Causal Reasoning in NLP近年来,深度学习在自然语言处理领域获得了巨大的成功。但是,质疑声也一直不绝于耳,尤其是关于深度学习容易学习到语言数据集上的伪关系(spurious relation)的问题一直没有得到解决。因果推断理论告诉我们,这是由于混杂因子(confounding)造成的。然而,将因果推断方法应用到自然语言处理目前仍然面临着一些困难:什么是自然语言当中的随机变量?如何从表示中找出混杂因子?如何让学习结果更加稳定,避免受训练集中的伪关系影响?其中最大的困难,在于如何定义自然语言中的因果关系。
Chi Zhang, Baoxiong Jia, Mark Edmonds, Song-Chun Zhu, Yixin Zhu
We present a new dataset for Abstract Causal REasoning, aiming to measure and improve causal induction in visual reasoning systems
Cited by0BibtexViews22
0
0
Vasili Ramanishka,Yi-Ting Chen, Teruhisa Misu,Kate Saenko
CVPR, (2018): 7699-7707
We introduce the Honda Research Institute Driving Dataset, which aims to stimulate the community to propose novel algorithms to capture the driver behavior
Cited by94BibtexViews77DOI
0
0
Jinyoung Yeo, Geungyu Wang,Hyunsouk Cho, Seungtaek Choi,Seung-won Hwang
THIRTY-SECOND AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTIETH INNOVATIVE APPLICATIONS OF ARTI..., pp.2021-2028, (2018)
That the cross-lingual knowledge transfer between native English and machinetranslated English can be an effective technique in growing the causal reasoning ability on resource-poor languages without any target language corpus
Cited by5BibtexViews63
0
0
AAAI, pp.1180-1185, (2011)
We have shown that the personal stories that people write in their weblogs are a good source of commonsense causal information
Cited by42BibtexViews19
0
0
Cognitive Science, no. 5 (2003): 709-748
This article has proposed that the causal relations that link category features are represented in terms of asymmetric and probabilistic causal mechanisms, and that category membership is evaluated on the basis of whether objects were likely to have been generated by those mechan...
Cited by97BibtexViews21DOI
0
0
IJCAI, pp.131-136, (2001)
This paper presents an extended system EPDL of propositional dynamic logic by allowing a proposition as a modality for representing and specifying direct and indirect effects of actions in a unified logical structure. A set of causal logics based on the framework are proposed to ...
Cited by43BibtexViews15
0
0
(2000)
Consider that it has further been discovered that it's not just the children's and the parents' birth dates that are correlated with the children's careers. No matter how many more complex covariation patterns we discover, most well-educated people would dismiss these findings be...
Cited by175BibtexViews14
0
0
Current biology : CB, no. 12 (2000): 723-726
In Experiment 1, S.A. received a number of tests, including a causal knowledge test, the picture arrangement test from the Wechsler adult intelligence scale, and a theory of mind’ task consisting of ‘changed container’ false-belief measures
Cited by89BibtexViews13DOI
0
0
uncertainty in artificial intelligence, no. 1 (1998): 317-337
A more general approach to modeling causality would allow there to be more than one value of X once we have set all the other variables. This would be appropriate if we model things at a somewhat coarser level of granularity, where the values of all the variables other than X do ...
Cited by149BibtexViews16DOI
0
0
JOURNAL OF MEMORY AND LANGUAGE, no. 3 (1988): 235-244
Several important findings emerge from this research
Cited by163BibtexViews24DOI
0
0