A group of researchers from the Allen Institute of Artificial Intelligence has proposed the Aristo challenge that requires answering science questions. The goal of the challenge is to aid in the development of machines that can understand natural language, use knowledge and reason. In this work, we take a subset of those questions, namely the questions from the chapters of food web. We model a consequence operator for the food webs that given a food web and a perturbation to some of the populations aims to compute possible effects on the other populations in the food web. We then use this operator to answers questions of the kind, ‘Explain why the population of rabbits might decrease if the population of mice decreased.’ or ‘Explain why the population of rabbits might change if the population of mice decreased.’ Unlike the previous works which deal with only direct predator-prey situations, here we aim to characterize the effect(s) even when the two populations in the question are indirectly related. Developing intelligent agents that can understand natural language, reason and use commonsense knowledge has been one of the long term goals of AI. To track the progress towards this goal, several question answering challenges have been proposed. The Winograd Schema Challenge (Levesque, Davis, and Morgenstern 2012), Allen Institute of Artificial Intelligence (AI2)’s flagship project Aristo (Clark et al. 2018), Natural Language Inference (Bowman et al. 2015), and the Stanford Question Answering Dataset, SQuAD (Rajpurkar et al. 2016) are all examples of this. Among these challenges the science question answering challenge, Aristo from AI2 is of particular interest to the community of Knowledge Representation and Reasoning (KR) for several reasons. First, answering science questions requires a wide variety of reasoning skills including qualitative, quantitative and counter factual reasoning. Second, majority of the knowledge needed to answer the questions can be found in the science textbooks. The later is an important property as it allows one to solely focus on the task of knowledge representation and reasoning without worrying about the challenging task of knowledge extraction. In this work, we focus on answering a particular type science questions, namely the food web questions. Table 1 shows two such examples. Copyright c © 2018, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. Question Type 1: Explain how a perturbation leads to a certain change in a population. The first example Q1 in Table 1 is an example of a type 1 question which is taken from the 8 grade New York Science Regents exam. This type of questions looks for an explanation to an outcome given a perturbation to a food web. For example, the question Q1 in Table 1 seeks an explanation for the decreasing rabbit population. A possible answer to the question could be that “the population of rabbits decreases because the snakes would eat more rabbits”. Note that there is also a reason for the rabbit population to increase as the competition for grass is less after the decline in mouse population. In this conflicting scenario, it is difficult to eliminate any of these two possibilities due to the absence of proper knowledge. Thus if the question asks to explain the possible effect on the rabbit population one has to list down all the possibilities. Question 2 from table 1 makes this expectation clear. Question Type 2: Enumerate and justify all the changes in a population due to a given perturbation. Another type of question that is often asked in New York Science 8th grade exams is about explaining the possible effects on a population due to a perturbation in other populations. Question 2 in Table 1 shows an example of this kind. Unlike to the type 1 questions, this type of questions looks for all possible effects on a population. An answer to this type of questions includes exploring the possible effects and one or more justification for each scenario. For e.g. according to the solution manual provided in the Regents portal, an answer to the Question 2 in Table 1 could be as follows: • The squid population could decrease because there would be fewer cod for them to eat. • The squid population would increase because there would be more small animals and krill for them to eat. • There would be less food for the elephant seals, so they would eat more squid and the squid population would decrease • There would be less food for the penguins, so they would eat more squid, so the squid population would decrease. Proceedings of the Sixteenth International Conference on Principles of Knowledge Representation and Reasoning (KR 2018)
[1]
Y. Takeuchi,et al.
Existence and bifurcation of stable equilibrium in two-prey, one-predator communities
,
1983
.
[2]
Daniel S. Weld.
Comparative Analysis
,
1987,
IJCAI.
[3]
Kenneth D. Forbus.
Qualitative Process Theory
,
1984,
Artificial Intelligence.
[4]
Daniel S. Weld.
Theories of comparative analysis
,
1990
.
[5]
B. Bredeweg,et al.
Qualitative Reasoning about Food Webs: Exploring Alternative Representations
,
2004
.
[6]
Hector J. Levesque,et al.
The Winograd Schema Challenge
,
2011,
AAAI Spring Symposium: Logical Formalizations of Commonsense Reasoning.
[7]
Christopher Potts,et al.
A large annotated corpus for learning natural language inference
,
2015,
EMNLP.
[8]
Jian Zhang,et al.
SQuAD: 100,000+ Questions for Machine Comprehension of Text
,
2016,
EMNLP.
[9]
Oren Etzioni,et al.
Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
,
2018,
ArXiv.