INTEGRATING QUALITATIVE REASONING AND TEXT PLANNING TO GENERATE CASUAL EXPLANATIONS I N T E G R A T I N G Q U A L I T A T I V E R E A S O N I N G A N D T E X T P L A N N I N G T O G E N E R A T E C A[.]
The growing success of argumentation-based approaches has caused a rich crossbreeding with other disciplines, providing interesting results in different areas such as legal reasoning, medical diagnosis and decision support systems. Many of these approaches rely on quantita- tive aspects (such as numeric attributes, probabilities or certainty values). As argumentation provides mostly a non-numerical, qualitative setting for commonsense reasoning, integrat- ing both quantitative and qualitative features has shown to be highly desirable [TP01]. Remarkably, numerical reasoning has been long neglected in the defeasible argumentation community. This is maybe due to the historical origins of the discipline, which were more related to legal (qualitative) reasoning rather than to number-based attributes as those used in rule-based production systems.
In this paper, we present a novel approach to physical modelling in heat transfer analysis which aims to address many of the issues raised in the first paragraph including: What is the nature of modelling in PDE analysis? How do engineers carry out modelling and how does this influence our approach? What do engineers require from modelling systems? What type of tools assist engineers best with the model generation task? Our examination of these questions has led us to view model generation as an iterative design task that uses both experiential and model-based knowledge. Consequently we have developed a physical modelling system called CoBRA which exploits both model-based and case-based reasoning techniques within a derivational analogy framework. We argue that this approach has a number of advantages over other work including; cognitive plausibility, computational tractability, ease of knowledge acquisition and a more pragmatic engineering approach to model generation. Finally, we believe that it addresses some of concerns raised by researchers from the qualitativereasoning community about the need to firstly, focus more clearly on significant engineering problems, and secondly, to tackle these problems in a manner that is beneficial to the engineering community .
When artificial intelligence (AI) techniques are used in the development or assessment of alternatives, the resulting systems are referred to as intelligent decision support systems. These techniques attempt to understand and explain the skill of human beings in reasoning without precise knowledge (Doumpos and Grigoroudis, 2013). Qualitativereasoning (QR) techniques and fuzzy systems, which are considered subfields of research in AI, offer systematic tools for criteria assessment. Frequently, this uncertainty is captured by using linguistic terms or fuzzy numbers to evaluate the set of criteria or indicators. In different studies, fuzzy MCDM approaches have been developed to help energy planners and policy makers. In fact, fuzzy and qualitativereasoning techniques are capable of representing uncertainty, emulating skilled humans, and handling vague situations (Dubois and Prade, 1980; Tuzkaya et al., 2009). Application of the fuzzy set theory, established by Zadeh (1965), plays an important role in overcoming uncertainties. Qualitative absolute order-of-magnitude models were introduced into the qualitativereasoning field with the aim of using a linguistic approach to work with different levels of precision (Travé- Massuyès et al. 2005).
The archaeologists interact with 3D underwater surveys through queries to an OWL2 knowledge base consisting of an ontology combining underwater archaeological knowledge and photogrammetric measurement knowledge, and a set of observations on underwater archaeological sites . The web- based tools proposed in  allow the users to perform advanced SparQL  queries to the OWL2 knowledge base concerning typological and photogrammetric features. Furthermore, in order to express spatial queries, rules in SWRL  have been added to the OWL2 knowledge base . However, performing qualitativereasoning leads to queries stemming from non-monotonic rules that cannot be expressed in SPARQL nor in SQWRL (SWRL-based query language).
of work by Gardin and Meltzer [Gardin and Meltzcr, 1989], both take a very different approach to reasoning about liquid from those based on symbolic qualitativereasoning. Instead of representing a body of liquid or a solid object as one entity as is usually done in symbolic reasoning systems, they represent both types of things as a one- and two dimensional collection of particles. Each particle represents a small piece of liquid or solid stuff. They use a two-dimensional array to represent the position of each piece, and simulate the movement of each piece to predict the behavior of the collection. For simulation of the movement, Decuyper et al. apply physics laws to each cell, while Gardin and Meltzer use local rules, which Govern the exchange of messages between neighboring particles. By changing the rules restricting the permissible angle between particles, Gardin and Meltzer can also simulate the behavior of solid objects, such as rods and rings, of different flexibility. As with Forbus' Metric Diagram, their analogical representation is not intended u> be a
Following a detailed study, a frame work has been designed to implement the intelligent system for diagnosis of diseases. There have been many approaches in implementing an expert system for diagnosis of various diseases. The first and foremost one is using the rule-based reasoning technique. This is based on the rules and facts stored in the database. The problem with this approach was that the rate of diagnosis was quite low. The disease could be diagnosed only if the rules could be fired. Adaptation to the change in the course of diagnosis according to the cases already reported was not there. Then the case-based reasoning techniques were used for diagnosis of diseases like Diabetes . Integrating the two reasoning systems, a reasoning method named the hybrid reasoning system was used in a number applications including medical diagnosis of various diseases , , , . The hybrid system is designed by encompassing the Rule-based system and Case-based system. The hybrid reasoning system has been added to the framework to make the diagnosis much more accurate than when there is only one reasoning system. The rate of diagnosis increases as the CBR module takes care of cases which cannot be solved by the RBR module. There are different modules for arriving at a diagnosis and for suggesting an appropriate treatment for diseases
Our second contribution is the proposal and analysis of two orthogonal approaches to automatically deciding satisfiability in multi-context systems. By providing a tractable encoding of contextual satisfiability problems into purely propositional ones, a solid groundwork has been laid for Sat -based implementations of contextual reasoning systems. On the other hand, we have also proposed a distributed algorithm, called ContextSat , which seeks to exploit the potential benefit of localizing reasoning and restricting it to relevant contexts only. ContextSat has been shown to be more efficient, in general, than our translation based procedure, and to be implementable using off-the-shelf efficient reasoning platforms, such as bdd s and propositional Sat solvers. Moreover, ContextSat has been designed to suit a possible distributed peer-to-peer implementation. It is modular, i.e. global reasoning is made up of local reasoning procedures, and it is backtrack-free, i.e. solutions are build - or rather confined - incrementally, imposing a minimal restriction at each step. These features support a natural implementation in a peer- to-peer architecture, in which peers perform local reasoning and propagate their conclusions to neighbor peers via bridge rules. Modularity supports local reasoning, while backtrack-freeness avoids infinite loops.
Thus, reasoning is a paramount cognitive ability, which is not innate in students (Ates & Cataloglu, 2007). Before prospective scientists can use this kind of reasoning as a thinking tool, they must pass through several steps. They think about descriptions of the world and build up small concepts and schemes. The more they find out about the relations of the concepts, the larger the schemes get. These schemes, when sufficiently developed, enable hypothetical and scientific reasoning. In consequence, students develop a better ability for doing science and the construction of science concepts (Lawson, Clark, Meldrum, Falconer, Sequist, & Kwon, 2000). Hence, Lawson et al. (2000) found a positive relationship between reasoning skills and achievement in education.
Most teachers who participated in the study had experienced a traditional education in statistics, emphasising procedural rather than conceptual competence, and had typically not been exposed to activities that could build their understanding of and reasoning about repeated sampling (Wessels & Nieuwoudt, 2011). Research about variability in a repeated sampling environment that included an intervention (Canada, 2006) has shown an improvement of participants’ descriptions of what they expected (description) as well as of reasons for their expectations (causality). After this intervention, participants increasingly appreciated how variation occurred in multiple trials whilst reasons for their expectations improved, and progressively emphasised proportional reasoning coupled with a realisation of what is likely in the presence of variation. The results of our study concur with the results reported by Canada. Canada emphasises that to be effective, teacher education programs need to include an environment where teachers ‘can learn in a similar way that they themselves will aim to teach’ (p. 61). Teachers must get the opportunity to draw real samples and discuss differences and similarities in the distributions to develop their own skills about this topic so as to enable them to facilitate the development of proportional reasoning skills and a sense for expected variability in learners. The value of proportional thinking in repeated sampling situations also needs to be made explicit (Shaughnessy et al., 2004; Watson & Shaughnessy, 2004).
In addition to discovering laws that directly en- able predictions, XPERO aspires to advance the understanding of the mechanisms that enable new insights. In XPERO, an “insight” means something conceptually more general than a law. One possible definition of an insight in the spirit of XPERO is the following: an insight is a new piece of knowledge that makes it possi- ble to simplify the current agent’s theory about its environment. Examples of insights are the discoveries of notions like absolute coordinate system, arithmetic operations, notion of grav- ity, notion of support between objects, etc. An insight would thus ideally be a new concept that makes the current domain theory more flexi- ble and enables more efficient reasoning about the domain. Thus, insights should also make further learning in this domain easier and more effective.
Arocha suggested that diagnosis occurs through rapid non-analytical matching of clinical presentation with a pattern previously formed of constructs of clinical signs and symptoms (or pattern) in memory (18-20). He believes that the retrieval of these pattern is triggered by recognition of key features within the case (18). New cases are classified according to their resemblance to clinical patterns (19, 20). The idea of pattern recognition was further strengthened with the observation that the clinical reasoning of medical experts in familiar situations does not frequently include hypotheses testing (19-22). A question may be raised as to the nature of the pattern stored in the memory against which new cases are categorized. Two answers to this question were supported by research evidence. A new case is matched against a specific instance or “exemplar” or to a more abstract construct or “prototype”. In an exemplar-based recognition, a new case is matched by its resemblance to memories of specific instances previously encountered (19, 22). The observation that diagnosis is strongly influenced by the context even when the context is irrelevant (for example the age and occupation of a patient with a left bundle branch block on an ECG) supported the notion of exemplar (20). In the prototype-based recognition, a new case is matched to a prototype (23, 24). The basis of prototype model is that an abstraction process is used to construct a common memory trace of two or more patients, seen with the same disease or set of signs and symptoms (25).
What are the lessons that the assessment of scientific reasoning can learn from this approach? First, it might be worthwhile to conceptualize scientific reasoning skills in a network model instead of a latent variable model. It is possible to imagine, for instance, that hypothesis generation influences evidence generation, which in turn influences evidence evaluation and drawing conclusions, and finally a feedback process might occur in which a conclusion leads to hypothesis generation again. Such a conceptualization seems not too unreasonable since it would be similar to the already used problem-solving approaches of scientific reasoning. Second, the idea to include a high number of scientific reasoning skills from different assessments into the network and to give special attention to central skills instead of simply trusting a sum score would fit the results about item-specific factors from Study 2. Third, there are constructs that are conceptually close to scientific reasoning, like NOS, and it would be possible to include items from these constructs into the network to see how distinct the constructs really are. Fourth, a network model, if it can be established, might be useful for the training of scientific reasoning. It would provide recommendations about which central skill should be trained first, since they would activate many other skills and thus increase the training efficiency. The last noteworthy aspect of network models is that they have proven to be useful in more than one field already: Researchers also already used similar approaches to model intelligence (van der Maas et al., 2006).
Pseudocode 1 shows the implementation of analogical reasoning in our knowledge- based system. As mentioned, this process is driven by failed justification premises. In this case, the process starts by looking for plausible rules that can resolve a missing premise (line 1). At this step, the concept hierarchy is leveraged to expand the search space. In this example, the direct superclass of the missing premise firmLiver (i.e., (type LiverIssue)) will match the consequent of the plausible rule from Code 5. The knowledge base is then searched for facts that unify the rule; i.e., match the rule condition and consequent (line 2). For each found fact, the algorithm checks whether its instantiated knowledge-transfer condition matches the original entity (line 3). If so, its consequent is added to the entity, at least in the system’s working memory (line 4). If the expert confirms the overall justifi- cation, this working memory will be materialized in the knowledge base (see section Implementation of SeMS-KBS).
Research about emotions in the workplace has increased rapidly over the last twenty years or so (see Grandey 2007 for a review), but little research has examined emotions, “an understudied yet powerful contributor to the decision-making process” (Monin et al. 2007, 99; see also Haidt 2007, 2010; Narvaez, 2010). Research about emotions affecting auditors’ judgments has received even less attention since the accounting profession promotes objectivity as a fundamental principle of the profession (Chua 1986; Hines 1991; IFAC, 2005). Moreover, the accounting profession overtly values objectivity and representational faithfulness to service corporate clients who are steeped in instrumental rationality (McPhail 2004). Little research has examined the notion of emotions affecting professional judgment although other auditors’ attributes such as moral reasoning has been extensively investigated (see Bailey et al. 2010). We argue that emotions can motivate moral reasoning particularly when auditors’ decisional processes are first sensitized by a highly intense moral situation.
Many modern machine learning approaches require vast amounts of training data to learn new concepts; conversely, human learning often requires few examples—sometimes only one—from which the learner can abstract structural con- cepts. We present a novel approach to introducing new spa- tial structures to an AI agent, combining deep learning over qualitative spatial relations with various heuristic search al- gorithms. The agent extracts spatial relations from a sparse set of noisy examples of block-based structures, and trains convolutional and sequential models of those relation sets. To create novel examples of similar structures, the agent be- gins placing blocks on a virtual table, uses a CNN to pre- dict the most similar complete example structure after each placement, an LSTM to predict the most likely set of remain- ing moves needed to complete it, and recommends one using heuristic search. We verify that the agent learned the concept by observing its virtual block-building activities, wherein it ranks each potential subsequent action toward building its learned concept. We empirically assess this approach with human participants’ ratings of the block structures. Initial re- sults and qualitative evaluations of structures generated by the trained agent show where it has generalized concepts from the training data, which heuristics perform best within the search space, and how we might improve learning and execution.
The Psychology of Reasoning had influenced AI in its effort to automate human reasoning. From the early stages of AI an approach based on production rules was developed, influenced by the psychological findings on the nature of implications in human reasoning. Cognitive architectures [1, 30] for systems were proposed whose baseline computation is given by the application of production rules and following onto their conclusions when these were drawn. Despite their relative success (and their re-emergence today in the new era of Cognitive Computing that we will examine in Section 4) these systems were considered to be lacking a proper formal foundation. For example, how was the firing of a production rule to be interpreted? On the one hand when its conditions hold it must fire to give its conclusion and yet the conclusion could be at odds with conclusions of other production rules that have also fired. Attempts to provide formal foundations for these rules have been made .
lable but purchasing Raven SPM scoring sheets from international suppliers made the offering of this test to clients expensive. Thus, the Hirebright team decided to develop their own executive func- tion test, the H-CAT (Hirebright Cognitive Ability Test) to reduce the operational costs of the evaluations and thereby making their pre-employment services more affordable to the South African market. The H-CAT team consisted of the second author, who is an occupational therapist with a Masters Degree in Occupational Therapy in Vocational Rehabilitation, an actuary and an IT develo- per who managed the development of the on-line test application. The objectives of the H-CAT measuring instrument development were to measure logical, verbal and abstract reasoning in a working population to establish levels of general cognitive ability.