Visual signals in text comprehension: How to restore them when oralizing a text via a speech synthesis?

It has been assumed theoretically and established empirically that text signals exert an influence on text memorization and comprehension. The study investigates whether the restoration of the text visual signals improve text memorization and comprehension when automatically converting a text into speech. Participants listened to a restaurant menu oralized by a text-to-speech synthesis. The visual signals used in the menu were restored either with discursive segments, with prosodic cues, or with a picture of text, displayed before or during the listening. Participants had to perform tasks assessing their text memorization and comprehension. The restoration of text visual signals exerts an influence on the participants’ recall but these effects vary according to the restoration mean used and to the task. When visual signals are not restored, individuals construct an erroneous representation of the situation described in the text leading to a misinterpretation of the text meaning, whereas the discursive and prosodic restorations involve the construction of an adequate representation.     

基于事件的文本表示方法研究

在传统文本表示模型的研究基础上,针对叙事类文本,考虑以事件作为基本语义单元,并结合图结构表示的 特点,提出了一种基于事件的文本表示方法—事件网络。该方法利用事件和事件间的关系来表示文本,能够较大程 度地保留文本的结构信息及语义信息。实验结果表明,基于该方法的自动摘要取得了较好的效果。     

文本层次分析与文本浏览

本文简要描述了文本的物理结构和逻辑结构以及相应的向量空间模型。研制了具有导航机制的文本浏览系统。提出了文本结构分析中的层次分析方法,它采用有序划分层次的方法。并在此基础上,给出了文本结构中各单元的标记信息,由此形成了文本的可视化表示。利用文本、层次、段落的超文本连接,根据浏览的需要,逐级展现文本细节,帮助用户有目的、有选择地浏览文本。最后给出评价的结果。     

面向文本分类的特征词选取方法研究与改进

中文特征词的选取是中文信息预处理内容之一,对文档分类有重要影响。中文分词处理后,采用特征词构建的向量模型表示文档时,导致特征词的稀疏性和高维性,从而影响文档分类的性能和精度。在分析、总结多种经典文本特征选取方法基础上,以文档频为主,实现文档集中的特征词频及其分布为修正的特征词选取方法(DC)。采用宏F值和微F值为评价指标,通过实验对比证明,该方法的特征选取效果好于经典文本特征选取方法。     

数字视频中文字分割算法的研究

文章对三种文字分割算法,包括基于阈值的二值化算法、基于分裂/合并的算法和基于纹理的算法,进行了分析、实验与评价,最后提出一种边缘检测——投影——局部区域二值化的文字分割算法,并给出与前面一些算法的比较。     

Reading Chinese text on a small screen with RSVP

This study examines the effects of Chinese typography (true type and standard Kai type), speed (171, 260, and 350 wpm), font size (12- and 28-point), and presentation method (serial one word, phrase-oriented, and sentence-by-sentence method) on users' reading comprehension for reading Chinese text on a small screen using rapid serial visual presentation (RSVP). Scores for reading comprehension indicated that typography style and font size did not significantly affect reading comprehension. Speed and presentation method, however, were significant factors in terms of the reading comprehension. In addition, the interaction between the speed and the presentation method also significantly affected the reading comprehension. When the presentation methods were serial one word and sentence-by-sentence formats, reading comprehension revealed no significant difference under different speed levels. Nonetheless, when the presentation method was in the phrase-oriented format, reading comprehension with a lower speed of 171 wpm was significantly higher than that with the speeds of 350 and 260 wpm.     

结合自注意力机制的神经网络文本分类算法研究

在长文本数据中存在很多与主题不相关词汇,导致这些文本数据具有信息容量大、特征表征不突出等特点。增加这些文本中关键词汇的特征影响,是提高文本分类器性能需要解决的问题。提出一种结合自注意力机制的循环卷积神经网络文本分类模型RCNN_A。注意力机制对文本词向量计算其对正确分类类别的贡献度,得到注意力矩阵,将注意力矩阵和词向量矩阵相结合作为后续结构的输入。实验结果表明,RCNN_A在10类搜狗新闻数据集上,得到了97.35%的分类正确率,比Bi-LSTM(94.75%)、Bi-GRU(94.25%)、TextCNN(93.31%)、RCNN(95.75%)具有更好的文本分类表现。通过在深度神经网络模型中引入注意力机制,能够有效提升文本分类器性能。     

Video text detection and segmentation for optical character recognition

In this paper, we present approaches to detecting and segmenting text in videos. The proposed video-text-detection technique is capable of adaptively applying appropriate operators for video frames of different modalities by classifying the background complexities. Effective operators such as the repeated shifting operations are applied for the noise removal of images with high edge density. Meanwhile, a text-enhancement technique is used to highlight the text regions of low-contrast images. A coarse-to-fine projection technique is then employed to extract text lines from video frames. Experimental results indicate that the proposed text-detection approach is superior to the machine-learning-based (such as SVM and neural network), multiresolution-based, and DCT-based approaches in terms of detection and false-alarm rates. Besides text detection, a technique for text segmentation is also proposed based on adaptive thresholding. A commercial OCR package is then used to recognize the segmented foreground text. A satisfactory character-recognition rate is reported in our experiments.Published online: 14 December 2004     

Visual design guidelines for improving learning from dynamic and interactive digital text

Despite the dynamic and interactive features of digital text, the visual design guidelines for digital text are similar to those for printed text. The purpose of this study was to develop visual design guidelines for improving learning from dynamic and interactive digital text and to validate them by controlled testing. Two structure design guidelines (for enhancing text structure comprehension) and two selective-attention design guidelines (for maintaining the learners' attention on the essential contents) were developed based on the psychological and instructional, technological foundations that can affect the visual design of digital text. In this study, a 2 × 2 factorial design with 141 university students was used to examine the effectiveness of the visual design guidelines. The university students had 20 min to study a piece of digital text with the structure design guidelines, selective-attention design guidelines, both, or no design guidelines applied. Both the structure and selective-attention design guidelines had a positive influence on text structure understanding, essential contents comprehension and usability of digital text. The suggested visual design guidelines were found to be useful for enhancing text comprehension.