ID Task Research Articles

ID-SF-Fusion: a cooperative model of intent detection and slot filling for natural language understanding

PurposeIntent detection (ID) and slot filling (SF) are two important tasks in natural language understanding. ID is to identify the main intent of a paragraph of text. The goal of SF is to extract the information that is important to the intent from the input sentence. However, most of the existing methods use sentence-level intention recognition, which has the risk of error propagation, and the relationship between intention recognition and SF is not explicitly modeled. Aiming at this problem, this paper proposes a collaborative model of ID and SF for intelligent spoken language understanding called ID-SF-Fusion.Design/methodology/approachID-SF-Fusion uses Bidirectional Encoder Representation from Transformers (BERT) and Bidirectional Long Short-Term Memory (BiLSTM) to extract effective word embedding and context vectors containing the whole sentence information respectively. Fusion layer is used to provide intent–slot fusion information for SF task. In this way, the relationship between ID and SF task is fully explicitly modeled. This layer takes the result of ID and slot context vectors as input to obtain the fusion information which contains both ID result and slot information. Meanwhile, to further reduce error propagation, we use word-level ID for the ID-SF-Fusion model. Finally, two tasks of ID and SF are realized by joint optimization training.FindingsWe conducted experiments on two public datasets, Airline Travel Information Systems (ATIS) and Snips. The results show that the Intent ACC score and Slot F1 score of ID-SF-Fusion on ATIS and Snips are 98.0 per cent and 95.8 per cent, respectively, and the two indicators on Snips dataset are 98.6 per cent and 96.7 per cent, respectively. These models are superior to slot-gated, SF-ID NetWork, stack-Prop and other models. In addition, ablation experiments were performed to further analyze and discuss the proposed model.Originality/valueThis paper uses word-level intent recognition and introduces intent information into the SF process, which is a significant improvement on both data sets.

The Impact of Phonetic Training in the Perception and Poduction of European Portuguese Stops by Mandarin Speakers

This study aims to improve knowledge regarding the process of L2 consonants acquisition and investigate the effectiveness of training strategies that can favor its learning. It compares the effect of a phonetic training using two High Variability Phonetic Training (HVPT) methods on training the perception of European Portuguese (EP) stop consonants by Mandarin native speakers, learners of EP as a L2. More specifically, it will investigate if the expected improvement is greater under the HVPT approach using an identification task (ID) or through an AX discrimination task (DIS). In a pretest-training-posttest design, thirty L2 learners of EP will be divided into 2 experimental groups and a control group. Experimental groups will differ in the training method: ID task and DIS task. Participants will be tested on their ability to perceive the target sounds presented in non-words and real words before training, after training and two months later. Thus, this study will address the following questions: a) whether training can improve the perception and production of trained as well as untrained segments, b) whether improvement generalizes to novel stimuli and talkers, c) if improvement is retained over time, and d) which training method is more effective (ID or DIS).

Transcutaneous auricular vagus nerve stimulation enhances learning of novel letter-sound relationships in adults

BackgroundReading is a critical skill in modern society but is significantly more difficult to acquire during adulthood. Many adults are required to learn a new orthography after this window closes for personal or vocational reasons and while many programs and training methods exist for learning to read in adulthood, none result in native-like fluency. Implantable cervical vagus nerve stimulation is capable of driving neural plasticity but is invasive and not practical as a reading intervention. ObjectiveThe goal of the current study was to evaluate whether non-invasive transcutaneous auricular vagus nerve stimulation (taVNS) is effective at enhancing novel orthography acquisition in young adults. MethodsWe enrolled 37 typically developing participants and randomly assigned them to a computer control, device sham control, earlobe stimulation control, or experimental transcutaneous auricular stimulation (taVNS) group. Participants then learned novel letter-sound correspondences in Hebrew over five training lessons. Performance was assessed using three measures to evaluate various aspects of reading: Letter ID, Automaticity, and Decoding. ResultsThe taVNS group significantly outperformed the three control groups on both the Automaticity and Decoding tasks. There was no difference on the Letter ID task. ConclusionsThese results demonstrate, for the first time, that taVNS is capable of improving aspects of reading acquisition in adults. These findings have potential implications for a wide range of cognitive tasks.

Unsupervised Person Re-identification via Cross-camera Similarity Exploration.

Most person re-identification (re-ID) approaches are based on supervised learning, which requires manually annotated data. However, it is not only resource-intensive to acquire identity annotation but also impractical for large-scale data. To relieve this problem, we propose a cross-camera unsupervised approach that makes use of unsupervised style-transferred images to jointly optimize a convolutional neural network (CNN) and the relationship among the individual samples for person re-ID. Our algorithm considers two fundamental facts in the re- ID task, i.e., variance across diverse cameras and similarity within the same identity. In this paper, we propose an iterative framework which overcomes the camera variance and achieves across-camera similarity exploration. Specifically, we apply an unsupervised style transfer model to generate style-transferred training images with different camera styles. Then we iteratively exploit the similarity within the same identity from both the original and the style-transferred data. We start with considering each training image as a different class to initialize the Convolutional Neural Network (CNN) model. Then we measure the similarity and gradually group similar samples into one class, which increases similarity within each identity. We also introduce a diversity regularization term in the clustering to balance the cluster distribution. The experimental results demonstrate that our algorithm is not only superior to state-of-the-art unsupervised re-ID approaches, but also performs favorably compared with other competing unsupervised domain adaptation methods (UDA) and semi-supervised learning methods.

Language ID for a Thousand Languages

ODIN, the Online Database of INterlinear text, is a resource built over language data harvested from linguistic documents (Lewis, 2006). It currently holds approximately 190,000 instances of Interlinear Glossed Text (IGT) from over 1100 languages, automatically extracted from nearly 3000 documents crawled from the Web. A crucial step in building ODIN is identifying the languages of extracted IGT, a challenging task due to the large number of languages and the lack of training data. We demonstrate that a coreference approach to the language ID task significantly outperforms existing algorithms as it provides an elegant solution to the unseen language problem. We also discuss several issues that make automated Language ID and the maintenance of ODIN very difficult.

Neural codes for somatosensory two-point discrimination in inferior parietal lobule: An fMRI study

This is the first functional magnetic resonance imaging (fMRI) study to investigate the hemodynamic response related to somatosensory spatial discrimination, so-called two-point discrimination. During scanning, we examined two discrimination tasks using four types of electrical stimuli applied to one or two points with strong or weak intensity on the right and left forearm, respectively. In the two-point discrimination task (TPD), subjects reported whether they thought the stimulus was applied to one point or two. In the intensity discrimination task (ID), subjects were required to judge whether the stimulus was strong or weak. In each task, they pressed a button to report their choice. Comparing TPD with the control, we found activated regions in the inferior parietal lobule (IPL) around the supramarginal gyrus (SMG) (Brodmann’s area 40) and anterior cingulate cortex (ACC). These areas were significantly activated irrespective of the forearm stimulated. Comparing ID with the control, there were no significantly activated regions. By comparing the TPD and ID, we identified that the left IPL was significantly activated, specifically in TPD, irrespective of the forearm stimulated. In contrast, there were no significantly activated regions in the ID task. Therefore, the left IPL is considered to play an important role in two-point discrimination.

Perception of a non-native speech contrast: Voiced and voiceless stops as perceived by Tamil speakers

The effect of linguistic experience plays a significant role in how speech sounds are perceived. The findings of many studies imply that the perception of non-native contrasts depends on their status in the native language of the listener. Tamil is a language with a single voicing category. All stop consonants in Tamil are phonemically voiceless, though allophonic voicing has been observed in spoken Tamil. The present study examined how native Tamil speakers and English controls perceived voiced and voiceless bilabial, alveolar, and velar stops in English. Voice onset time (VOT) was manipulated for editing of naturally produced stimuli with increasingly longer continuum. Perceptual data was collected from 16 Tamil and 16 English speakers. Experiment 1 was an AX task in which subjects responded same or different to 162 pairs of stimuli. Experiment 2 was a forced choice ID task in which subjects identified 99 individually presented stimuli as pa, ta, ka or ba, da, ga. Experiments show statistically significant differences between Tamil and English speakers in their perception of English stop consonants. Results of the study imply that the allophonic status of voiced stops in Tamil does not aid the Tamil speakers in perceiving phonemically voiced stops in English.

Quantitative EEG patterns during nose versus mouth inhalation of filtered room air in young adults with and without self-reported chemical odor intolerances

Individuals who report illness (e.g. nausea, headache) from common chemical odors tend to report CNS symptoms suggestive of olfactory-limbic system involvement. This study compared the resting quantitative electroencephalographic (qEEG) patterns of young adult college students reporting subjectively elevated chemical odor intolerance ratings (HICI) with those of controls reporting little or no odor intolerance (LOCI). Each group was subdivided into those with higher (HIDEP) vs. lower (LODEP) ratings of concomitant depression. Nineteen channels of EEG were recorded during a single session over four separate rest periods, respectively, following baseline, cognitive, chemical exposure and olfactory identification tests. Each recording involved two 30-s, eyes-closed, filtered room air breathing conditions: (1) nose inhalation followed by mouth exhalation and (2) mouth inhalation followed by mouth exhalation. HICI showed significantly less beta 1 (β1) over the temporal–central region during nose than during mouth inhalation. Over some temporal and central leads, task, DEP and CI interacted to influence β1 as well. For theta (θ), CI differences emerged during nose inhalation after the cognitive task at Cz, after chemical exposures at C3, Cz and C4 and after the olfactory ID task at C4. CI differences emerged during mouth breathing after the olfactory ID task at Cz, C4 and T4. The T5–T6 coronal array showed significant CI differences after chemical exposures during nose breathing and during mouth breathing after the cognitive and olfactory ID tasks. The θ findings in the HICI may be related to reports of disturbed attention in CI.