New Frontiers in Artificial Intelligence

Neg-Raising (NR) verbs form a class of verbs with a clausal complement that show the following behavior: when a negation syntactically attaches to the matrix predicate, it can semantically attach to the embedded predicate. This paper presents an account of NR predicates within Tree Adjoining Grammar (TAG). We propose a lexical semantic interpretation that heavily relies on a Montague-like semantics for TAG and on higher-order types.

This paper investigates measures of semantic similarity between conversations from an axiomatic perspective. We abstract away from real conversations, representing them as sequences of formulas, equipped with a notion of semantic interpretation that maps them into a different space. An example we use to illustrate our approach is the language of propositional logic with its classical semantics. We introduce and study a range of different candidate properties for metrics on such conversations, for the structure of the semantic space, and for the behavior of the interpretation function, and their interactions. We define four different metrics and explore their properties in this setting.

This paper introduces a world history ontology that supports reasoning of truth/falsehood of historical descriptions in natural languages. The core of the ontology includes an event classification according to certain basic properties such as necessary/sufficient conditions for the existence of events in the real world. We will discuss how this ontology functions in solving world history problems in Japan’s National Center Test for University Admissions, especially in the reasoning of “falsehood” of sentences and bridging of the “granularity difference” between target sentences and knowledge resources.

In this paper, we introduce hypersequent calculi for some modal logics extending S4 modal logic. In particular, we uniformly characterize hypersequent calculi for S4, S4.2, S4.3, S5 in terms of what are called “external modal structural rules” for hypersequent calculi. In addition to the monomodal logics, we also introduce simple bimodal logics combing S4 modality with another modality from each of the rest of logics. Using a proof-theoretic method, we prove cut-elimination for the hypersequent calculi for these logics and, as applications of it, we show soundness and faithfulness of Gödel embedding for the monomodal logics and the bimodal logic combining S4 and S5.

This paper considers politeness at the discourse level in terms of strategic choice. We begin with a discussion of the nature and levels of linguistic politeness from semantic and pragmatic perspectives, then turning to the way in which such strategies can be realized in natural language. A distinction is drawn between formal and polite linguistic behavior. We then provide a formal analysis in terms of the topological analysis of game strategies in an infinitely repeated game. This analysis extends that of [2]. It improves on that earlier work in three ways: (i) by considering a wider range of player ‘types’, (ii) by implementing the distinction between formality and politeness, and (iii) by analyzing a much wider range of kinds of politeness strategies, together with their positions in the Borel hierarchy [8].

Chierchia’s comparative analysis of nominals based on the two features [\(\pm \) arg] and [\(\pm \) pred] has lead to many discussions on the semantics of nominals in argument positions and predicate positions. On the other hand, many syntactic results about left dislocation and topicalization have been accumulated. In this paper, we will try to elucidate possibilities of bare plurals in the left periphery and differentiate their readings. By examining data from Italian and German we claim that different demands on foregoing contexts are organized as constructions, as far as the left periphery is concerned. In addition, those constructions also reflect an organization of discourse.

In a conversation, different kinds of speech acts are performed. Logic for communication has to deal with these various kinds of speech acts ([5]: 52). Additionally, for interpretation of conversations, it will be appropriate to take shared beliefs among communication partners into consideration. In this paper, we show that this problem can be dealt with in a framework that is a dynamic extension of the logic for normative systems.

This paper proposes a proof-theoretic definition for generalized quantifiers (GQs). Sundholm first proposed a proof-theoretic definition of GQs in the framework of constructive type theory. However, that definition is associated with three problems: the proportion problem, absence of strong interpretation and lack of definitional uniformity. This paper presents an alternative definition for “most” based on polymorphic dependent type theory and shows strong potential to serve as an alternative to the traditional model-theoretic approach.

This work deals with sentiment analysis on a corpus of French product reviews. We first introduce the corpus and how it was built. Then we present the results of two classification tasks that aimed at automatically detecting positive, negative and neutral reviews by using various machine learning techniques. We focus on methods that make use of feature selection techniques. This is done in order to facilitate the interpretation of the models produced so as to get some insights on the relative importance of linguistic items for marking sentiment and opinion. We develop this topic by looking at the output of the selection processes on various classes of lexical items and providing an explanation of the selection in argumentative terms.

I show that the exhaustive interpretation of answers can be explained as a conversational implicature through the Maxim of Relation, dealing with the problematic epistemic step (Sauerland 2004). I assume a fairly standard Maxim of Relation, that captures the same intuition as Roberts’ (1996) contextual entailment. I show that if a richer notion of meaning is adopted, in particular that of attentive semantics (Roelofsen 2011), this Maxim of Relation automatically becomes strong enough to enable exhaustivity implicatures. The results suggest that pragmatic reasoning is sensitive not only to the information an utterance provides, but also to the possibilities it draws attention to. Foremost, it shows that exhaustivity implicatures can be genuine conversational implicatures.

(Cannot access the full version? Download it from the author’s website.)

In this study, we define the neighborhood-sheaf semantics (NSS).

This paper discusses the challenges of legal norms in specialist domains - the interplay between industry/professional standards and legal norms, the information gap between legal and specialist domains and the need for interpretation at all stages of compliance - design, operation and justification. We propose extensions to the Eunomos legal knowledge management tool to help address the information gap, with particular attention to aligning norms with operational procedures, and the use of domain-specific specialist ontologies from multiple domains to help users understand and reason with norms on specialist topics. The paper focuses mainly on medical law and clinical guidelines.

The development of Question Answering (QA) systems has become important because it reveals research issues that require insight from a variety of disciplines, including Artificial Intelligence, Information Extraction, Natural Language Processing, and Psychology. Our goal here is to develop a QA approach to answer yes/no questions relevant to civil laws in legal bar exams. A bar examination is intended to determine whether a candidate is qualified to practice law in a given jurisdiction. We have found that the development of a QA system for this task provides insight into the challenges of formalizing reasoning about legal text, and about how to exploit advances in computational linguistics. We separate our QA approach into two steps. The first step is to identify legal documents relevant to the exam questions; the second step is to answer the questions by analyzing the relevant documents. In our initial approach described here, the first step has been already solved for us: the appropriate articles for each question have been identified by legal experts. So here, we focus on the second task, which can be considered as a form of Recognizing Textual Entailment (RTE), where input to the system is a question sentence and its corresponding civil law article(s), and the output is a binary answer: whether the question sentence is entailed from the article(s). We propose a hybrid method, which combines simple rules and an unsupervised learning model using deep linguistic features. We first construct a knowledge base for negation and antonym words for the legal domain. We then identify potential premise and conclusion components of input questions and documents, based on text patterns and separating commas. We further classify the questions into easy and difficult ones, and develop a two-phase method for answering yes/no questions. We answer easy questions by negation/antonym detection. For more difficult questions, we adapt an unsupervised machine learning method based on morphological, syntactic, and lexical semantic analysis on identified premises and conclusions. This provides the basis to compare the semantic correlation between a question and a legal article. Our experimental results show reasonable performance, which improves the baseline system, and outperforms an SVM-based supervised machine learning model.

Cooking and eating on a table is known as a Japanese dining style. As we cook “monja-yaki” on a table, how do we communicate with others? This paper indicates that cooking acts cause utterances to overlap and generate silence more frequently than when not cooking. The order of overlaps in table cooking is shown in two aspects: (1) accidental overlaps are not always repaired in cooking, and (2) co-telling of how to cook sometimes allows utterances to overlap. Besides, while cooking, there occur some kinds of sequence organization with bodily actions: (1) adjacency pairs are organized not only by language but also bodily actions, and (2) even if adjacency pairs are not sufficiently organized with language, bodily actions could complement the absence or insufficiency. Such orders of sequence organization of actions may make silence occur more frequently. Repeated occurrences of overlaps and silence in cooking may result from embodied interaction.

To deal with the question of what a sociable robot is, we describe how an educational robot is encountered by children, teachers and designers in a preschool. We consider the importance of the robot’s body by focusing on how its movements are contingently embedded in interactional situations. We point out that the effects of agency that these movements generate are inseparable from their grounding in locally coordinated, multimodal actions and interactions.

This paper studies an abduction problem in formal argumentation frameworks. Given an argument, an agent verifies whether the argument is justified or not in its argumentation framework. If the argument is not justified, the agent seeks conditions to explain the argument in its argumentation framework. We formulate such abductive reasoning in argumentation semantics and provide its computation in logic programming. Next we apply abduction in argumentation frameworks to reasoning by players in debate games. In debate games, two players have their own argumentation frameworks and each player builds claims to refute the opponent. A player may provide false or inaccurate arguments as a tactic to win the game. We show that abduction is used not only for seeking counter-claims but also for building dishonest claims in debate games.

An assurance case provides an argument that certain claims (usually concerning safety or other critical properties) are justified, based on given evidence concerning the context, design, and implementation of a system. An assurance case serves two purposes: reasoning and communication. For the first, the argument in the case should approach the standards of mathematical proof (though it may be grounded on premises—i.e., evidence—that are equivocal); for the second it must assist human stakeholders to grasp the essence of the case, to explore its details, and to challenge it. Because of the scale and complexity of assurance cases, both purposes benefit from mechanized assistance. We propose simple ways in which an assurance case, formalized in a mechanized verification system to support the first purpose, can be adapted to serve the second.

In this paper, we introduce an agreement subtree mapping kernel counting all of the agreement subtree mappings and design the algorithm to compute it for phylogenetic trees, which are unordered leaf-labeled full binary trees, in quadratic time. Then, by applying the agreement subtree mapping kernel to trimmed phylogenetic trees obtained from all the positions in nucleotide sequences for A (H1N1) influenza viruses, we classify pandemic viruses from non-pandemic viruses and viruses in one region from viruses in the other regions. On the other hand, for leaf-labeled trees, we show that the problem of counting all of the agreement subtree mappings is #P-complete.

Tree kernels are an effective method to capture the structural information of tree data of various applications and many algorithms have been proposed. Nevertheless, we do not have sufficient knowledge about how to select good kernels. To answer this question, we focus on 32 tree kernel algorithms defined within a certain framework to engineer positive definite kernels, and investigate them under two different parameter settings. The result is amazing. Three of the 64 tree kernels outperform the others, and their superiority proves statistically significant through t-tests. These kernels include the benchmark tree kernels proposed in the literature, while many of them are introduced and tested for the first time in this paper.

Outlier detection in mixed-type data, which contain both discrete and continuous features, is still a challenging problem. Here we newly introduce concept-based outlierness, which is defined on a hierarchy of clusters of data points and features, called the concept lattice, obtained by formal concept analysis (FCA). Intuitively, this outlierness is the degree of isolation of clusters on the hierarchy. Moreover, we investigate discretization of continuous features to embed the original continuous (Euclidean) space into the concept lattice. Our experiments show that the proposed method which detects concept-based outliers is more effective than other popular distance-based outlier detection methods that ignore the discreteness of features and do not take cluster relationships into account.