centres. For this, I will discuss the presence of foreign languages in Japanese society and its language (loanwords, writing systems, foreign language in the media), after which, the article alludes to the ethnolinguistic vitality of French in the society. This vitality of the French language in Japan’s urban linguistic landscapes is described by the form and presence of metalinguistic variation, that is, the form and meaning of elements of French in signage in Japan’s cities.
Listening to music prompts strong emotional reactions in the listeners but relatively little research has focused on individual differences. This study addresses the role of musical preference and familiarity on emotions induced through music. A sample of 50 healthy participants (25 women) listened to 42 excerpts from the FMMS during 8 s while their autonomic and facial EMG responses were continuously recorded. Then, affective dimensions (hedonic valence, tension arousal, and energy arousal) and musical preference were rated using a 9-point scale, as well as familiarity using a 3-point scale. It was hypothesized that preferred and familiar music would be evaluated as more pleasant, energetic and less tense, and would prompt an increase of autonomic and zygomatic responses, and a decrease of corrugator activity. Results partially confirmed our hypothesis showing a strong effect of musical preference but not familiarity on emotion correlates. Specifically, musical preference predicted valence ratings, as well as HR acceleration and facial EMG activity. Overall, current findings suggested a great influence of musical preference on music-induced emotions, particularly modulating hedonic valence correlates. Our findings add evidence about the role of individual differences in the emotional processing through music and suggest the importance of considering those variables in future studies.
Inflammation could impact on the formation and persistence of interoceptive fear and hypervigilance, with relevance to psychiatric disorders and chronic pain. To systematically analyze effects of inflammation on fear learning and extinction, we performed two complementary randomized, double-blind, placebo-controlled functional magnetic resonance imaging (fMRI) studies combining experimental endotoxemia as a translational model of acute systemic inflammation with a two-day multiple-threat fear conditioning paradigm involving interoceptive and exteroceptive unconditioned stimuli (US). Healthy volunteers (N = 95) were randomized to receive intravenous injections of either endotoxin (lipopolysaccharide, LPS; 0.4 ng/kg) or placebo prior to fear acquisition (study 1) or extinction training (study2). Treatment effects on behavioral and neural responses to conditioned stimuli (CS) predicting interoceptive or exteroceptive threat were assessed during fear learning and extinction phases, along with US valence ratings. Despite robust inflammatory and emotional responses triggered by LPS, no direct effects of inflammation on US ratings or on the formation or extinction of conditioned fear, as assessed with CS valence ratings, were observed. However, in the group treated with LPS prior to acquisition (i.e., study 1), we found enhanced neural responses to the interoceptive but not the exteroceptive CS in key regions of the central fear circuitry during extinction learning. After extinction, this group further showed enhanced negative valence ratings selectively for the interoceptive US during unexpected US re-exposure when compared to the placebo group. Together, inflammation during fear acquisition may promote the establishment of a more robust neural signature of the interoceptive fear memory trace, which may contribute to altered interoceptive pain perception. The fear extinction circuitry engaged during interoceptive fear memory processing may be particularly vulnerable to inflammation, with transdiagnostic implications for gut-brain mechanisms underlying disturbed interoception in psychiatric conditions and chronic visceral pain.
PURPOSE: Adolescence is a delicate phase during life in which self-stigmatization increases and acceptance by peers becomes more important. However, little is known about how adolescents with cleft lip and/or palate experience this stage of life. Therefore, the purpose of this study was to investigate how cleft lip and/or palate (CLP) adolescents are looked at by their peers and how they look at others with/without CLP. METHODS: In this prospective, cross-sectional study 54 observers (CLP versus control) performed an eye-tracking task and gave attractiveness/ valence ratings. For this purpose, they were shown pictures of patients with and without CLP with neutral or smiling facial expressions. RESULTS: Adolescents with CLP were looked at differently compared to their unaffected peers, with shorter fixations of the eyes and longer fixations of the nose and mouth. Smiling altered the scan path toward the mouth for all faces. Contrary to the control group, adolescents with CLP tended to spend less time fixating the eyes. In the attractiveness/valence ratings, CLP adolescents were rated more negatively. CONCLUSIONS: Adolescents with cleft lip and/or palate look differently at peers and are also viewed with an alternate scan path.
This repository contains the evaluation sets used in <pre><code>A Benton, T Shi, O İrsoy, and I Malioutov."Weakly Supervised Headline Dependency Parsing". Findings of EMNLP. 2022. </code></pre> This dataset contains parse annotations for English news headlines and a script to produce conllu files joined with original headline text. Parse annotations are joined to the corresponding text by running: <pre> LDC_NYT_DIR="/PATH/TO/UNTARRED/LDC2008T19/" # path to untarred LDC2008T19 python build_eht.py --nyt_dir ${LDC_NYT_DIR} --num_proc 4</pre> This will download the Google sentence compression (GSC) dataset, and build conllu files for GSC examples. If you have the New York Times Annotated Corpus (LDC2008T19) untarred locally, this will also join annotations to the NYT examples (location passed via <code>--nyt_dir</code>). Increase the argument to <code>--num_procs</code> to process more shards from the NYT corpus in parallel and reduce build time. The above was tested with python 3.9.7. The EHT evaluation sets, with gold-annotated POS tags and dependency relations, are built as <code>EHT/gsc.test.conllu</code> and <code>EHT/nyt.test.conllu</code> Silver, projected, trees which we used to train and validate out models are built under <code>GSC_projected</code>. These are not gold parse trees (projected predictions from the article lead sentence), and are shared purely for reproducibility sake.
The convergence of artificial intelligence (AI) technology and natural language processing (NLP) has rapidly increased the demands for an analysis on the natural language that involves plenty of ambiguities not present in formal language. For this reason, the language model (LM), a statistical approach, has been used as a key role in this area. Recently, the emerging field of deep learning, which applies complex deep neural networks for machine learning tasks, has been applied to language modeling and achieved more remarkable results than traditional language models. One of the important techniques that have led neural network-based LM success is the attention mechanism. Attention mechanism makes neural networks pay attention to specific words in the input sentence when generating the output words. However, although the attention mechanism has improved the performance of many neural network models, it requires tons of parameters to achieve the state-of-art level performance. This is because attention mechanism encodes the context of a word by simply accumulating the outputs from the network for all the input words, which may cause information loss. To compensate for this limitation, we propose an extension of attention mechanism by adopting a convolutional neural network to replace the accumulation. With only far fewer parameters, our model achieved comparable performance to the recent state-of-the-art models on the very popular benchmark datasets, yielding perplexity scores of 58.4 on the Penn Treebank dataset and 50.1 on the Wikitext-2 dataset, respectively.
Facial expressions are indispensable in daily human communication. Previous neuroimaging studies investigating facial expression processing have presented pre-recorded stimuli and lacked live face-to-face interaction. Our paradigm alternated between presentations of real-time model performance and pre-recorded videos of dynamic facial expressions to participants. Simultaneous functional magnetic resonance imaging (fMRI) and facial electromyography activity recordings, as well as post-scan valence and arousal ratings were acquired from 44 female participants. Live facial expressions enhanced the subjective valence and arousal ratings as well as facial muscular responses. Live performances showed greater engagement of the right posterior superior temporal sulcus (pSTS), right inferior frontal gyrus (IFG), right amygdala and right fusiform gyrus, and modulated the effective connectivity within the right mirror neuron system (IFG, pSTS, and right inferior parietal lobule). A support vector machine algorithm could classify multivoxel activation patterns in brain regions involved in dynamic facial expression processing in the mentalizing networks (anterior and posterior cingulate cortex). These results indicate that live social interaction modulates the activity and connectivity of the right mirror neuron system and enhances spontaneous mimicry, further facilitating emotional contagion.
Abstract Tupí-Guaraní is one of the largest branches of the Tupían language family, but despite its relevance there is no consensus about its origins in terms of age, homeland, and expansion. Linguistic classifications vary significantly, with archaeological studies suggesting incompatible date ranges while ethnographic literature confirms the close similarities as a result of continuous inter-family contact. To investigate this issue, we use a linguistic database of cognate data, employing Bayesian phylogenetic methods to infer a dated tree and to build a phylogeographic expansion model. Results suggest that the branch originated around 2500 BP in the area of the upper course of the Tapajós-Xingu basins, with a split between Southern and Northern varieties starting around 1750 BP. We analyse the difficulties in reconciling archaeological and linguistic data for this group, stressing the importance of developing an interdisciplinary unified model that incorporates evidence from both disciplines.
We illustrate an approach for multilingual treebanks explorations by introducing a novel adaptation to small treebanks of a methodology for identifying cross-lingual quantitative trends in the distribution of dependency relations. By relying on the principles of cross-validation, we reduce the amount of data required to execute the method, paving the way to expanding its use to low-resources languages. We validated the approach on 8 small treebanks, each containing less than 100,000 tokens and representing typologically different languages. We also show preliminary but promising evidence on the use of the proposed methodology for treebank expansion.
The affective variability of bipolar disorder (BD) is thought to qualitatively differ from that of borderline personality disorder (BPD), with changes in affect persisting longer in BD. However, quantitative studies have not been able to confirm this distinction. It has therefore not been possible to accurately quantify how treatments like lithium influence affective variability in BD. We assessed the affective variability associated with BD and BPD as well as the effect of lithium using a computational model that defines two subtypes of variability: affective changes that persist (volatility) and changes that do not (noise). We hypothesized that affective volatility would be raised in the BD group, noise would be raised in the BPD group, and that lithium would impact affective volatility. Daily affect ratings were prospectively collected for up to 3 y from patients with BD or BPD and nonclinical controls. In a separate experimental medicine study, patients with BD were randomized to receive lithium or placebo, with affect ratings collected from week -2 to +4. We found a diagnostically specific pattern of affective variability. Affective volatility was raised in patients with BD, whereas affective noise was raised in patients with BPD. Rather than suppressing affective variability, lithium increased the volatility of positive affect in both studies. These results provide a quantitative measure of the affective variability associated with BD and BPD. They suggest a mechanism of action for lithium, whereby periods of persistently low or high affect are avoided by increasing the volatility of affective responses.
Treebanks are one of the most needed and used linguistic resources in the fields of Natural language processing (NLP) and Natural language understanding (NLU). Arabic has only two constituency-based treebanks and a number of dependency treebanks. The current research presents the guidelines for building a parsed Arabic treebank for Modern Standard Arabic (MSA). The guidelines show, firstly the choice of the grammar formalism, then the genre and size of the treebank, and finally the annotation layers of the treebank. The study also shows that using the traditional Arabic grammar syntactic theory to describe the Arabic syntax has proven to be more suitable than using any of the modern syntax theories. Working with the traditional Arabic grammar also helps avoid the errors that the available treebank fell in as a result of using guidelines that don't suit the Arabic grammar. The study adopts three layers of annotations: the morphological layer, the syntactic layer, and the grammatical function layer. The resultant tree is a very detailed and rich syntactic tree, which is preferable by the researcher over having a huge amount of data poorly and shallowly annotated.
Lexical substitution is the task of generating meaningful substitutes for a word in a given textual context. Contextual word embedding models have achieved state-of-the-art results in the lexical substitution task by relying on contextual information extracted from the replaced word within the sentence. However, such models do not take into account structured knowledge that exists in external lexical databases.
Abstract Prior work suggests that more frequent or higher exposure to stressors relates to less positive affect and more negative affect in daily life. Limited knowledge exists about whether subjective appraisals of such stressors (i.e., perceived negative impacts on daily routine, personal health and safety, and finances) also have negative links to daily well-being. This study examines this link using data from an 8-day daily dairy study (n=675 days) in an online sample of older adults (n = 110 people, ages 60-90). We also explored potential psychological moderators particularly relevant to the experience of aging (i.e., self-views of aging, S-VOA). Results from multilevel models indicate that people reported more negative affect and less positive affect on days with more negative appraisals, especially on those days when they also had more negative self-views of aging. These findings highlight S-VOA as psychological resources that help people cope with stressful events in everyday life.
In the area of natural language processing, measuring sentence similarity is an essential problem. Searching for semantic meaning in natural language is a related issue. The task of measuring sentence similarity is to find semantic symmetry in two sentences, not matter how they are arranged. It is important to measure the similarity of sentences accurately. To compute the similarity between sentences, existing methods have been constructed from approaches for large texts. Since these methods work in very high-dimensional spaces, they are inefficient, require human input, and are not flexible enough for some applications. In this study, we propose a hybrid method (HydMethod) which considers not only semantic information including lexical databases, word embeddings, and corpus statistics, but also implied word order information. With lexical databases, our method models human common sense knowledge, and that knowledge can then be adapted to be used in different domains with the incorporation of corpus statistics. Therefore, the methodology is applicable across several domains. As part of our experiments, we used two standard datasets - Pilot Short Text Semantic Similarity Benchmark and MS paraphrase - in order to demonstrate the efficacy of our proposed method. As a result, the proposed method outperforms the existing approaches when tested on these two datasets, giving the highest correlation value for both word and sentence similarity. Moreover, it achieves a maximum of 32% higher increase than only using word vector or WorldNet based methodology. With Rubenstein and Goodenough word & sentence pairs, our algorithm's similarity measure shows a high Pearson correlation coefficient of 0.8953.
showed that, treebank size aside, overlap between training and test graphs (termed leakage) explains more of the observed variation in dependency parsing performance than other explanations. In this work we revisit this claim, testing it on more models and languages. We find that it only holds for zero-shot cross-lingual settings. We then propose a more fine-grained measure of such leakage which, unlike the original measure, not only explains but also correlates with observed performance variation.
In this paper, we discuss the development of treebanks for two low-resourced Indian languages - Magahi and Braj based on the Universal Dependencies framework. The Magahi treebank contains 945 sentences and Braj treebank around 500 sentences marked with their lemmas, part-of-speech, morphological features and universal dependencies. This paper gives a description of the different dependency relationship found in the two languages and give some statistics of the two treebanks. The dataset will be made publicly available on Universal Dependency (UD) repository (https://github.com/UniversalDependencies/UD_Magahi-MGTB/tree/master) in the next(v2.10) release.
Building computational resources and tools for the under-resourced languages is strenuous for any Natural Language Processing task. This article presents the first dependency parser for an under-resourced Indian language, Nepali. A prerequisite for developing a parser for a language is a corpus annotated with the desired linguistic representations known as a treebank. With an aim of cross-lingual learning and typological research, we use a Bengali treebank to build a Bengali-Nepali parallel corpus and apply the method of annotation projection from the Bengali treebank to build a treebank for Nepali. With the developed treebank, MaltParser (with all algorithms for projective dependency structures) and a Neural network-based parser have been used to build Nepali parser models. The Neural network-based parser produced state-of-the-art results with 81.2 Unlabeled Attachment Score, 73.2 Label Accuracy, and 66.1 Labeled Attachment Score on the gold test data. The parser models have also been evaluated with the predicted Part-of-speech (POS)-tagged test data. A statistical POS tagger using Conditional Random Field has been developed for predicting the POS tags of the test data.
This study sought to determine if hues overlayed on a video recording of a piano performance would systematically influence perception of its emotional arousal level. The hues were artificially added to a series of four short video excerpts of different performances using video editing software. Over two experiments 106 participants were sorted into 4 conditions, with each viewing different combinations of musical excerpts (two excerpts with nominally high arousal and two excerpts with nominally low arousal) and hue (red or blue) combinations. Participants rated the emotional arousal depicted by each excerpt. Results indicated that the overall arousal ratings were consistent with the nominal arousal of the selected excerpts. However, hues added to video produced no significant effect on arousal ratings, contrary to predictions. This could be due to the domination of the combined effects of other channels of information (e.g., the music and player movement) over the emotional effects of the hypothesized influence of hue on perceived performance (red expected to enhance and blue to reduce arousal of the performance). To our knowledge this is the first study to investigate the impact of these hues upon perceived arousal of music performance, and has implications for musical performers and stage lighting. Further research that investigates reactions during live performance and manipulation of a wider range of lighting hues, saturation and brightness levels, and editing techniques, is recommended to further scrutinize the veracity of the findings.
Visual processing of emotional words modulates early event-related potentials (ERPs) such as the early posterior negativity (EPN). Questions remain as to whether this modulation reflects modality-specific processing, preferentially elicited by emotional words of the native language (L1). This study investigates the modulation of early ERPs during rapid serial visual presentation (RSVP) of adjectives or nouns referring to emotional feeling states, neutral traits, to an overweight or lean body or to concrete body parts or neutral objects, presented in the L1 and the second language (L2). Word ratings in the L2 were assessed in a pilot study. The N100 and the EPN were modulated by the emotional valence of the stimuli irrespective of the word class or the task (silent reading vs. word counting). The results suggest that early affective appraisal is obligatory, not restricted to privileged categories of linguistic information (emotions) or solely found for the embodied language (L1).
This paper investigates linguistic complexity across natural languages from a corpus-based perspective and relies on the assumptions of linguistic profiling as a methodological framework. We focus in particular on the domain of syntactic complexity and analyze the distribution of a set of features taken as proxies of complexity phenomena at sentence level, which were extracted from 63 treebanks annotated according to the Universal Dependencies formalism. This dataset guarantees that the features considered are modeling the same linguistic phenomena in different treebanks, allowing reliable comparison among languages. We show that our approach is able to identify tendencies of structural proximity between languages not necessarily in line with typologically-supported classification, thus shedding light on new corpus-based findings.
With the advent of the WordNet (a lexical database), the field of linguistics ushered into the new realm where this structured database, based on ontology can be used for a variety of operations such as Word Sense Disambiguation (WSD), checking Semantic Similarity, performing Text Based Summarization etc. The development of various knowledge bases/lexicons like WordNet, with graphically structured information allowed many graph based approaches to be used for choosing the best sense in the Word Sense Disambiguation process. In these graph based approaches, local graph connectivity measures like Fuzzy Degree Centrality ($\mathrm{V}_{\mathrm{F}\mathrm{D}\mathrm{C}}$), Fuzzy Closeness Centrality ($\mathrm{V}_{\mathrm{F}\mathrm{C}\mathrm{C}}$), Fuzzy Betweenness Centrality ($\mathrm{V}_{\mathrm{F}\mathrm{B}\mathrm{C}}$), Fuzzy PageRank Centrality ($\mathrm{V}_{\mathrm{F}\mathrm{P}\mathrm{C}}$) and global graph connectivity measures like Fuzzy Compactness ($\mathrm{V}_{\mathrm{F}\mathrm{C}}$) and Fuzzy Graph Entropy ($\mathrm{V}_{\mathrm{F}\mathrm{G}\mathrm{E}}$) are widely used to perform WSD.Proposed and implemented here is a novel fuzzy graph connectivity measure named as Fuzzy Expected Force (FExF) to choose the correct word sense, by taking into consideration the influential as well as less influential nodes for the node under consideration in the graph formed using Fuzzy Hindi WordNet. FExF performs less computations than $\mathrm{V}_{\mathrm{F}\mathrm{G}\mathrm{E}}$ to give accurate results; thus becoming a reliable local graph connectivity measure, as overall structure of graph is not necessary for finding correct sense of the word.
For fans of popular cultural products, digitization has meant the configuration of affinity spaces online and opportunities for learning in the “digital wilds,” including incidental language learning and identity development. Through online multiparty written interaction, we explored how 15 Catalan-speaking gamers organized themselves to translate games from English into Catalan. Results indicated distributed roles, routinized translation practices, and, through translating and commenting on new members’ translation tests, the emergence of folk linguistic attitudes and beliefs regarding Spanish-Catalan interference, changes to Catalan linguistic norms, or positionings towards the group’s English-Catalan translation strategy: They favored creative expressions and what they believed were idiosyncratic language choices with symbolic power to signal what Catalan is or ought to be. The group of fan translators configured a site for metapragmatic discussion and identity development which navigates across monoglot/heteroglossic, prescriptive/descriptive/idiosyncratic conceptions of language, manifesting grassroots linguistic activism fueled by the diglossic status of the Catalan language.
A gentle touch is an essential part of human interaction that produces a positive care effect. Previously, robotics studies have shown that robots can reproduce a gentle touch that elicits similar, positive emotional responses in humans. However, whether the positive emotional effects of a robot's touch combined with speech can be enhanced using a multimodal approach remains unclear. This study supports the hypothesis that a multimodal interaction combining gentle touch and speech by a robot enhances positive emotional responses. Here, we conducted an experiment using a robotic arm to perform a gentle touch combined with speech and compared three conditions: touch alone, speech alone, and touch with speech. We assessed participants' subjective ratings of valence, arousal, and human likeliness using subjective emotional responses. Furthermore, we recorded facial electromyography (EMG) from the corrugator supercilii and zygomaticus major muscles and measured skin conductance levels (SCLs) as physiological emotional responses. Our results show that touch combined with speech elicited higher subjective valence and arousal ratings, stronger zygomaticus major EMG and SCL activities than touch alone. The results suggest that the positive emotional effects of robotic touch can be boosted by combining elements of speech.
INTRODUCTION: Drug addiction is characterized by impaired response inhibition and salience attribution (iRISA), where the salience of drug cues is postulated to overpower that of other reinforcers with a concomitant decrease in self-control. However, the neural underpinnings of the interaction between the salience of drug cues and inhibitory control in drug addiction remain unclear. METHODS: We developed a novel stop-signal functional magnetic resonance imaging task where the stop-signal reaction time (SSRT-a classical inhibitory control measure) was tested under different salience conditions (modulated by drug, food, threat, or neutral words) in individuals with cocaine use disorder (CUD; n = 26) versus demographically matched healthy control participants (n = 26). RESULTS: Despite similarities in drug cue-related SSRT and valence and arousal word ratings between groups, dorsolateral prefrontal cortex (dlPFC) activity was diminished during the successful inhibition of drug versus food cues in CUD and was correlated with lower frequency of recent use, lower craving, and longer abstinence (Z > 3.1, P < 0.05 corrected). DISCUSSION: Results suggest altered involvement of cognitive control regions (e.g. dlPFC) during inhibitory control under a drug context, relative to an alternative reinforcer, in CUD. Supporting the iRISA model, these results elucidate the direct impact of drug-related cue reactivity on the neural signature of inhibitory control in drug addiction.
The present study investigates the relations between L2-English proficiency and L1-Turkish lexical property evaluations. We asked whether L2 proficiency affects lexical properties, including imageability and concreteness ratings of 600 Turkish words selected from the Word Frequency Dictionary of Written Turkish. Seventy-two participants (L1-Turkish - L2-English) provided ratings of concreteness and imageability for 600 words on a 7-point scale. In order to assess their L2 proficiency, we administered Peabody Picture Vocabulary Test-IV (PPVT-IV). We divided categories into two subcategories as high and low for the frequency, concreteness, imageability, and age of acquisition (AoA). The relationship between these subcategories and imageability-concreteness was examined by mixed effects linear regression analyses. We found that L2 proficiency and imageability ratings were positively correlated and specifically, this positive association was evident for low-frequency words and later acquired words. Results are in line with the interaction of bilingual representation under the dual-coding theory which suggests that bilinguals develop an interconnected imaginal representation for two languages as opposed to separate verbal representations. As L2 proficiency increased, the imageability also increased. These findings have implications for literature investigating the relationship between L2 proficiency and linguistic outcomes. Additionally, findings point to the importance of considering the L2 proficiency of participants when lexical tasks that involve cue words or word lists are used.
Aging is accompanied by deterioration in both working memory (WM) and long-term memory (LTM), yet whether these changes are related is not understood. Sleep plays a role in the formation of LTM in young adults, but the findings in older adults are not as clear. The types of memories we store also shift with age as young adults preserve a higher proportion of negative experiences when compared to older adults. The reason for this age-related change in emotional memory bias is also not clear; however, some studies have suggested that WM changes across aging may be an important factor. In the current study, we examined performance in WM and emotional LTM in younger and older adults. We added a daytime nap in half the subjects to examine a possible role of sleep on emotional LTM. In the morning, 93 younger (18-39) and 121 older (60-85) adults completed a WM task. Subjects also encoded neutral or negative word pairs and provided valence and arousal ratings for each pair. After half the subjects took a daytime nap, LTM was examined, and valence and arousal ratings were reassessed. Results indicate that older adults showed worse recognition for negative word pairs compared with neutral, as well as decreased negative valence ratings in the afternoon. This decrease in emotional reactivity was correlated with better LTM performance. In contrast, younger adults performed better on the negative compared to neutral word pairs, with no change in emotional reactivity and no association between emotional reactivity and LTM. In addition, WM was positively related to LTM in younger, but not in older adults. Lastly, no differences were shown across sleep, regardless of age. Our findings suggest that the emotional memory bias may be associated with the emotional saliency of the information in older adults, and with WM capacity in younger adults. Supplementary Information: The online version contains supplementary material available at 10.1007/s42761-022-00134-5.
Unsupervised cross-lingual transfer has been shown great potentials for dependency parsing of the low-resource languages when there is no annotated treebank available. Recently, the self-training method has received increasing interests because of its state-of-the-art performance in this scenario. In this work, we advance the method further by coupling it with curriculum learning, which guides the self-training in an easy-to-hard manner. Concretely, we present a novel metric to measure the instance difficulty of a dependency parser which is trained mainly on a Treebank from a resource-rich source language. By using the metric, we divide a low-resource target language into several fine-grained sub-languages by their difficulties, and then apply iterative-self-training progressively on these sub-languages. To fully explore the auto-parsed training corpus from sub-languages, we exploit an improved parameter generation network to model the sub-languages for better representation learning. Experimental results show that our final curriculum-style self-training can outperform a range of strong baselines, leading to new state-of-the-art results on unsupervised cross-lingual dependency parsing. We also conduct detailed experimental analyses to examine the proposed approach in depth for comprehensive understandings.
In Chinese Internet spaces, "fishing" is a form of non-literal communication that tries to lure or bait others. We conducted a case study of fishing on a Chinese Q&A platform in which we deconstructed the linguistic structure of the fishing language and identified its features by employing analysis techniques informed by Critical Discourse Analysis (CDA). Additionally, we examined the relationship between fishing practices and the specifics of the underlying communication platform. We found that fishing can be characterized as a reaction to rigid norms of politeness and friendliness imposed on users by platform owners. Our analysis reveals that fishing is organically connected with other social interaction within the community and facilitates autonomous and active negotiation of boundaries and linguistic norms for online discussion of controversial topics. We challenge the characterization of non-literal communication as "abnormal," and contribute a more refined understanding of online linguistic norms, community cohesion, and civil engagement.
The prevailing practice in the academia is to evaluate the model performance on in-domain evaluation data typically set aside from the training corpus. However, in many real world applications the data on which the model is applied may very substantially differ from the characteristics of the training data. In this paper, we focus on Finnish out-of-domain parsing by introducing a novel UD Finnish-OOD out-of-domain treebank including five very distinct data sources (web documents, clinical, online discussions, tweets, and poetry), and a total of 19,382 syntactic words in 2,122 sentences released under the Universal Dependencies framework. Together with the new treebank, we present extensive out-of-domain parsing evaluation utilizing the available section-level information from three different Finnish UD treebanks (TDT, PUD, OOD). Compared to the previously existing treebanks, the new Finnish-OOD is shown include sections more challenging for the general parser, creating an interesting evaluation setting and yielding valuable information for those applying the parser outside of its training domain.
In this paper, we investigate to which extent contextual neural language models (LMs) implicitly learn syntactic structure. More concretely, we focus on constituent structure as represented in the Penn Treebank (PTB). Using standard probing techniques based on diagnostic classifiers, we assess the accuracy of representing constituents of different categories within the neuron activations of a LM such as RoBERTa. In order to make sure that our probe focuses on syntactic knowledge and not on implicit semantic generalizations, we also experiment on a PTB version that is obtained by randomly replacing constituents with each other while keeping syntactic structure, i.e., a semantically ill-formed but syntactically well-formed version of the PTB. We find that 4 pretrained transfomer LMs obtain high performance on our probing tasks even on manipulated data, suggesting that semantic and syntactic knowledge in their representations can be separated and that constituency information is in fact learned by the LM. Moreover, we show that a complete constituency tree can be linearly separated from LM representations.
Recent research into grammatical gender from the perspective of information theory has shown how seemingly arbitrary gender systems can ease processing demands by guiding lexical prediction. When the gender of a noun is revealed in a preceding element, the list of possible candidates is reduced to the nouns assigned to that gender. This strategy can be particularly effective if it eliminates words that are likely to compete for activation against the intended word. We propose syntax as the crucial context within which words must be disambiguated, hypothesizing that syntactically similar words should be less likely to share a gender cross-linguistically. We draw on recent work on syntactic information in the lexicon to define the syntactic distribution of a word as a probability vector of its participation in various dependency relations, and we extract such relations for 32 languages from the Universal Dependencies Treebanks. Correlational and mixed-effects regression analyses reveal that syntactically similar nouns are less likely to share a gender, the opposite pattern that is found for semantically and orthographically similar words. We interpret this finding as a design feature of language, and this study adds to a growing body of research attesting to the ways in which functional pressures on learning, memory, production, and perception shape the lexicon in different ways.
In this paper, we launch a new Universal Dependencies treebank for an endangered language from Amazonia: Kakataibo, a Panoan language spoken in Peru. We first discuss the collaborative methodology implemented, which proved effective to create a treebank in the context of a Computational Linguistic course for undergraduates. Then, we describe the general details of the treebank and the language-specific considerations implemented for the proposed annotation. We finally conduct some experiments on part-of-speech tagging and syntactic dependency parsing. We focus on monolingual and transfer learning settings, where we study the impact of a Shipibo-Konibo treebank, another Panoan language resource.
While scholars increasingly link affective polarization to the rise of populist parties, existing empirical studies are limited to the effects of radical right parties, without considering the possible effects of leftist populist parties or of parties' varying degrees of populism. Analyzing novel survey data across eight European publics, we analyze whether citizens' affective party evaluations broadly map onto these parties' varying degrees of populism, along with their Left-Right ideologies. We scale survey respondents' party feeling thermometer evaluations and social distance ratings of rival partisans using multidimensional scaling (MDS) to estimate a two-dimensional affective partisan space for each mass public, finding that in most (though not all) publics our mappings are strongly related to the parties' varying degrees of populism, as well as to Left-Right ideology. We substantiate these conclusions via analyses regressing respondents' affective ratings against exogenous measures of the parties' Left-Right ideologies and their degrees of populism. Our findings suggest that in many European publics, populism structures citizens' affective ratings of parties (and of their supporters) to roughly the same degree as Left-Right ideology.
Chills are a peak emotional experience induced by stimuli such as music, films, and speeches. The emotional consequences of chills in terms of valence and arousal are still debated and the existing empirical data is conflicting. In this study, we tested the effects of ChillsDB, an open-source repository of chills-inducing stimuli, on the emotional ratings of 600+ participants. We found that chills have a positive effect on valence and arousal as participants experiencing chills reported a significantly greater valence and arousal after the experience than before, as compared to participants who did not experience chills. Furthermore, we found an interesting pattern in the valence ratings of participants, which tended to harmonize towards a similar mean after the experiment, though initially disparately distributed.
International audience
We analyze Recurrent Neural Network (RNN) architectures to handle the problem of Part-of-Speech (POS) Tagging. When linguistic rules are inserted ad-hoc into the decision algorithm, there is a difficulty in understanding the role of prior information and learning. The real potential of recurrent networks is demonstrated in this paper on the Italian language in a purely data-driven approach, where we can reach the state-of-the-art on the UD_Italian-ISTD (Italian Stanford Dependency Treebank) dataset in comparison to TINT. We propose a methodology for splitting words that are mapped to embedding spaces and fed to forward-backward networks.
With the development of deep learning, neural networks are widely used in various fields, and the improved model performance also introduces a considerable number of parameters and computations. Model quantisation is a technique that turns floating-point computing into low-specific-point computing, which can effectively reduce model computation strength, parameter size, and memory consumption but often bring a considerable loss of accuracy. This paper mainly addresses the problem where the distribution of parameters is too concentrated during quantisation aware training (QAT). In the QAT process, we use a piecewise function to statistics the parameter distributions and simulate the effect of quantisation noise in each round of training, based on the statistical results. Experimental results show that by quantising the Transformer network, we lose less precision and significantly reduce the storage cost of the model; compared with the full precision LSTM network, our model has higher accuracy under the condition of a similar storage cost. Meanwhile, compared with other quantisation methods on language modelling task, our approach is more accurate. We validated the effectiveness of our policy on the WikiText-103 and PENN Treebank datasets. The experiments show that our method extremely compresses the storage cost and maintains high model performance.