人工知能学会論文誌
検索
OR
閲覧
検索
28 巻 , 4 号
選択された号の論文の5件中1~5を表示しています
    • |<
    • <
    • 1
    • >
    • >|
一般論文
原著論文
  • 高野 渉, 中村 仁彦
    28 巻 (2013) 4 号 p. 361-369
    公開日: 2013/05/10
    ジャーナル フリー
    Language is a symbolic system unique to human being. Human understands a real world through this symbolic system, makes inference, and communicates with others. Language underlies human intelligence. This paper describes a novel approach to combining motion primitives with words, and extracting relations among the words based on the motion primitives. The motion primitives are symbolized by Hidden Markov Models (HMMs). The HMM is hereafter referred to as ``motion symbol". Observation of human whole body motions is automatically segmented into motion primitives, recognized by using the motion symbols and converted to a sequence of the motion symbols. A sequence of words is also manually assigned to the observation. The association between motion symbols and words can be derived from pairs of these sequences as probability parameters, and dissimilarities among the words can be extracted. Words are located in a multidimensional space so that distances between the words in the space can become as close as possible to the dissimilarities. Thus, ``language space" is formed. The mapping of motion primitives onto the language space enables robots to understand human behaviors as words. The association between motion symbols and words can be also applied to generation of motion primitives from words. The validities of our proposed methods are demonstrated on a motion capture dataset.
    抄録全体を表示
  • 木村 泰知, 関根 聡
    28 巻 (2013) 4 号 p. 370-378
    公開日: 2013/05/23
    ジャーナル フリー
    This paper presents an information extraction method of political problems in minutes of local council which include councilor's utterances impartially. We focus on lexical heads of noun phrases in order to extract political problems. In this paper, the lexical heads are limited to noun. Our method is divided into two steps as follows: First step, our system classifies whether a phrase of lexical head includes political problems or not. Second step, our system find the beginning position of phrase which was classified as a lexical head of political problems. We confirmed that effectiveness of our method in the evaluation experiment.
    抄録全体を表示
  • 原 一夫, 鈴木 郁美, 新保 仁, 松本 裕治
    28 巻 (2013) 4 号 p. 379-390
    公開日: 2013/05/23
    ジャーナル フリー
    We propose a new measure of semantic similarity between words in context, which exploits the syntactic/semantic structure of the context surrounding each target word. For a given pair of target words and their sentential contexts, labeled directed graphs are made from the output of a semantic parser on these sentences. Nodes in these graphs represent words in the sentences, and labeled edges represent syntactic/semantic relations between them. The similarity between the target words is then computed as the sum of the similarity of walks starting from the target words (nodes) in the two graphs. The proposed measure is tested on word sense disambiguation and paraphrase ranking tasks, and the results are promising: The proposed measure outperforms existing methods which completely ignore or do not fully exploit syntactic/semantic structural co-occurrences between a target word and its neighbors.
    抄録全体を表示
  • 榎堀 優, 間瀬 健二
    28 巻 (2013) 4 号 p. 391-399
    公開日: 2013/05/23
    ジャーナル フリー
    On-demand, skill-level self-checks are required to establish effective training, and this is the same for metal-filing. However, such a training system has not been established yet for metal-filing because the current skill-level check depends on subjective skill-level estimation by experts. Such subjective skill-level estimation is composed of various complex viewpoints, and its estimation mechanisms cannot be represented in language because they are generated and supported by the experience of experts. That is why sensor-based, on-demand systems cannot faithfully imitate or replace the current skill-level checks of metal-filing. To solve this problem, we analyze the relationships among the subjective skill-level estimation mechanisms of experts and metal-filing mechanics structures. Our analysis yielded three simple viewpoints and related measures: class 2 lever-like movement measure (L), upper body rigidity measure (R), and pre-acceleration measure (A). Surveys of experts also yielded another viewpoint and a related measure: stability measure (S). These four measures successfully reproduced the subjective skill-level estimation of experts (adjusted-R2 = 0.90, p < 0.1, N = 10). The coefficients for the measures, which suggest that A is the main factor of the subjective skill-level estimation of experts, also suggest that effective training must emphasize these points in this order: A > L > R. S's coefficient suggests that the skill-level scores of experts are reduced by 69% when the filings of learners fail. In addition, since these four measures can be calculated with three small wearable hybrid sensors, they can be implemented on scalable wearable sensor-based skill-training systems. In future works, we will implement one such system, integrate it into a skill-training center's teaching plan, and assess how much the system improved the learning speeds of the students.
    抄録全体を表示
  • 小嵜 耕平, 新保 仁, 小町 守, 松本 裕治
    28 巻 (2013) 4 号 p. 400-408
    公開日: 2013/06/20
    ジャーナル フリー
    Graph construction is an important step in graph-based semi-supervised classification. While the k-nearest neighbor graphs have been the de facto standard method of graph construction, this paper advocates using the less well-known mutual k-nearest neighbor graphs for high-dimensional natural language data. To evaluate the quality of the graphs apart from classification algortihms, we measure the assortativity of graphs. In addition, to compare the performance of these two graph construction methods, we run semi-supervised classification methods on both graphs in word sense disambiguation and document classification tasks. The experimental results show that the mutual k-nearest neighbor graphs, if combined with maximum spanning trees, consistently outperform the k-nearest neighbor graphs. We attribute better performance of the mutual k-nearest neighbor graph to its being more resistive to making hub vertices. The mutual k-nearest neighbor graphs also perform equally well or even better in comparison to the state-of-the-art b-matching graph construction, despite their lower computational complexity.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top