New Frontiers in Artificial Intelligence

The thesis that the probability of a conditional is the corresponding conditional probability of C, given A, enjoys wide currency among philosophers and growing empirical support in psychology. In this paper I ask how a probabilisitic account of conditionals along these lines could be extended to unconditional sentences, i.e., conditionals with interrogative antecedents. Such sentences are typically interpreted as equivalent to conjunctions of conditionals. This raises a number of challenges for a probabilistic account, chief among them the question of what the probability of a conjunction of conditionals should be. I offer an analysis which addresses these issues by extending the interpretation of conditonals in Bernoulli models to the case of unconditionals.

The purpose of this paper is to solve the simple sentence puzzle about proper names. (1) Superman leaps more tall buildings than Clark Kent. (2) Superman = Clark Kent. (3) Superman leaps more tall buildings than Superman. Even when (1) and (2) are true, (3) is false. It will be shown that this is not a real puzzle, because (i) (1) and (3) do not express singular propositions, and (ii) the identity statement in (2) only concerns singular propositions. In (1) and (3), the proper names refer to aspects of an individual at the level of explicature, while identity statements of the form X = Y mean that Y can be substituted for X salva veritate, only in singular propositions about X /Y. Given this difference in reference between (1)/(3) and (2), the conjunction of (1) and (2) does not entail (3), in accordance with our intuition.

Cantonese has a number of sentence-final particles which serve various communicative functions. This paper looks into two of the most frequently used particles, wo3 and lo1. We propose that wo3 and lo1 are expressive items: Wo3 indicates unexpectedness of the propositional content or the current discourse move, while lo1 indicates expectedness of the propositional content or the current discourse move. We employ Default Logic to characterize the notion of (un)expectedness by normality conditionals. The analysis has a further implication on the Gricean Cooperative Principle in that the use of wo3 and lo1 makes reference to the general world knowledge which includes conditions on how the discourse should normally proceed.

We consider two hypotheses about how rhetorical structure and QUD structure might come together to provide a more general pragmatic theory. Taking SDRT ([2]) and some basic principles from [18]’s QUD framework as starting points, we first consider the possibility that rhetorical relations can be modelled as QUDs, and vice versa. We ultimately reject this hypothesis in favor of the possibility that QUDs correspond to topics that bind together the members of complex discourse units.

This paper introduces a collection of inference problems intended for use in evaluation of semantic theories and semantic processing systems for Japanese. The problem set categorizes inference problems according to semantic phenomena that they involve, following the general policy of the FraCaS test suite. It consists of multilingual and Japanese subsets, which together cover both universal semantic phenomena and Japanese-specific ones. This paper outlines the design policy used in constructing the problem set and the contents of a beta version, currently available online.

Recently introduced Applicative Abstract Categorial Grammars (AACG) extend the Abstract Categorial Grammar (ACG) formalism to make it more suitable for semantic analyses while preserving all of its benefits for syntactic analyses. The surface form of a sentence, the abstract (tecto-) form, as well as the meaning are all uniformly represented in AACG as typed terms, or trees. The meaning of a sentence is obtained by applying to the abstract form a composition of elementary, deterministic but generally partial transformations. These term tree transformations are specified precisely and can be carried out mechanically. The rigor of AACG facilitates its straightforward implementation, in Coq, Haskell, Grammatical Framework, etc.We put AACG through its paces, illustrating its expressive power and positive as well as negative predictions on the wide range of analyses: gender-marked pronouns, quantifier ambiguity, scoping islands and binding, crossover, topicalization, and inverse linking. Most of these analyses have not been attempted for ACG.AACG offers a different perspective on linguistic side-effects, demonstrating compositionality not just of meanings but of transformations.

The scope parallelism in the so-called Geach sentences in right-node raising (Every boy admires, and every girl detests, some saxophonist) poses a difficult challenge to many analyses of right-node raising, including ones formulated in the type-logical variants of categorial grammar (e.g. Kubota and Levine (2015)). In this paper, we first discuss Steedman’s (2012) solution to this problem in Combinatory Categorial Grammar, and point out some empirical problems for it. We then propose a novel analysis of the Geach problem within Hybrid Type-Logical Categorial Grammar (Kubota and Levine 2015), by incorporating Dependent Type Semantics (Bekki 2014) as the semantic component of the theory. The key solution for the puzzle consists in linking quantifiers to the argument positions that they correspond to via an anaphoric process. Independently motivated mechanisms for anaphora resolution in DTS then automatically predicts the scope parallelism in Geach sentences as a consequence of binding parallelism independently observed in right-node raising sentences.

This paper proposes an analysis of declaratives with the German discourse particles doch and ja as context change potentials (CCPs), imposing restrictions on input and output contexts consisting of public beliefs of the discourse participants. The analysis accounts for a wider range of data than previous approaches and makes novel predictions on the distribution of doch and ja by defining public beliefs as independent of fist-order beliefs, and modeling the difference between ja and doch in terms of whether a public belief of the addressee is presupposed (ja), or not (doch).

Kuno (1973) and others describe the Japanese junctor ya as conjunction. But, Sudo (2014) analyzes ya as a disjunction with a conjunctive implicature. We compare ya with other junctors and implicature triggers experimental using mouse-tracking. Our two main results are: (1) ya differs from lexical conjunctions corroborating Sudo’s (2014) proposal. (2) The time-course of the conjunctive implicature of ya argues against the details of Sudo’s (2014) implementation, and instead favors an account similar to other cases of conjunctive implicatures.

This paper discusses the phenomenon of what I call the “projection of non-at issue meaning via modal support” shown in the Japanese counter-expectational intensifier yoppodo and the counter expectational scale-reversal adverb kaette, and considers the variation of projective content from a new perspective. I show that, unlike the typical conventional implicatures (CIs) like appositives and expressives (e.g., Potts [19]), kaette and yoppodo can project out of the complement of a belief predicate only if there is a modal in the main clause. I argue that yoppodo and kaette belong to a new class of projective content that requires consistency between an at-issue meaning and a CI meaning in terms of a judge. This paper provides a new perspective for the typology of projective content.

This paper has two aims. The first is to provide a characterization of evaluative predicates (‘good’, ‘horrible’, ‘beautiful’). The second is to explain how ordinary predicates, such as ‘intense’ or ‘insane’, may be used evaluatively, and how they convey sometimes a positive and sometimes a negative evaluation, depending on the context. I propose a semantic account, which, in a nutshell, relies on the fact that evaluative predicates are typically multidimensional adjectives, and that the choice, as well as the respective weights of the relevant dimensions, may vary with the context. Thus in a context in which a negative dimension has been brought to salience, the overall evaluation carried by the use of the predicate will likely be negative; mutatis mutandis for the positive case. The paper ends with a comparison between this approach and the pragmatic approach, and suggests that rather than compete, the two complement each other.

This paper formulates strong permission in prescriptive causal models. The key features of this formulation are that (a) strong permission is encoded in causal models in a way suitable for interaction with functional equations, (b) the logic is simpler and more straightforward than other formulations of strong permission such as those utilizing defeasible reasoning or linear logic, (c) when it is applied to the free choice permission problem, it avoids paradox formation in a satisfactory manner, and (d) it also handles the embedding of strong permission, e.g. in conditionals, by exploiting interventionist counterfactuals in causal models.

Some truth theories allow to represent and prove generalized statements as “all that you said is true” or “all theorems of $${\mathbf {PA}}$$ are true” in the sense of deflationism.

Abductive Logic Programming (ALP) has been exploited to formalize societies of agents, commitments and norms, taking advantage from ALP operational support as a (static or dynamic) verification tool. In [7], the most common deontic operators (obligation, prohibition, permission) are mapped into the abductive expectations of an ALP framework for agent societies. Building upon such correspondence, in [5], authors introduced $$Deon^{+}$$, a language where obligation and prohibition deontic operators are enriched with quantification over time, by means of ALP and Constraint Logic Programming (CLP).In recent work [30, 31], we have shown that the same ALP framework can be suitable to represent Datalog$$^\pm $$ ontologies. Ontologies are a fundamental component of both the Semantic Web and knowledge-based systems, even in the legal setting, since they provide a formal and machine manipulable model of a domain.In this work, we show that ALP is a suitable framework for representing both norms and ontologies. Normative reasoning and ontological query answering are obtained by applying the same abductive proof procedure, smoothly achieving their integration. In particular, we consider the ALP framework named $$\mathcal {S}\text {CIFF}$$ and derived from the IFF abductive framework, able to deal with existentially (and universally) quantified variables in rule heads and CLP constraints.The main advantage is that this integration is achieved within a single language, grounded on abduction in computational logic.

A scenario that was not considered at the time of enforcing a civil code article may be discovered later. In case application of the civil code article in the discovered scenario is not consistent with the intention of the article, it is necessary that the article be appropriately updated. We show that this kind of civil code update that is induced upon reaction to augmentation of knowledge can be modelled in a belief revision theory. We develop our formal framework, and show one instantiation of the framework with case application of a civil code article.

In this paper, we propose a new approach to formalize real-world obligations that may be found in existing legislation. Specifically, we propose to formalize real-world obligations by combining insights of two logical frameworks: Input/Output logic, belonging to the literature in deontic logic and normative reasoning, and the Reification-based approach of Jerry R. Hobbs, belonging to the literature in Natural Language Semantics. The present paper represents the first step of the ProLeMAS project, whose main goal is the one of filling the gap between the current logical formalizations of legal text, mostly propositional, and the richness of Natural Language Semantics.

Data protection, currently under the limelight at the European level, is undergoing a long and complex reform that is finally approaching its completion. Consequently, there is an urgent need to customize semantic standards towards the prospective legal framework. The aim of this paper is to provide a bottom-up ontology describing the constituents of data protection domain and its relationships. Our contribution envisions a methodology to highlight the (new) duties of data controllers and foster the transition of IT-based systems, services, tools and businesses to comply with the new General Data Protection Regulation. This structure may serve as the foundation for the design of data protection compliant information systems.

Statutory sentences are generally difficult to read because of their complicated expressions and length. Such difficulty is one reason for the low quality of statistical machine translation (SMT). Multi-word expressions (MWEs) also complicate statutory sentences and extend their length. Therefore, we proposed a method that utilizes MWEs to improve the SMT system of statutory sentences. In our method, we extracted the monolingual MWEs from a parallel corpus, automatically acquired these translations based on the Dice coefficient, and integrated the extracted bilingual MWEs into an SMT system by the single-tokenization strategy. The experiment results with our SMT system using the proposed method significantly improved the translation quality. Although automatic translation equivalent acquisition using the Dice coefficient is not perfect, the best system’s score was close to a system that used bilingual MWEs whose equivalents are translated by hand.

In legal debates, it is a matter of importance whether one’s own argument is accepted or not. For this, we propose evaluation method for calculating the acceptability of arguments, and a tool developed based on the measures. This method is called reliability-based argumentation framework (RAFs), extended from argumentation framework, seeking for multivalued dialectical validities of arguments reliable to some extent. The modular reliability-based argumentation framework (MRAF) based on RAFs is able to integrate the RAF semantics in every module. This leads to an over-all valuation of the acceptability of argumentations including several local arguments. The argumentation-support tool can represent the utterance logs of those who join an debate, the argumentation diagram its users made, and the argumentation framework converted from this, contributing to the intuitive comprehension of the logical structures of arguments and their acceptability. This tool also enables represented argumentation framework to be converted into modular structures of local AFs, leading to an overall valuation of the acceptability of arguments.

Our legal question answering system combines legal information retrieval and textual entailment, and we describe a legal question answering system that exploits a deep convolutional neural network. We have evaluated our system using the training/test data from the competition on legal information extraction/entailment (COLIEE). The competition focuses on the legal information processing related to answering yes/no questions from Japanese legal bar exams, and it consists of three phases: ad-hoc legal information retrieval, textual entailment, and a learning model-driven combination of the two phases. Phase 1 requires the identification of Japan civil law articles relevant to a legal bar exam query. For that phase, we have implemented a combined TF-IDF and Ranking SVM information retrieval component. Phase 2 requires the system to answer “Yes” or “No” to previously unseen queries, by comparing extracted meanings of queries with relevant articles. Our training of an entailment model focuses on features based on word embeddings, syntactic similarities and identification of negation/antonym relations. We augment our textual entailment component with a convolutional neural network with dropout regularization and Rectified Linear Units. To our knowledge, our study is the first to adapt deep learning for textual entailment. Experimental evaluation demonstrates the effectiveness of the convolutional neural network and dropout regularization. The results show that our deep learning-based method outperforms our baseline SVM-based supervised model and K-means clustering.

In the context of the Competition on Legal Information Extraction/Entailment (COLIEE), we propose a method comprising the necessary steps for finding relevant documents to a legal question and deciding on textual entailment evidence to provide a correct answer. The proposed method is based on the combination of several lexical and morphological characteristics, to build a language model and a set of features for Machine Learning algorithms. We provide a detailed study on the proposed method performance and failure cases, indicating that it is competitive with state-of-the-art approaches on Legal Information Retrieval and Question Answering, while not needing extensive training data nor depending on expert produced knowledge. The proposed method achieved significant results in the competition, indicating a substantial level of adequacy for the tasks addressed.

In the current paper we examine how to assess knowledge and expertise in an argumentation based setting. In particular, we are looking for a formal criterion to determine whether one agent is more informed than another. Several such criteria are discussed, each of which has its own advantages and disadvantages.

An assurance case provides a structured argument to establish a claim for a system based on evidence about the system and its environment. I propose a simple interpretation for the overall argument that uses epistemic methods for its evidential or leaf steps and logic for its reasoning or interior steps: evidential steps that cross some threshold of credibility are accepted as premises in a classical deductive interpretation of the reasoning steps. Thus, all uncertainty is located in the assessment of evidence. I argue for the utility of this interpretation.

This paper introduces argument-based decision-tree for learning acceptability of arguments. We specifically examine an attack relation existing between arguments, without referring to any contents, either sentences or words, existing in individual arguments. This idea is formalized using decision trees in which their attributes are instantiated by complete, preferred, stable and grounded extensions, respectively, defined by acceptability semantics. This study extracted 38 arguments and 4 utterers from an argument about euthanasia that actually took place on a social media site. Also, 21 training data were collected by asking them to express their attitudes either for or against the individual 38 arguments. By stratifying audiences in accordance with consistency with utterers, leave-two-out cross validation yielded results with a 0.73 AUC value, on average. This fact demonstrates that our argument-based decision-tree learning is expected to be fairly useful for agents who have a definite position on an issue of argument.

There are many sleep tracking technologies in the consumer market nowadays. These technologies offer rich functions ranging from sleep pattern tracking to smart alarm clock. However, previous study indicates that users find these technologies of little use in facilitating sleep quality improvement, as simply making a user aware of how poor his/her sleep is provides no actionable information on how to improve it. Armed with such understanding, we proposed an architecture for designing intelligent sleep analysis systems and developed a prototype called SleepExplorer to help users automatically analyse and visualize the interrelationship of his/her sleep quality and the context (i.e., psychological states, physiological states, lifestyle, and environment). Such contextual information is crucial in helping users understand what the potential reasons for their sleep problems might be. We conducted a 2-week field study with 10 diverse participants, learning that SleepExplorer help users make sense of their sleep-tracking data and reflect on their lifestyle, and that the system has potentially positive impact on sleep behaviour change.

In Japan, the ratio of people with lifestyle-related diseases has increased to approximately 30%. Individuals as well as the Nation are getting more and more health-conscious, and special attention has been made to body trunk because it is vital for injury prevention, physical strength, and beauty. Various training methods have been proposed to increase the muscle mass of body trunk. However, for sports that emphasize somatoform such as dance, the strength of the trunk is mainly decided by smooth use of the trunk rather than its muscle mass. In this paper, in order to evaluate the use of the trunk torsion movement, we proposed a new trunk torsion model for the purpose of evaluating two trunk torsion standard movements. We also developed a mobile application named “Axis Visualizer” based on the proposed trunk torsion model analyzing sensor data in the device. Axis Visualizer generates higher score when a user rotates the shoulders or hips smoothly with axis fixed and high frequencies. This application can support trainers and coaches to visualize the use of customers’ trunk and to increase the training effect.

Time series data are found everywhere in the real world and their analysis is needed in many practical situations. Multivariate time series data poses problem for analysis due to its dynamic nature and traditional machine learning algorithms for static data become unsuitable for direct application. A measure to assess the similarity of two time series is essential in time series processing and a lot of measures have been developed. In this work a comparative study of some of the most popular similarity measures has been done with 43 benchmark data set from UCR time series repository. It has been found that, on the average over the different data sets, DTW performs better in terms of classification accuracy but it has high computational cost. A simple processing technique for reducing the data set to lower computational cost without much degradation in classification accuracy is proposed and studied. A new similarity measure is also proposed and its efficiency is examined compared to other measures.

In this paper, we formulate propagation patterns as the pairs of records in the same bacterial culture occurring within a fixed span in bacterial culture data. Then, we design the exhaustive search algorithm to extract all of the propagation patterns from bacterial culture data based on the extended principle of the 2-dimensional career map to determine whether two records in bacterial culture data belong to the same bacterial culture or the different ones. In particular, we focus on infectious propagation patterns, in which two patients are not identical, and they are in the same room and/or treated by the same physician. Finally, we give the experimental results to extract all of the propagation patterns and analyze them.

In this paper, we proposed an efficient heuristic algorithm for real-time anomaly detection of periodic bio-signals. We introduced a new concept, “mother signal” which is the average of normal subsequences of one period length. Their number is overwhelmingly large compared to anomalies. From the time series, first we find the fundamental time period, assuming the period to be stable over the whole time. Next, we find the normal subsequence of length equal to time-period and call it the “mother signal”. When the distance of a subsequence of same length is large from the mother signal, we identify it as anomaly. While calculating the distance, we ensure that it is not large due to time shift. To ensure that, we shift-and-rotate the subsequence in step of one slot at a time and find the minimum distance of all such comparisons. The proposed heuristic algorithm using mother signal is efficient. Results are compared and found to be similar to that obtained using brute force comparisons of all possible pairs. Computational costs are compared to show that the proposed method is more efficient compared to existing works.

In the top news stories, the commenting activity is rising and falling until it stops. In some ongoing news stories such as disasters like the disappearance of flight MH370, global warming or climate change, political turmoil or economic crisis, this commenting activity cycle can repeat and last many years. To our knowledge, a study and analysis of those data does not exist up to now. There is a need to separate facts, opinions and junk within those comments data. In this paper, we present our framework for supporting readers in analyzing and visualizing facts, opinions and topics in the comments and its extension with comments aggregation and summarization for comments within several news articles for the same event. We added a time-series analysis and comments features such as surprising comments and a preferential threads attachment model.

We investigated whether it is easier for adolescents with ASD to establish communication using eye-gaze with an android than a human. Two-days-experiment was conducted to measure eye-gaze patterns of subjects during conversation with two types of interlocutor, a female type android robot and a human female, where the interlocutors used their gaze to identify what they were mentioning to. Fixation bias on the target object that the interlocutor was referring to by using her eye gaze showed that the adolescents with ASD followed the gaze of android more than human’s although the sample size was still small.

With the growth of robot technology, educational support robots have received increasing attention. Most of the previous studies have focused on collaborative learning effects between educational support robots and healthy children. However, few studies have addressed the use of educational-support robots for children with autism spectrum disorders. Because the number of students in primary school with autism spectrum disorders has increased, the need for educational support robots developed to assist autistic children becomes more imperative. Therefore, this study investigates the practicality of collaborative learning and work with educational support robots and reports the feasibility of collaborative learning between educational-support robots and children with autism spectrum disorders.

We examine whether teachers in special school accept communication robots as new educational tools and what they expect by introduction of them into class activities. We asked the teachers to experience demonstrations of three different types of robots, Telenoid, Hugvie, and M3-Synchy, which are available to assist teachers in educational situations and to fill in questionnaire about their impression on the robots and possible applications. The results of the questionnaire showed that more than half of teachers recognized that the robots show positive effect on their students and are useful tools not only in group activities but also in independent activities. Surprisingly, almost all teachers considered Hugvie, which is a cushion-like communication medium, have educational effects on their students, comparing it with the other robots. We also discuss what kind of robots are required by teachers.

Some children with social anxiety and autism spectrum disorders (ASD) are characterized by deficits in social communication consisting of a lack of speech in specific social situations in which there is an expectation of speaking, although these children do speak in other situations. Comorbid social anxiety and ASD is considered hard to treat. A variety of therapeutic and educational approaches have been developed, which have had educational benefits for some children with social anxiety and ASD. Thus, there is an urgent need for the development and application of novel and more efficacious treatment strategies. We introduced a “taking note system,” in which a subject operates an android robot and communicates with others through it. Post intervention, the participant showed a significant decrease in social anxiety. Based on our case report, an intervention using a “taking note system” is effective for decreasing stress. This finding suggests that this approach is promising and warrants further study.

Objectives: The purpose of this study is to assess the effectiveness of PARO assisted therapy for Autism Spectrum Disorder (ASD) in the child and adolescent psychiatric ward. Methods: In the Child and Adolescent Psychiatry Ward of Shikoku Medical Center for Children and Adults, we put PARO near the door of the nurse station and told the inpatients to play with it freely in the hall of the ward after getting permission from the staff. The interaction between patients and PARO was observed. Results: It was thought to be useful for same patients with ASD in developing good communication or in reduction of impulsive behaviors or anxiety. However, others hated it for some features such as its big eyes or a slight drive noise. Conclusions: Before animal type robot-assisted therapy is introduced as a tool for the patients with ASD, the kind of patients who are benefitted by this approach and how the approach would work in the treatment must be clarified.

This paper discusses the relationship between brain activity and improvement of skills during the process of embodied knowledge acquisition by imitation. Study subjects watched a video clip of a working procedure and then executed the same series of actions. Each experiment was conducted twice. After the first experiment, we set up three practice trials. Using near-infrared spectroscopy, we found that the trend in oxy-hemoglobin levels during the observation task shifted toward a low-level increase in the dorsolateral prefrontal area and a low-level decrease in the frontal lobe with improvement in performing the skill. In the execution task, the trend in oxy-hemoglobin shifted toward an increase in the dorsolateral prefrontal area and toward a decrease in the frontal pole with improvement in skill performance. These results suggest that activity in the frontal area changes during the process of embodied knowledge acquisition.

In this paper, we discuss the cognitive role of analogical abduction in skill acquisition. Abductive inference makes it possible to find missing links that explain a given knack in achieving a skillful task. We introduced meta level abduction to realize rule abduction which is mandatory in finding intermediate missing links to be added in knack explanation. Analogical abduction can be achieved by adding analogical inference rules to causality rules within meta level abduction. We have applied our analogical abduction method to the problem of explaining the difficult cello playing techniques of spiccato and rapid cross strings of the bow movement. Our method has constructed persuasive analogical explanations about how to play them. We have used a model of forced vibration mechanics as the analogy base world for spiccato, and the specification of the skeletal structure of the hand as the basis for the cross string bowing technique. We also have applied analogical abduction to show the effectiveness of a metaphorical expression of “eating pancake on the sly” to achieve forte-piano dynamics, and successfully identified an analogical explanation of how it works. Through these examples, we show the effectiveness of analogical abduction in skill acquisition. Furthermore we discuss the importance of meta level representation as a basis for providing rich human cognitive paradigm such as causality, analogy and metaphor. Finally we propose a cognitive architecture which gives a possible structure for realizing accommodation on our analogical abduction schema.

Past literature has suggested that reading text as a whole cannot be reduced to merely an aggregation of sentence processing, but instead there are expected to be some context-dependent stylistic differences in the reading process. It has been, however, difficult to capture such context-dependent reading styles or modes. In this study, under the hypothesis that the statistics of reading time reflects such reading modes, we introduce a new statistical approach to capture them. Our analysis of the distributions of reading times identified two distinct modes of reading. In further analysis, we found that the temporal profiles of the two reading modes were correlated to the reader’s degree of engagement. We discuss how the context dependency of the reading modes is related to dynamic construction of the reader’s knowledge of narratives.

The purpose of the present study is to reveal the fundamental skills for slacklining. A slackline is a flat belt tightly spanned between two anchor points. Because it bounces and swings in all directions, maintaining balance on it is difficult. In the practical field of slackline training, instructors share their skills based on personal experience. In a basic slackline course, they begin by teaching a fundamental skill, such as single-leg standing on a slackline, by explaining how they do it. However, such first-person perspectives on slacklining skills have not been scientifically investigated. According to instructors’ knowledge based on personal experience, we hypothesize the skills for single-leg standing on the slackline. The present study examines current hypotheses by comparing performances at different skill level (i.e., experienced vs. novice). This article introduces our pilot study, including current hypotheses and data from preliminary experiment, and discusses them.