Reasoning-Driven Question-Answering for Natural Language Understanding Daniel Khashabi University of Pennsylvania, [email protected]

Total Page:16

File Type:pdf, Size:1020Kb

Reasoning-Driven Question-Answering for Natural Language Understanding Daniel Khashabi University of Pennsylvania, Danielkh@Cis.Upenn.Edu University of Pennsylvania ScholarlyCommons Publicly Accessible Penn Dissertations 2019 Reasoning-Driven Question-Answering For Natural Language Understanding Daniel Khashabi University of Pennsylvania, [email protected] Follow this and additional works at: https://repository.upenn.edu/edissertations Part of the Artificial Intelligence and Robotics Commons Recommended Citation Khashabi, Daniel, "Reasoning-Driven Question-Answering For Natural Language Understanding" (2019). Publicly Accessible Penn Dissertations. 3271. https://repository.upenn.edu/edissertations/3271 This paper is posted at ScholarlyCommons. https://repository.upenn.edu/edissertations/3271 For more information, please contact [email protected]. Reasoning-Driven Question-Answering For Natural Language Understanding Abstract Natural language understanding (NLU) of text is a fundamental challenge in AI, and it has received significant attention throughout the history of NLP research. This primary goal has been studied under different tasks, such as Question Answering (QA) and Textual Entailment (TE). In this thesis, we investigate the NLU problem through the QA task and focus on the aspects that make it a challenge for the current state-of-the-art technology. This thesis is organized into three main parts: In the first part, we explore multiple formalisms to improve existing machine comprehension systems. We propose a formulation for abductive reasoning in natural language and show its effectiveness, especially in domains with limited training data. Additionally, to help reasoning systems cope with irrelevant or redundant information, we create a supervised approach to learn and detect the essential terms in questions. In the second part, we propose two new challenge datasets. In particular, we create two datasets of natural language questions where (i) the first one requires reasoning over multiple sentences; (ii) the second one requires temporal common sense reasoning. We hope that the two proposed datasets will motivate the field to address more complex problems. In the final part, we present the first formal framework for multi-step reasoning algorithms, in the presence of a few important properties of language use, such as incompleteness, ambiguity, etc. We apply this framework to prove fundamental limitations for reasoning algorithms. These theoretical results provide extra intuition into the existing empirical evidence in the field. Degree Type Dissertation Degree Name Doctor of Philosophy (PhD) Graduate Group Computer and Information Science First Advisor Dan Roth Keywords Language Understanding, Natural Language Processing Subject Categories Artificial Intelligence and Robotics | Computer Sciences This dissertation is available at ScholarlyCommons: https://repository.upenn.edu/edissertations/3271 REASONING-DRIVEN QUESTION-ANSWERING FOR NATURAL LANGUAGE UNDERSTANDING Daniel Khashabi A DISSERTATION in Computer and Information Sciences Presented to the Faculties of the University of Pennsylvania in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy 2019 Supervisor of Dissertation Dan Roth, Professor of Computer and Information Science Graduate Group Chairperson Rajeev Alur, Professor of Computer and Information Science Dissertation Committee Dan Roth, Professor, Computer and Information Science, University of Pennsylvania Mitch Marcus, Professor of Computer and Information Science, University of Pennsylvania Zachary Ives, Professor of Computer and Information Sciences, University of Pennsylvania Chris Callison-Burch, Associate Professor of Computer Science, University of Pennsylvania Ashish Sabharwal, Senior Research Scientist, Allen Institute for Artificial Intelligence REASONING-DRIVEN QUESTION-ANSWERING FOR NATURAL LANGUAGE UNDERSTANDING c COPYRIGHT 2019 Daniel Khashabi This work is licensed under the Creative Commons Attribution NonCommercial-ShareAlike 3.0 License To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-sa/3.0/ Dedicated to the loving memory of my gramma, An'nah Your patience and kindness will forever stay with me. iii ACKNOWLEDGEMENT I feel incredibly lucky to have Dan Roth as my advisor. I am grateful to Dan for trusting me, especially when I had only a basic understanding of many key challenges in natural language. It took me a while to catch up with what is important in the field and be able to communicate the challenges effectively. During these years, Dan's vision has always been the guiding principle to many of my works. His insistence on focusing on the long-term progress, rather than \easy" wins, shaped the foundation of many of the ideas I pursued. This perspective pushed me to think differently than the popular trends. It has been a genuine privilege to work together. I want to thank my thesis committee at UPenn, Mitch Marcus, Zach Ives and Chris Callison- Burch for being a constant source of invaluable feedback and guidance. Additionally, I would like to thanks the many professors who have touched parts of my thinking: Jerry DeJong, for encouraging me read the classic literature; Chandra Chekuri and Avrim Blum, for their emphasis on intuition, rather than details; and my undergraduate advisor Hamid Sheikhzadeh Nadjar, for encouraging me to work on important problems. A huge thank you to the Allen Institute for Artificial Intelligence (AI2) for much support during my PhD studies. Any time I needed any resources (computing resources, crowdsourc- ing credits, engineering help, etc), without any hesitation, AI2 has provided me what was needed. Special thanks to Ashish Sabhwaral and Tushar Khot for being a constant source of wisdom and guidance, and investing lots of time and effort. They both have always been present to listen to my random thoughts, almost on a weekly basis. I am grateful to other members of AI2 for their help throughout my projects: Oren Etzioni, Peter Clark, Oyvind Tafjord, Peter Turney, Ingmar Ellenberger, Dirk Groeneveld, Michael Schmitz, Chandra Bhagavatula and Scott Yih. Moreover, I would like to remember Paul Allen (1953-2018): his vision and constant generous support has tremendously changed our field (and my life, in particular). iv My collaborators, especially past and present CogComp members, have been major con- tributors and influencers throughout my works. I would like to thank Mark Sammons, Vivek Srikumar, Christos Christodoulopoulos, Erfan Sadeqi Azer, Snigdha Chaturvedi, Kent Quanrud, Amirhossein Taghvaei, Chen-Tse Tsai, and many other CogComp members. Furthermore, I thank Eric Horn and Jennifer Sheffield for their tremendous contributions to many of my write-ups. And thank you to all the friends I have made at Penn, UIUC, and elsewhere, for all the happiness you've brought me. Thanks to Whitney, for sharing many happy and sad moments with me, and for helping me become a better version of myself. Last, but never least, my family, for their unconditional sacrifice and support. I wouldn't have been able to go this far without you. v ABSTRACT REASONING-DRIVEN QUESTION-ANSWERING FOR NATURAL LANGUAGE UNDERSTANDING Daniel Khashabi Dan Roth Natural language understanding (NLU) of text is a fundamental challenge in AI, and it has received significant attention throughout the history of NLP research. This primary goal has been studied under different tasks, such as Question Answering (QA) and Textual Entailment (TE). In this thesis, we investigate the NLU problem through the QA task and focus on the aspects that make it a challenge for the current state-of-the-art technology. This thesis is organized into three main parts: In the first part, we explore multiple formalisms to improve existing machine comprehension systems. We propose a formulation for abductive reasoning in natural language and show its effectiveness, especially in domains with limited training data. Additionally, to help reasoning systems cope with irrelevant or redundant information, we create a supervised approach to learn and detect the essential terms in questions. In the second part, we propose two new challenge datasets. In particular, we create two datasets of natural language questions where (i) the first one requires reasoning over multiple sentences; (ii) the second one requires temporal common sense reasoning. We hope that the two proposed datasets will motivate the field to address more complex problems. In the final part, we present the first formal framework for multi-step reasoning algorithms, in the presence of a few important properties of language use, such as incompleteness, ambiguity, etc. We apply this framework to prove fundamental limitations for reasoning algorithms. These theoretical results provide extra intuition into the existing empirical evidence in the field. vi TABLE OF CONTENTS ACKNOWLEDGEMENT . iv ABSTRACT . vi LIST OF TABLES . xv LIST OF ILLUSTRATIONS . xx PUBLICATION NOTES . xxi CHAPTER 1 : Introduction . 1 1.1 Motivation . 1 1.2 Challenges along the way to NLU . 1 1.3 Measuring the progress towards NLU via Question Answering . 4 1.4 Thesis outline . 6 CHAPTER 2 : Background and Related Work . 8 2.1 Overview . 8 2.2 Terminology . 8 2.3 Measuring the progress towards NLU . 10 2.3.1 Measurement protocols . 10 2.4 Knowledge Representation and Abstraction for NLU . 14 2.4.1 Early Works: \Neats vs Scruffies”1 ................... 14 2.4.2 Connectionism . 16 2.4.3 Unsupervised representations . 16 2.4.4 Grounding of meanings . 17 1Terms originally made by Roger Schank to characterize two different camps: the first group that repre- sented commonsense knowledge
Recommended publications
  • Handling Ambiguity Problems of Natural Language Interface For
    IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 3, No 3, May 2012 ISSN (Online): 1694-0814 www.IJCSI.org 17 Handling Ambiguity Problems of Natural Language InterfaceInterface for QuesQuesQuestQuestttionionionion Answering Omar Al-Harbi1, Shaidah Jusoh2, Norita Norwawi3 1 Faculty of Science and Technology, Islamic Science University of Malaysia, Malaysia 2 Faculty of Science & Information Technology, Zarqa University, Zarqa, Jordan 3 Faculty of Science and Technology, Islamic Science University of Malaysia, Malaysia documents processing, and answer processing [9], [10], Abstract and [11]. The Natural language question (NLQ) processing module is considered a fundamental component in the natural language In QA, a NLQ is the primary source through which a interface of a Question Answering (QA) system, and its quality search process is directed for answers. Therefore, an impacts the performance of the overall QA system. The most accurate analysis to the NLQ is required. One of the most difficult problem in developing a QA system is so hard to find an exact answer to the NLQ. One of the most challenging difficult problems in developing a QA system is that it is problems in returning answers is how to resolve lexical so hard to find an answer to a NLQ [22]. The main semantic ambiguity in the NLQs. Lexical semantic ambiguity reason is most QA systems ignore the semantic issue in may occurs when a user's NLQ contains words that have more the NLQ analysis [12], [2], [14], and [15]. To achieve a than one meaning. As a result, QA system performance can be better performance, the semantic information contained negatively affected by these ambiguous words.
    [Show full text]
  • Why Machines Don't
    KI - Künstliche Intelligenz https://doi.org/10.1007/s13218-019-00599-w SURVEY Why Machines Don’t (yet) Reason Like People Sangeet Khemlani1 · P. N. Johnson‑Laird2,3 Received: 15 February 2019 / Accepted: 19 June 2019 © This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply 2019 Abstract AI has never come to grips with how human beings reason in daily life. Many automated theorem-proving technologies exist, but they cannot serve as a foundation for automated reasoning systems. In this paper, we trace their limitations back to two historical developments in AI: the motivation to establish automated theorem-provers for systems of mathematical logic, and the formulation of nonmonotonic systems of reasoning. We then describe why human reasoning cannot be simulated by current machine reasoning or deep learning methodologies. People can generate inferences on their own instead of just evaluating them. They use strategies and fallible shortcuts when they reason. The discovery of an inconsistency does not result in an explosion of inferences—instead, it often prompts reasoners to abandon a premise. And the connectives they use in natural language have diferent meanings than those in classical logic. Only recently have cognitive scientists begun to implement automated reasoning systems that refect these human patterns of reasoning. A key constraint of these recent implementations is that they compute, not proofs or truth values, but possibilities. Keywords Reasoning · Mental models · Cognitive models 1 Introduction problems for mathematicians). Their other uses include the verifcation of computer programs and the design of com- The great commercial success of deep learning has pushed puter chips.
    [Show full text]
  • A Survey of Top-Level Ontologies to Inform the Ontological Choices for a Foundation Data Model
    A survey of Top-Level Ontologies To inform the ontological choices for a Foundation Data Model Version 1 Contents 1 Introduction and Purpose 3 F.13 FrameNet 92 2 Approach and contents 4 F.14 GFO – General Formal Ontology 94 2.1 Collect candidate top-level ontologies 4 F.15 gist 95 2.2 Develop assessment framework 4 F.16 HQDM – High Quality Data Models 97 2.3 Assessment of candidate top-level ontologies F.17 IDEAS – International Defence Enterprise against the framework 5 Architecture Specification 99 2.4 Terminological note 5 F.18 IEC 62541 100 3 Assessment framework – development basis 6 F.19 IEC 63088 100 3.1 General ontological requirements 6 F.20 ISO 12006-3 101 3.2 Overarching ontological architecture F.21 ISO 15926-2 102 framework 8 F.22 KKO: KBpedia Knowledge Ontology 103 4 Ontological commitment overview 11 F.23 KR Ontology – Knowledge Representation 4.1 General choices 11 Ontology 105 4.2 Formal structure – horizontal and vertical 14 F.24 MarineTLO: A Top-Level 4.3 Universal commitments 33 Ontology for the Marine Domain 106 5 Assessment Framework Results 37 F. 25 MIMOSA CCOM – (Common Conceptual 5.1 General choices 37 Object Model) 108 5.2 Formal structure: vertical aspects 38 F.26 OWL – Web Ontology Language 110 5.3 Formal structure: horizontal aspects 42 F.27 ProtOn – PROTo ONtology 111 5.4 Universal commitments 44 F.28 Schema.org 112 6 Summary 46 F.29 SENSUS 113 Appendix A F.30 SKOS 113 Pathway requirements for a Foundation Data F.31 SUMO 115 Model 48 F.32 TMRM/TMDM – Topic Map Reference/Data Appendix B Models 116 ISO IEC 21838-1:2019
    [Show full text]
  • Artificial Intelligence
    BROAD AI now and later Michael Witbrock, PhD University of Auckland Broad AI Lab @witbrock Aristotle (384–322 BCE) Organon ROOTS OF AI ROOTS OF AI Santiago Ramón y Cajal (1852 -1934) Cerebral Cortex WHAT’S AI • OLD definition: AI is everything we don’t yet know how program • Now some things that people can’t do: • unique capabilities (e.g. Style transfer) • superhuman performance (some areas of speech, vision, games, some QA, etc) • Current AI Systems can be divided by their kind of capability: • Skilled (Image recognition, Game Playing (Chess, Atari, Go, DoTA), Driving) • Attentive (Trading: Aidyia; Senior Care: CareMedia, Driving) • Knowledgeable, (Google Now, Siri, Watson, Cortana) • High IQ (Cyc, Soar, Wolfram Alpha) GOFAI • Thought is symbol manipulation • Large numbers of precisely defined symbols (terms) • Based on mathematical logic (implies (and (isa ?INST1 LegalAgreement) (agreeingAgents ?INST1 ?INST2)) (isa ?INST2 LegalAgent)) • Problems solved by searching for transformations of symbolic representations that lead to a solution Slow Development Thinking Quickly Thinking Slowly (System I) (System II) Human Superpower c.f. other Done well by animals and people animals Massively parallel algorithms Serial and slow Done poorly until now by computers Done poorly by most people Not impressive to ordinary people Impressive (prizes, high pay) "Sir, an animal’s reasoning is like a dog's walking on his hind legs. It is not done well; but you are surprised to find it done at all.“ - apologies to Samuel Johnson Achieved on computers by high- Fundamental design principle of power, low density, slow computers simulation of vastly different Computer superpower c.f. neural hardware human Recurrent Deep Learning & Deep Reasoning MACHINE LEARNING • Meaning is implicit in the data • Thought is the transformation of learned representations http://karpathy.github.io/2015/05/21/rnn- effectiveness/ .
    [Show full text]
  • Knowledge Graphs on the Web – an Overview Arxiv:2003.00719V3 [Cs
    January 2020 Knowledge Graphs on the Web – an Overview Nicolas HEIST, Sven HERTLING, Daniel RINGLER, and Heiko PAULHEIM Data and Web Science Group, University of Mannheim, Germany Abstract. Knowledge Graphs are an emerging form of knowledge representation. While Google coined the term Knowledge Graph first and promoted it as a means to improve their search results, they are used in many applications today. In a knowl- edge graph, entities in the real world and/or a business domain (e.g., people, places, or events) are represented as nodes, which are connected by edges representing the relations between those entities. While companies such as Google, Microsoft, and Facebook have their own, non-public knowledge graphs, there is also a larger body of publicly available knowledge graphs, such as DBpedia or Wikidata. In this chap- ter, we provide an overview and comparison of those publicly available knowledge graphs, and give insights into their contents, size, coverage, and overlap. Keywords. Knowledge Graph, Linked Data, Semantic Web, Profiling 1. Introduction Knowledge Graphs are increasingly used as means to represent knowledge. Due to their versatile means of representation, they can be used to integrate different heterogeneous data sources, both within as well as across organizations. [8,9] Besides such domain-specific knowledge graphs which are typically developed for specific domains and/or use cases, there are also public, cross-domain knowledge graphs encoding common knowledge, such as DBpedia, Wikidata, or YAGO. [33] Such knowl- edge graphs may be used, e.g., for automatically enriching data with background knowl- arXiv:2003.00719v3 [cs.AI] 12 Mar 2020 edge to be used in knowledge-intensive downstream applications.
    [Show full text]
  • The Question Answering System Using NLP and AI
    International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016 ISSN 2229-5518 55 The Question Answering System Using NLP and AI Shivani Singh Nishtha Das Rachel Michael Dr. Poonam Tanwar Student, SCS, Student, SCS, Student, SCS, Associate Prof. , SCS Lingaya’s Lingaya’s Lingaya’s Lingaya’s University, University,Faridabad University,Faridabad University,Faridabad Faridabad Abstract: (ii) QA response with specific answer to a specific question The Paper aims at an intelligent learning system that will take instead of a list of documents. a text file as an input and gain knowledge from the given text. Thus using this knowledge our system will try to answer questions queried to it by the user. The main goal of the 1.1. APPROCHES in QA Question Answering system (QAS) is to encourage research into systems that return answers because ample number of There are three major approaches to Question Answering users prefer direct answers, and bring benefits of large-scale Systems: Linguistic Approach, Statistical Approach and evaluation to QA task. Pattern Matching Approach. Keywords: A. Linguistic Approach Question Answering System (QAS), Artificial Intelligence This approach understands natural language texts, linguistic (AI), Natural Language Processing (NLP) techniques such as tokenization, POS tagging and parsing.[1] These are applied to reconstruct questions into a correct 1. INTRODUCTION query that extracts the relevant answers from a structured IJSERdatabase. The questions handled by this approach are of Factoid type and have a deep semantic understanding. Question Answering (QA) is a research area that combines research from different fields, with a common subject, which B.
    [Show full text]
  • Using Linked Data for Semi-Automatic Guesstimation
    Using Linked Data for Semi-Automatic Guesstimation Jonathan A. Abourbih and Alan Bundy and Fiona McNeill∗ [email protected], [email protected], [email protected] University of Edinburgh, School of Informatics 10 Crichton Street, Edinburgh, EH8 9AB, United Kingdom Abstract and Semantic Web systems. Next, we outline the process of GORT is a system that combines Linked Data from across guesstimation. Then, we describe the organisation and im- several Semantic Web data sources to solve guesstimation plementation of GORT. Finally, we close with an evaluation problems, with user assistance. The system uses customised of the system’s performance and adaptability, and compare inference rules over the relationships in the OpenCyc ontol- it to several other related systems. We also conclude with a ogy, combined with data from DBPedia, to reason and per- brief section on future work. form its calculations. The system is extensible with new Linked Data, as it becomes available, and is capable of an- Literature Survey swering a small range of guesstimation questions. Combining facts to answer a user query is a mature field. The DEDUCOM system (Slagle 1965) was one of the ear- Introduction liest systems to perform deductive query answering. DE- The true power of the Semantic Web will come from com- DUCOM applies procedural knowledge to a set of facts in bining information from heterogeneous data sources to form a knowledge base to answer user queries, and a user can new knowledge. A system that is capable of deducing an an- also supplement the knowledge base with further facts.
    [Show full text]
  • Why Has AI Failed? and How Can It Succeed?
    Why Has AI Failed? And How Can It Succeed? John F. Sowa VivoMind Research, LLC 10 May 2015 Extended version of slides for MICAI'14 ProblemsProblems andand ChallengesChallenges Early hopes for artificial intelligence have not been realized. Language understanding is more difficult than anyone thought. A three-year-old child is better able to learn, understand, and generate language than any current computer system. Tasks that are easy for many animals are impossible for the latest and greatest robots. Questions: ● Have we been using the right theories, tools, and techniques? ● Why haven’t these tools worked as well as we had hoped? ● What other methods might be more promising? ● What can research in neuroscience and psycholinguistics tell us? ● Can it suggest better ways of designing intelligent systems? 2 Early Days of Artificial Intelligence 1960: Hao Wang’s theorem prover took 7 minutes to prove all 378 FOL theorems of Principia Mathematica on an IBM 704 – much faster than two brilliant logicians, Whitehead and Russell. 1960: Emile Delavenay, in a book on machine translation: “While a great deal remains to be done, it can be stated without hesitation that the essential has already been accomplished.” 1965: Irving John Good, in speculations on the future of AI: “It is more probable than not that, within the twentieth century, an ultraintelligent machine will be built and that it will be the last invention that man need make.” 1968: Marvin Minsky, technical adviser for the movie 2001: “The HAL 9000 is a conservative estimate of the level of artificial intelligence in 2001.” 3 The Ultimate Understanding Engine Sentences uttered by a child named Laura before the age of 3.
    [Show full text]
  • An Overview of Automated Reasoning
    An overview of automated reasoning John Harrison Intel Corporation 3rd February 2014 (10:30{11:30) Talk overview I What is automated reasoning? I Early history and taxonomy I Automation, its scope and limits I Interactive theorem proving I Hobbes (1651): \Reason . is nothing but reckoning (that is, adding and subtracting) of the consequences of general names agreed upon, for the marking and signifying of our thoughts." I Leibniz (1685) \When there are disputes among persons, we can simply say: Let us calculate [calculemus], without further ado, to see who is right." Nowadays, by `automatic and algorithmic' we mean `using a computer program'. What is automated reasoning? Attempting to perform logical reasoning in an automatic and algorithmic way. An old dream! I Leibniz (1685) \When there are disputes among persons, we can simply say: Let us calculate [calculemus], without further ado, to see who is right." Nowadays, by `automatic and algorithmic' we mean `using a computer program'. What is automated reasoning? Attempting to perform logical reasoning in an automatic and algorithmic way. An old dream! I Hobbes (1651): \Reason . is nothing but reckoning (that is, adding and subtracting) of the consequences of general names agreed upon, for the marking and signifying of our thoughts." Nowadays, by `automatic and algorithmic' we mean `using a computer program'. What is automated reasoning? Attempting to perform logical reasoning in an automatic and algorithmic way. An old dream! I Hobbes (1651): \Reason . is nothing but reckoning (that is, adding and subtracting) of the consequences of general names agreed upon, for the marking and signifying of our thoughts." I Leibniz (1685) \When there are disputes among persons, we can simply say: Let us calculate [calculemus], without further ado, to see who is right." What is automated reasoning? Attempting to perform logical reasoning in an automatic and algorithmic way.
    [Show full text]
  • Quester: a Speech-Based Question Answering Support System for Oral Presentations Reza Asadi, Ha Trinh, Harriet J
    Quester: A Speech-Based Question Answering Support System for Oral Presentations Reza Asadi, Ha Trinh, Harriet J. Fell, Timothy W. Bickmore Northeastern University Boston, USA asadi, hatrinh, fell, [email protected] ABSTRACT good support for speakers who want to deliver more Current slideware, such as PowerPoint, reinforces the extemporaneous talks in which they dynamically adapt their delivery of linear oral presentations. In settings such as presentation to input or questions from the audience, question answering sessions or review lectures, more evolving audience needs, or other contextual factors such as extemporaneous and dynamic presentations are required. varying or indeterminate presentation time, real-time An intelligent system that can automatically identify and information, or more improvisational or experimental display the slides most related to the presenter’s speech, formats. At best, current slideware only provides simple allows for more speaker flexibility in sequencing their indexing mechanisms to let speakers hunt through their presentation. We present Quester, a system that enables fast slides for material to support their dynamically evolving access to relevant presentation content during a question speech, and speakers must perform this frantic search while answering session and supports nonlinear presentations led the audience is watching and waiting. by the speaker. Given the slides’ contents and notes, the system ranks presentation slides based on semantic Perhaps the most common situations in which speakers closeness to spoken utterances, displays the most related must provide such dynamic presentations are in Question slides, and highlights the corresponding content keywords and Answer (QA) sessions at the end of their prepared in slide notes. The design of our system was informed by presentations.
    [Show full text]
  • Logic-Based Technologies for Intelligent Systems: State of the Art and Perspectives
    information Article Logic-Based Technologies for Intelligent Systems: State of the Art and Perspectives Roberta Calegari 1,* , Giovanni Ciatto 2 , Enrico Denti 3 and Andrea Omicini 2 1 Alma AI—Alma Mater Research Institute for Human-Centered Artificial Intelligence, Alma Mater Studiorum–Università di Bologna, 40121 Bologna, Italy 2 Dipartimento di Informatica–Scienza e Ingegneria (DISI), Alma Mater Studiorum–Università di Bologna, 47522 Cesena, Italy; [email protected] (G.C.); [email protected] (A.O.) 3 Dipartimento di Informatica–Scienza e Ingegneria (DISI), Alma Mater Studiorum–Università di Bologna, 40136 Bologna, Italy; [email protected] * Correspondence: [email protected] Received: 25 February 2020; Accepted: 18 March 2020; Published: 22 March 2020 Abstract: Together with the disruptive development of modern sub-symbolic approaches to artificial intelligence (AI), symbolic approaches to classical AI are re-gaining momentum, as more and more researchers exploit their potential to make AI more comprehensible, explainable, and therefore trustworthy. Since logic-based approaches lay at the core of symbolic AI, summarizing their state of the art is of paramount importance now more than ever, in order to identify trends, benefits, key features, gaps, and limitations of the techniques proposed so far, as well as to identify promising research perspectives. Along this line, this paper provides an overview of logic-based approaches and technologies by sketching their evolution and pointing out their main application areas. Future perspectives for exploitation of logic-based technologies are discussed as well, in order to identify those research fields that deserve more attention, considering the areas that already exploit logic-based approaches as well as those that are more likely to adopt logic-based approaches in the future.
    [Show full text]
  • Knowledge on the Web: Towards Robust and Scalable Harvesting of Entity-Relationship Facts
    Knowledge on the Web: Towards Robust and Scalable Harvesting of Entity-Relationship Facts Gerhard Weikum Max Planck Institute for Informatics http://www.mpi-inf.mpg.de/~weikum/ Acknowledgements 2/38 Vision: Turn Web into Knowledge Base comprehensive DB knowledge fact of human knowledge assets extraction • everything that (Semantic (Statistical Web) Web) Wikipedia knows • machine-readable communities • capturing entities, (Social Web) classes, relationships Source: DB & IR methods for knowledge discovery. Communications of the ACM 52(4), 2009 3/38 Knowledge as Enabling Technology • entity recognition & disambiguation • understanding natural language & speech • knowledge services & reasoning for semantic apps • semantic search: precise answers to advanced queries (by scientists, students, journalists, analysts, etc.) German chancellor when Angela Merkel was born? Japanese computer science institutes? Politicians who are also scientists? Enzymes that inhibit HIV? Influenza drugs for pregnant women? ... 4/38 Knowledge Search on the Web (1) Query: sushi ingredients? Results: Nori seaweed Ginger Tuna Sashimi ... Unagi http://www.google.com/squared/5/38 Knowledge Search on the Web (1) Query:Query: JapaneseJapanese computerscomputeroOputer science science ? institutes ? http://www.google.com/squared/6/38 Knowledge Search on the Web (2) Query: politicians who are also scientists ? ?x isa politician . ?x isa scientist Results: Benjamin Franklin Zbigniew Brzezinski Alan Greenspan Angela Merkel … http://www.mpi-inf.mpg.de/yago-naga/7/38 Knowledge Search on the Web (2) Query: politicians who are married to scientists ? ?x isa politician . ?x isMarriedTo ?y . ?y isa scientist Results (3): [ Adrienne Clarkson, Stephen Clarkson ], [ Raúl Castro, Vilma Espín ], [ Jeannemarie Devolites Davis, Thomas M. Davis ] http://www.mpi-inf.mpg.de/yago-naga/8/38 Knowledge Search on the Web (3) http://www-tsujii.is.s.u-tokyo.ac.jp/medie/ 9/38 Take-Home Message If music was invented Information is not Knowledge.
    [Show full text]