Automatic Question Generation (AQG) aims to generate valid, coherent questions based on given text passages in pre-trained language models. While AQG has been a significant area of retrieval augmentation and agent-based systems, current QG models face limitations, especially with sequential models like Transformer, which struggle with modeling complex logical structures and limited question coherence and depth. This paper proposes Q-Chain, a framework designed to optimize logic and educational values. Q-Chain features: (1)Differentiable logic layers in GMP model conditional dependencies and counterfactuals, ensuring logical rigor. (2) Bloom's taxonomy-guided gating adjusts question difficulty to align with pedagogical goals. (3) Direct generation of logic-structured question graphs enhanced by counterfactual training. Experimental results show Q-Chain outperforms GPT-3.5, T5, and BART on the MedQuAD dataset (0.82 vs 0.80 vs 0.72 and 0.74 F1-score) and shows superior robustness to noisy inputs, achieving a 4.1/5 human rating on counterfactual questions, 40% better than BART.

As robots are increasingly integrated into everyday life, their ability to communicate emotions becomes crucial to foster positive human-robot interactions. However, the key factors for the recognition of artificial emotions are still poorly understood. This study explores whether children can better recognize emotions expressed by robots using gestures or by robots using facial expressions. We observed ten five-year-olds and fourteen six-year-olds during an in-the-wild workshop. The results revealed significant differences in emotion recognition based on robot type, age, and emotion. Children who interact with the gesture robot were 90% less likely to accurately recognize emotions than the robot with facial expressions. Six-year-olds demonstrated significantly higher accuracy, with 8.30 times greater chances of recognizing emotions than five-year-olds. Regarding emotion type, children were significantly more likely to identify sadness and happiness correctly than anger. These findings highlight the importance of incorporating facial expressions to aid in emotion recognition and make robots more approachable for young children.

Abstract Skeleton Ground Truth (GT) is critical to the success of supervised skeleton extraction methods, especially with the popularity of deep learning techniques. Furthermore, we see skeleton GTs used not only for training skeleton detectors with Convolutional Neural Networks (CNN), but also for evaluating skeleton-related pruning and matching algorithms. However, most existing shape and image datasets suffer from the lack of skeleton GT and inconsistency of GT standards. As a result, it is difficult to evaluate and reproduce CNN-based skeleton detectors and algorithms on a fair basis. In this paper, we present a heuristic strategy for object skeleton GT extraction in binary shapes and natural images. Our strategy is built on an extended theory of diagnosticity hypothesis, which enables encoding human-in-the-loop GT extraction based on clues from the target’s context, simplicity, and completeness. Using this strategy, we developed a tool, SkeView, to generate skeleton GT of 17 existing shape and image datasets. The GTs are then structurally evaluated with representative methods to build viable baselines for fair comparisons. Experiments demonstrate that GTs generated by our strategy yield promising quality with respect to standard consistency, and also provide a balance between simplicity and completeness.

We present here some lessons learned from observations and applications of child-robot interaction research in diverse ecosystems such as schools, therapy centers, and hospitals, where the interaction was facilitated in real-world circumstances rather than lab settings. Specifically, we use observational results from our reflections on multiple child-robot interaction practices in the wild conducted over a 9-year research period. Using these exploratory studies, we outline some general design considerations and adaptation guidelines for improving the design and implementation of robotic systems in healthcare and education that might lead to more practical, feasible, and ethically sustainable results.

It seems intuitive that a social robot should communicate in the native language of the user. On the other hand, English has become lingua franca among scientific research groups. We conducted a study to see whether native Czech speakers (fluent in English) prefer to interact with a robot in English or in Czech. We found that our participants perceived an English-speaking robot as more intelligent than a Czech-speaking robot. However, those who had previous experience with robots perceived a Czech-speaking robot to be safer.

Telling lies and faking emotions is quite common in human-human interactions: though there are risks, in many situations such behaviours provide social benefits. In recent years, there have been many social robots and chatbots that fake emotions or behave deceptively with their users. In this paper, I present a few examples of such robots and chatbots, and analyze their ethical aspects. Three scenarios are presented where some kind of lying or deceptive behaviour might be justified. Then five approaches to deceptive behaviours – no deception, blatant deception, tactful deception, nudging, and self deception are discussed and their implications are analyzed. I conclude by arguing that we need to develop localized and culture specific solutions to incorporating deception in social robots and chatbots.

As child-robot interactions become more and more common in daily life environment, it is important to examine how robot’s errors influence children’s behavior. We explored how a robot’s unexpected behaviors affect child-robot interactions during two workshops on active reading: one in a modern art museum and one in a school. We observed the behavior and attitudes of 42 children from three age groups: 6-7 years, 8-10 years, and 10-12 years. Through our observations, we identified six different types of surprising robot behaviors: personality, movement malfunctions, inconsistent behavior, mispronunciation, delays, and freezing. Using a qualitative analysis, we examined how children responded to each type of behavior, and we observed similarities and differences between the age groups. Based on our findings, we propose guidelines for designing age-appropriate learning interactions with social robots.

This article introduces a novel approach to generate visually promising skeletons automatically without any manual tuning. In practice, it is challenging to extract promising skeletons directly using existing approaches. This is because they either cannot fully preserve shape features, or require manual intervention, such as boundary smoothing and skeleton pruning, to justify the eye-level view assumption. We propose an approach here that generates backbone and dense skeletons by shape input, and then extends the backbone branches via skeleton grafting from the dense skeleton to ensure a well-integrated output. Based on our evaluation, the generated skeletons best depict the shapes at levels that are similar to human perception. To evaluate and fully express the properties of the extracted skeletons, we introduce two potential functions within the high-order matching protocol to improve the accuracy of skeleton-based matching. These two functions fuse the similarities between skeleton graphs and geometrical relations characterized by multiple skeleton endpoints. Experiments on three high-order matching protocols show that the proposed potential functions can effectively reduce the number of incorrect matches.

Abstract The cold-start scenario is a critical problem for recommendation systems, especially in dynamically changing domains such as online news services. In this research, we aim at addressing the cold-start situation by adapting an unsupervised neural User2Vec method to represent new users and articles in a multidimensional space. Toward this goal, we propose an extension of the Doc2Vec model that is capable of representing users with unknown history by building embeddings of their metadata labels along with item representations. We evaluate our proposed approach with respect to different parameter configurations on three real-world recommendation datasets with different characteristics. Our results show that this approach may be applied as an efficient alternative to the factorization machine-based method when the user and item metadata are used and hence can be applied in the cold-start scenario for both new users and new items. Additionally, as our solution represents the user and item labels in the same vector space, we can analyze the spatial relations among these labels to reveal latent interest features of the audience groups as well as possible data biases and disparities.

We address the problem of executing tool-using manipulation skills in scenarios where the objects to be used may vary. We assume that point clouds of the tool and target object can be obtained, but no interpretation or further knowledge about these objects is provided. The system must interpret the point clouds and decide how to use the tool to complete a manipulation task with a target object; this means it must adjust motion trajectories appropriately to complete the task. We tackle three everyday manipulations: scraping material from a tool into a container, cutting, and scooping from a container. Our solution encodes these manipulation skills in a generic way, with parameters that can be filled in at run-time via queries to a robot perception module; the perception module abstracts the functional parts of the tool and extracts key parameters that are needed for the task. The approach is evaluated in simulation and with selected examples on a PR2 robot.

In child-robot interaction, the element of trust towards the robot is critical. This is particularly important the first time the child meets the robot, as the trust gained during this interaction can play a decisive role in future interactions. We present an in-the-wild study where Polish kindergartners interacted with a Pepper robot. The videos of this study were analyzed for the issues of trust, anthropomorphization, and reaction to malfunction, with the assumption that the last two factors influence the children's trust towards Pepper. Our results reveal children's interest in the robot performing tasks specific for humans, highlight the importance of the conversation scenario and the need for an extended library of answers provided by the robot about its abilities or origin and show how children tend to provoke the robot.

In recent years, there has been a growing interest in applying deep learning techniques for automatic generation of software. To achieve this ambitious objective, a number of smaller research goals need to be reached, one of which is automatic categorization of software, used in numerous tasks of software intelligence. We present here an approach to this problem using a set of low-level features derived from lexical analysis of software code. We compare different feature sets for categorizing software and also apply different supervised machine learning algorithms to perform the classification task. The representation allows us to identify the most relevant libraries used for each class, and we use the best-performing classifier to accomplish this. We evaluate our approach by applying it to categorize popular Python projects from Github.

We explored how emotional cues presented in visual and haptic modalities interact. We constructed an affective haptic dataset, and used the emotional visual stimuli from the International Affective Picture System (IAPS). Participants were asked to rate the visual stimuli, haptic stimuli and visualhaptic stimuli. Analysis of the results indicates that the presence of haptic stimulus affects the arousal of the visual stimulus, but does not affect the valence significantly. We further explored this interaction in terms of the intensity, frequency, waveform and rhythm of the haptic stimuli. We then provide a set of guidelines on visual-haptic interaction that could be used in design of multimodal affective feedback.

In recent years, many empirical studies of legal decision-making process have shown that it incorporates many cognitive, affective, and supra-legal factors. Our goal is to design artificial intelligence systems that model these aspects of legal decision-making. Our vision is to implement a kind of legal assistant that can be used by lawyers and judges to run through different scenarios and produce arguments for different, and possibly contradictory, decisions. We propose a multi-agent blackboard architecture for such an assistive system, employing some insights from our previous work on a context-aware recommender system.

Recent research in human computation has focused on improving the quality of work done by crowd workers on crowdsourcing platforms. Multiple approaches have been adopted like filtering crowd workers through qualification tasks, and aggregating responses from multiple crowd workers to obtain consensus. We investigate here how improving the presentation of the task itself by using cognitively inspired features affects the performance of crowd workers. We illustrate this with a case-study for the task of extracting text from scanned images. We generated six task-presentation designs by modifying two parameters - visual saliency of the target fields and working memory requirements - and conducted experiments on Amazon Mechanical Turk (AMT) and with an eye-tracker in the lab setting. Our results identify which task-design parameters (e.g. highlighting target fields) result in improved performance, and which ones do not (e.g. reducing the number of distractors). In conclusion, we claim that the use of cognitively inspired features for task design is a powerful technique for maximizing the performance of crowd workers.

Current models of web navigation focus only on the influence of textual information and ignore the role of graphical information. We studied the differential role of text and graphics in identifying web page widgets classified into two kinds: textual and graphical. Four different versions of web pages were created by systematically removing textual and graphical information from each page. The task of the participants was to locate either textual or graphical widgets on the displayed web page. Results show that for any widget, the task-completion time and the number of clicks were significantly less in web pages with graphics than in those with no graphics. This demonstrates the importance of graphical information. However, textual information is also important because performance in locating graphical widgets under no-graphics conditions was better when text was present than with no text. Since, for identifying graphical widgets, text and graphics interact and complement each other, we conclude that cognitive models on web navigation should include the role of graphical information next to textual information.

The aim of this special issue is to contribute to the improvement of the existing cognitive models of web navigation like SNIF-ACT (Fu and Pirolli 2007), CoLiDeS (Blackmon et al. 2005, Kitajima et ...

CoLiDeS + Pic is a cognitive model of web-navigation that incorporates semantic information from pictures intoCoLiDeS. In our earlier research, we have demonstrated that by incorporating semantic information from pictures, CoLiDeS + Pic can predict the hyperlinks on the shortest path more frequently, and also with greater information scent, compared to earlier cognitive models of web-navigation like CoLiDeS that relied only on textualinformation from hyperlinks. In this article, we investigate the following research questions. First, would the increase in information scent have an impact on the actual user navigation behaviour? Second, do users actually follow the navigation path predicted by CoLiDeS + Pic? In other words, would CoLiDeS + Pic predict actual user navigation behaviour more accurately than CoLiDeS? We investigate these questions by varying the relevance of pictures on aweb page and studying the impact of varying relevance on the user navigation patterns. We found that under the highly relevant picture condition, users were more accurate and took less time to finish their tasks. Also, under thehighly relevant picture condition, CoLiDeS + Pic predicts significantly greater number of actual user clicks. There was no significant difference in model predictions between the lowly relevant picture condition and no-picture condition. These results validate the predictions made by CoLiDeS + Pic.

Comprehension-based linked model of deliberate search (CoLiDeS) + Pic is a cognitive model of web-navigation that takes into account the semantic information from graphical elements present on a web-page to compute the information scent value of the hyperlinks. The model is based on CoLiDeS, which has a two-phase processing cycle: (a) attention phase, which first parses the web-page and focuses attention on the region of the web-page that is semantically most similar to the goal, and (b) action-selection phase, which evaluates the available actions in the focused region and selects a particular action such as clicking a link. The graphical elements are important both for attracting attention to a region of the web-page and for communicating semantic meaning that may alter or enhance the meaning of the hyperlink labels. In the first part of this article, we give a theoretical explanation of the CoLiDeS + Pic model and describe the methodology followed to implement it. In the second part, we run a simulation on a mock-up website and evaluate the effect of pictures on information scent of hyperlinks by means of the CoLiDeS + Pic model on basis of the simulation results. It was found that CoLiDeS + Pic with highly relevant pictures increases the values of information scent of task-relevant hyperlinks, and therefore it increases the probability of selecting those hyperlinks compared to CoLiDeS (without pictures) or CoLiDeS + Pic with lowly relevant pictures. These results confirm the importance of including information from pictures into the modelling of web-navigation.

Metaphors based on perceptual similarity are considered to be one of the hallmarks of creativity. We propose here an image-based retrieval system for generating pairs of perceptually similar images. A number of examples of images paired by the system are presented and discussed. Often, these images are conceptually very different, and hence can serve as anchors for perceptual metaphors that create new conceptual similarities.

This paper proposes a perceptually motivated method for solving proportional analogy problems involving sequential patterns. Our method is based on an algebraic model of pattern perception that determines the gestalt structure of sequential patterns. The gestalts of sequential patterns are represented as algebraic terms in the model. An analogical relation between sequential patterns appearing in proportional analogy is then formalized as a mapping between algebras that generate terms representing the gestalts of these patterns. Based on this formalism, an algorithm is proposed that solves proportional analogy problems: given three sequential patterns, the algorithm computes a fourth pattern such that the resulting two pairs of patterns are perceived as having an identical analogical relation.

Following the trend in using computer-based learning tools to support articulate reflection, we report here our experiences on using an interactive simulator to help children learn about the role of decentralized control in social communication. A decentralized control system is a parallel system in which many local processes interact with the neighboring processes, and macro-characteristics emerge from it. In order to make children learn and appreciate the role of decentralized control in social communication we designed and implemented an interactive simulator, and experimented with using the simulator in a classroom with 6/sup th/ grade children (11-12years). Our results show that it is unreasonable to support articulate reflection from a computer-based learning system alone. Instead, a more integrated approach that combines traditional classroom teaching and discussion with the computer-based learning-systems is necessary.

In the paper we describe two cognitive experiments performed using the web and virtual reality to study the alignment effect. In particular, we investigated if the presence of a landmark has any influence in reducing the errors due to maps being displayed in a misaligned orientation. We conducted two psychological experiments to ascertain the validity of this approach was. In our experiments, the participants were asked to memorize their starting position and a destination on a map, and then were asked to navigate themselves to their destination in a three-dimensional virtual-reality simulation of the map. During the memorization phase, the maps were displayed in different orientations with respect to the starting position of the participant and the destination. The results of these experiments show that the destination orientation and self-orientation has a much more pronounced effect.

This paper proposes an approach to enhance dialogue capabilities between humans and the computer. In order for a computer system to sustain a natural conversation without constraining the topic, it must be able to generate a natural response to a wide variety of topics. In particular, a good general-purpose dialogue system must be able to infer the user's intention from an utterance and be able to respond to it appropriately. A case-based dialogue system could provide an approach to solve these problems because it can work as an empirical language learner, and can become a better hearer and speaker for the human user as the case database becomes larger during dialogues. The goal of this research is to design an efficient case-base for a dialogue system, which contains dialogues between humans or between a human and a computer. We define "dialogue act", which is a key to finding the user's intention and for searching appropriate cases in the case-base. We describe how to extract useful knowledge from example dialogues and how to choose an optimal case to generate an appropriate response.

In this paper we address the problem of how new categories and new interpretations can be created in legal reasoning.We present some examples of such creativity, and then analyze them to identify the mechanisms that must be modeled for incorporating creativity.We point out that many of these mechanisms can be implemented by using components of already existing systems.Based on this analysis, we outline our approach, which uses a blackboard style architecture to generate creative arguments.Top-down processes are activated by portions of rules or ratio decidendi of precedents.Bottom-up processes are activated by the facts of the current case and precedents.The retrieval of precedents is also modeled as a blackboard process, so that which precedents are examined is determined dynamically depending on the contents of the blackboard.We compare our approach with existing research and then briefly mention future research problems.

An investigation is made of an optimal task-assignment policy for a random-graph model of a distributed program. The model of the distributed computer system assumes that communications overhead adds to total run time and that total run time decreases as the number of processors running the program is increased. When the processors are homogeneous, the optimal task-assignments are external in the sense that tasks are totally distributed among all processors as evenly as possible or not distributed at all. The point at which the policy shows a sharp change of behavior depends upon the ratio of run times to communication times. The authors derive two important properties of the optimal task-assignments for heterogeneous processors. The first property is that an optimal policy distributes the cost of processing among the processors as evenly as possible so that a processor with higher speed gets more tasks and vice versa. The second property determines the number of processors among which to distribute the tasks evenly. In the special case when there is a uniform degradation of processing speed, it is shown that the optimal policy again exhibits an extremal characteristic.