当前位置：网站首页>[AI vision · quick review of NLP natural language processing papers today, issue 29] Mon, 14 Feb 2022

[AI vision · quick review of NLP natural language processing papers today, issue 29] Mon, 14 Feb 2022

2022-04-23 03:45:00 【hitrjj】

AI View · Today, CS.NLP Natural language processing papers
Mon, 14 Feb 2022
Totally 16 papers
Quick view of last issue For more highlights, please move to the home page

Insert picture description here

Daily Computation and Language Papers

Evaluating MT Systems: A Theoretical Framework
Authors Rajeev Sangal
This paper outlines a theoretical framework , Using this framework, different automatic indicators can be designed to evaluate machine translation systems . It introduces em Recognize the concept of ease , It depends on the em Adequacy and em The lack of fluency . therefore , Cognitive difficulty becomes the main parameter to measure , Not comprehensibility .

White-Box Attacks on Hate-speech BERT Classifiers in German with Explicit and Implicit Character Level Defense
Authors Shahrukh Khan, Mahnoor Shahid, Navdeeppal Singh
In this work , We evaluated training on the German hate speech dataset BERT The robustness of the model . We also complement our assessment with two novel white box character and word level attacks , Thus increasing the range of available attacks .

Using Random Perturbations to Mitigate Adversarial Attacks on Sentiment Analysis Models
Authors Abigail Swenor, Jugal Kalita
Deep learning attacks on models are often difficult to identify , So it's hard to guard against . This problem is exacerbated by the use of public data sets that are usually not manually checked before use . In this paper , By using random perturbation during the test （ Make spelling corrections if necessary 、 Random synonym replacement or simple deletion of words ） To provide a solution to this vulnerability . These perturbations are applied to random words in random sentences , To protect NLP Models are protected from adversarial attacks . Our random disturbance defense and increased randomness defense methods successfully restore the attacked model to the similar accuracy of the model before the attack . The original accuracy of the model used in this work is 80 Used for emotion classification . After being attacked , The accuracy dropped to 0 To 44 Accuracy between .

Dual Task Framework for Debiasing Persona-grounded Dialogue Dataset
Authors Minju Kim, Beong woo Kwak, Youngwook Kim, Hong in Lee, Seung won Hwang, Jinyoung Yeo
This paper introduces a simple and effective data centric method , Tasks for improving role conditional dialog agents . The previous model centric approach undoubtedly relied on the original crowdsourcing benchmark data set , for example Persona Chat. by comparison , Our goal is to fix annotation artifacts in the benchmark , This orthogonality applies to any conversation model . say concretely , By using the original binary structure of two tasks , Enhance relevant roles based on each other's predicted dialog responses and roles to improve the dialog dataset agent .

Including Facial Expressions in Contextual Embeddings for Sign Language Generation
Authors Carla Viegas, Mert nan, Lorna Quandt, Malihe Alikhani
The most advanced sign language generation framework lacks expressiveness and naturalness , This is an emotion that focuses only on manual symbols and ignores facial expressions 、 The result of grammatical and semantic functions . The purpose of this work is to enhance the semantic representation of sign language based on facial expression . We studied the text 、 Modeling the relationship between gloss and facial expression and its impact on the performance of logo generation system . especially , We put forward a kind of Dual Encoder Transformer, It can generate manual symbols and facial expressions by capturing the similarities and differences between text and symbol gloss annotations . We first use facial action units in sign language generation , Thus, the function of facial muscle activity in expressing sign language intensity is considered .

Multi-Modal Knowledge Graph Construction and Application: A Survey
Authors Xiangru Zhu, Zhixu Li, Xiaodan Wang, Xueyao Jiang, Penglei Sun, Xuwu Wang, Yanghua Xiao, Nicholas Jing Yuan
In recent years , Knowledge engineering characterized by the rapid growth of knowledge map is rising again . However , Most of the existing knowledge maps are expressed by pure symbols , This undermines the machine's ability to understand the real world . The multimodality of knowledge map is an inevitable key step to realize man-machine intelligence . The result of this effort is a multimodal knowledge map MMKGs. In this project built from text and images MMKG In the survey , We first gave MMKG The definition of , Then there is a preliminary introduction to multimodal tasks and technologies . then , We systematically reviewed MMKG The challenges of building and applying 、 Progress and opportunities , The advantages and disadvantages of different solutions are analyzed in detail .

Chinese Abs From Machine Translation

Papers from arxiv.org

For more highlights, please move to the home page

pic from pexels.com

版权声明
本文为[hitrjj]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/04/202204220600582993.html