Publikationen

Smartphone users are beyond two billion worldwide. Heavy users of the texting application rely on input prediction to reduce typing effort. In languages based on the Roman alphabet, many techniques are available. However, Japanese text is based on multiple character sets such as Kanji (Chinese-like word symbols), Hiragana and Katakana syllable sets. For its time/labor intensive input, next word prediction is crucial. It is still an open challenge. To tackle this, a hybrid language model is proposed. It integrates a Recurrent Neural Network (RNN) with an n-gram model. RNNs are powerful models for learning long sequences for next word prediction. N-gram models are best at current word completion. Our RNN language model (RNN-LM) predicts the next words. According the “price” of the performance gain paid by a higher time complexity, our model best deploys on a client-server architecture. Heavily-loaded RNN-LM deploys on the server while the n-gram model on the client. Our RNN-LM consists of an input layer equipped with word embedding, an output layer, and hidden layers connected with LSTMs (Long Short-Term Memories). Training is done via BPTT (Back Propagation Through Time). For robust training, BPTT is elaborated by learning rate refinement and gradient norm scaling. To avoid overfitting, the dropout technique is applied except for LSTM. Our novel model is compact (2 LSTMs, 650 units per layer), indeed. Due to synergetic elaboration, it shows 10 % lower perplexity than Zaremba's excellent conventional models in our Japanese text prediction experiment. Our model has been incorporated into IME (Input Method Editor) we call Flick. On the Japanese text input experiment, Flick outperforms Mozc (Google Japanese Input) by 16 % in time and 34 % in the number of keystrokes.

The authors’ key area of application is training for the prevention of accidents in the process technology industries. They run a professional training center with own 3D virtual environments. At TLIC 2021, four of the present authors delivered a contribution advocating planning of human training experiences as dynamically as managing some severely disturbed technical system back into a normal operation - such as an out of control chemical reactor - and enabling human trainees who failed to complete a risky task - thereby possibly ruining a (fortunately only virtual) technical installation - to virtually travel back in time to make good the damage. At TLIC 2022, they introduced cascades of gradually more intricate categories of time travel games. With every step from one category to the next, the deployed AI gets more powerful and effective in providing adaptive guidance of human trainees. The most advanced time travel games are those that allow for the dynamic modification of events experienced in the virtual past. In this way, the game system evolves over time and adapts to the needs of human trainees with emphasis on guidance for trainees who fail repeatedly. The extended team of authors presents a novel perspective at time travel prevention games that leads to a more human-centered adaptive guidance. Training is seen through the lens of deontic modal logic. The focus is on undesired events such as explosions, fire, health hazards due to toxic vapors, and the like. The game system’s AI is reasoning about necessity and possibility of such events. It offers to human trainees/players helpful chats about modalities of decisive events of training.

This paper describes and discusses methods for improving character depth and sentence diversification in automated storytelling systems. The fAIble III system that is the subject of this paper addresses a major limitation of its immediate predecessor (fAIble II) in that the characters in its stories seemed to act in a vacuum, without any apparent reasons for their choices or emotions. This is accomplished through generating character backstories. fAIble III also addresses the diversity of generated sentences with a pattern recognition system that removes many of the awkward and repetitive sentences that drew negative comments in the testing of fAIble I and II. Lastly, stories generated by fAIble II and fAIble III are compared and empirical test results are presented.

Cloud computing is currently receiving much attention from the industry, government, and academia. It has changed the way computation is performed and how services are delivered to customers. Most importantly, cloud services change the way we design software, handle data, and perform testing. In cloud computing, testing is delivered as a service (TaaS). Case testing is one of the most common validation approaches for software. However, executing test cases on a software system could be expensive and time consuming. Therefore, test case reduction is performed to minimize the number of test cases to be executed on the system. In this paper, we introduce a validation method called Context-Assisted Test Case Reduction (CATCR) for systems that are deployed on the cloud. In CATCR, test cases are reduced based on the context of the validation process. The results of previous test cases are used to select the next set of test cases while the validation process is ongoing. The minimized set of test cases needs to have effective coverage of the entire system. To evaluate CATCR, an experimental evaluation is performed through Amazon's Cloud and a Java validation tool. Experimental results are recorded and presented.

In Japan, many people suffer from bad conditions of their mental health with an increasing tendency that should be alarmed. For this reason, a counseling robot was created in our former work and enhanced more and more. Our former quantitative evaluation based on the utterance amount. Here, we provide a new solution on evaluating this system. We introduce an evaluation polarity of the utterance contents by using a polarity dictionary This evaluates the feeling polarity of utterances by positive through negative degree on a scale from -1 (negative) to +1 (positive). Combined with generally used conventional impression evaluation by a questionnaire, it provides a much better way to evaluate the performance of our Counseling Agent Robot.