Most artificial intelligence agents in interactive fiction games are implemented using reinforcement learning. Considering the recent rapid development of large language models, we propose an approach that utilizes a large language model to tackle interactive fiction game tasks. The chosen test dataset is TextWorld Commonsense, an interactive fiction game environment designed for artificial intelligence agents. In these games, the AI agent’s task is to organize rooms and place items in appropriate locations. To achieve a high score in the game, common sense knowledge about “which items belong to which locations” is important. Our approach is based on GPT-4 and a carefully designed prompt. Experimental results demonstrate that our approach outperforms prior research. Specifically, GPT-4 with feedback-augmented prompt successfully completed all tasks in both simple and medium level game environments without fine-tuning. In hard level game environments, our approach achieved a normalized score of 0.70, surpassing the best baseline score of 0.57.

In Japan, the number of single-person households, particularly among the elderly, is increasing. Consequently, opportunities for people to narrate are being reduced. To address this issue, conversational agents, e.g., communication robots and smart speakers, are expected to play the role of the listener. To realize these agents, this paper describes the collection of conversational responses by listeners that demonstrate attentive listening attitudes toward narrative speakers, and a method to annotate existing narrative speech with responsive utterances is proposed. To summarize, 148,962 responsive utterances by 11 listeners were collected in a narrative corpus comprising 13,234 utterance units. The collected responsive utterances were analyzed in terms of response frequency, diversity, coverage, and naturalness. These results demonstrated that diverse and natural responsive utterances were collected by the proposed method in an efficient and comprehensive manner. To demonstrate the practical use of the collected responsive utterances, an experiment was conducted, in which response generation timings were detected in narratives.

Nowadays, spoken dialogue agents such as communication robots and smart speakers listen to narratives of humans. In order for such an agent to be recognized as a listener of narratives and convey the attitude of attentive listening, it is necessary to generate responsive utterances. Moreover, responsive utterances can express empathy to narratives and showing an appropriate degree of empathy to narratives is significant for enhancing speaker’s motivation. The degree of empathy shown by responsive utterances is thought to depend on their type. However, the relation between responsive utterances and degrees of the empathy has not been explored yet. This paper describes the classification of responsive utterances based on the degree of empathy in order to explain that relation. In this research, responsive utterances are classified into five levels based on the effect of utterances and literature on attentive listening. Quantitative evaluations using 37,995 responsive utterances showed the appropriateness of the proposed classification.

With the development of speech and language processing, speech translation systems have been developed. These studies target spoken dialogues, and employ consecutive interpretation, which uses a sentence as the translation unit. On the other hand, there exist a few researches about simultaneous interpreting, and recently, the language resources for promoting simultaneous interpreting research, such as the publication of an analytical large-scale corpus, has been prepared. For the future, it is necessary to make the corpora more practical toward realization of a simultaneous interpreting system. In this paper, we describe the construction of a bilingual corpus which can be used for simultaneous lecture interpreting research. Simultaneous lecture interpreting systems are required to recognize translation units in the middle of a sentence, and generate its translation at the proper timing. We constructed the bilingual lecture corpus by the following steps. First, we segmented sentences in the lecture data into semantically meaningful units for the simultaneous interpreting. And then, we assigned the translations to these units from the viewpoint of the simultaneous interpreting. In addition, we investigated the possibility of automatically detecting the simultaneous interpreting timing from our corpus.

One of the essential factors in community sites is anonymous submission. This is because anonymity gives users chances to submit messages (questions, problems, answers, opinions, etc.) without regard to shame and reputation. However, some users abuse the anonymity and disrupt communications in a community site. These users and their submissions discourage other users, keep them from retrieving good communication records, and decrease the credibility of the communication site. To solve this problem, we conducted an experimental study to detect submitters suspected of pretending to be someone else to manipulate communications in a community site by using machine learning techniques. In this study, we used messages in the data of Yahoo! chiebukuro for data training and examination.

As a first step to developing systems that enable non-native speakers to output near-perfect English sentences for given mixed English-Japanese sentences, we propose new approaches for selecting English equivalents by using the number of hits for various contexts in large English corpora. As the large English corpora, we not only used the huge amounts of Web data but also the manually compiled large, high-quality English corpora. Using high-quality corpora enables us to accurately select equivalents, and using huge amounts of Web data enables us to resolve the problem of the shortage of hits that normally occurs when using only high-quality corpora. The types and lengths of contexts used to select equivalents are variable and optimally determined according to the number of hits in the corpora, so that performance can be further refined. Computer experiments showed that the precision of our methods was much higher than that of the existing methods for equivalent selection.

Japanese adverbs are classified as either declarative or normal; the former declare the communicative intention of the speaker, while the latter convey a manner of action, a quantity, or a degree by which the adverb modifies the verb or adjective that it accompanies. We have automatically classified adverbs as either declarative or not declarative using a machine-learning method such as the maximum entropy method. We defined adverbs having positive or negative connotations as the positive data. We classified adverbs in the EDR dictionary and IPADIC used by Chasen using this result and built an adverb dictionary that contains descriptions of the communicative intentions of the speaker.

We describe a new model for dependency structure analysis. This model learns the relationship between two phrasal units called bunsetsus as three categories; ‘between’, ‘dependent’, and ‘beyond’, and estimates the dependency likelihood by considering not only the relationship between two bunsetsus but also the relationship between the left bunsetsu and all of the bunsetsus to its right. We implemented this model based on the maximum entropy model. When using the Kyoto University corpus, the dependency accuracy of our model was 88%, which is about 1% higher than that of the conventional model using exactly the same features.