Recognizing lexical entailment (RLE) always plays an important role in inference of natural language, i.e., identifying whether one word entails another, for example, fox entails animal. In the literature, automatically recognizing lexical entailment for word pairs deeply relies on words' contextual representations. However, as a "prototype" vector, a single representation cannot reveal multifaceted aspects of the words due to their homonymy and polysemy. In this paper, we propose a supervised Context-Enriched Neural Network (CENN) method for recognizing lexical entailment. To be specific, we first utilize multiple embedding vectors from different contexts to represent the input word pairs. Then, through different combination methods and attention mechanism, we integrate different embedding vectors and optimize their weights to predict whether there are entailment relations in word pairs. Moreover, our proposed framework is flexible and open to handle different word contexts and entailment perspectives in the text corpus. Extensive experiments on five datasets show that our approach significantly improves the performance of automatic RLE in comparison with several state-of-the-art methods.

From Young 2010), frames (Zhu and Ontañón 2014), plotpoints early games like Zork, to the text-based interactive Victorian (Weyhrauch and Bates 1997; Nelson and Mateas 2005; dramas generated by Versu (Evans and Short 2014) Sharma et al. 2010) or social models (McCoy et al. 2011), to 3D RPG games like Skyrim (Ruch 2011), the quality of the problem of how to computationally model narratives the stories play a crucial role in engaging the player and and story spaces remains open.

This paper presents an approach to automatically extracting and representing narrative information from stories written in natural language. Specifically, we present our results in extracting story graphs, a formalism that captures the entities (e.g., characters, props, locations) and their interactions in a story. The long-term goal of this research is to automatically extract this narrative information in order to use it in computational narrative systems such as story generators or interactive fiction systems. Our approach combines narrative domain knowledge and off-the-shelf natural language processing (NLP) tools into a machine learning framework to build story graphs by automatically identifying entities, actions, and narrative roles. We report the performance of our fully automated system in a corpus of 21 stories and provide examples of the extracted story graphs and their uses in computational narrative systems.

Prior work on viewing plan recognition as parsing of grammars has assumed completely observable actions. This paper provides an algorithm to rewrite plan grammars to allow for recognizing partially observable actions.  For the ELEXIR (Geib 2009) system, the impact of this rewriting on plan recognition runtime is shown to be limited to those plans that actually use the partially observable actions.

Applying game-tree search techniques to RTS games poses a significant challenge, given the large branching factors involved. This paper studies an approach to incorporate knowledge learned offline from game replays to guide the search process. Specifically, we propose to learn Naive Bayesian models predicting the probability of action execution in different game states, and use them to inform the search process of Monte Carlo Tree Search. We evaluate the effect of incorporating these models into several Multiarmed Bandit policies for MCTS in the context of StarCraft, showing a significant improvement in gameplay performance.

Statistical models, such as Markov Chains, have been recently studied in the context of procedural content generation (PCG). These models can capture statistical regularities of a set of training data and use them to sample new content. However, these techniques assume the existence of sufficient training data with which to train the models. In this paper we study the setting in which we might not have enough training data from the target domain, but we have ample training data from another, similar domain. We propose an algorithm to discover a mapping between domains, so that out-of-domain training data can be used to train the statistical model. Specifically, we apply this to two-dimensional level generation, and experiment with three classic video games: Super Mario Bros., Kid Icarus and Kid Kool.

Recent work has shown that incorporating action probability models (models that given a game state can predict the probability with which an expert will play each move) into MCTS can lead to significant performance improvements in a variety of adversarial games, including RTS games. This paper presents a collection of experiments aimed at understanding the relation between the amount of training data, the predictive performance of the action models, the effect of these models in the branching factor of the game and the resulting performance gains in MCTS. Experiments are carried out in the context of the microRTS simulator, showing that more accurate predictive models do not necessarily result in better MCTS performance.

This paper presents a new terrain analysis algorithm for RTS games. The proposed algorithms significantly improves the analysis time of the state of the art via contour tracing, and also offers better chokepoint detection. We demonstrate that our approach (BWTA2) is at least 10 times faster than the commonly used BWTA in a collection of StarCraft maps. Additionally, we show the usefulness of terrain analysis in tasks such as pathfinding and discuss potential applications to strategic decision making tasks.

Computational narrative systems usually require knowledge about the story world and narrative theory to be encoded in some form of structured knowledge representation formalism, a notoriously time-consuming task requiring expertise in both storytelling and knowledge engineering. In this paper we present an approach that combines supervised machine learning with narrative domain knowledge toward automatically extracting such knowledge from natural language stories, focusing specifically on predicting Proppian narrative functions. Our experiments on a dataset of Russian fairy tales show that our system outperforms an informed baseline and that combining top-down narrative theory and bottom-up statistical models inferred from an annotated dataset increases prediction accuracy with respect to using them in isolation.

This paper presents a new model of cooperative behavior based on the interaction of plan recognition and automated planning.  Based on observations of the actions of an "initiator" agent, a  "supporter" agent uses plan recognition to hypothesize the plans  and goals of the initiator.  The supporter agent then proposes and  plans for a set of subgoals it will achieve to help the initiator.  The approach is demonstrated in an open-source, virtual robot  platform.