[论文解读] Frequency of Occurrence and Information Entropy of American Sign Language

Andrew Chong, Lalitha Sankar+1 more|ArXiv.org|Dec 9, 2009

Hand Gesture Recognition Systems参考文献 15被引用 5

一句话总结

本文通过经验频率分析，量化了美国手语（ASL）手部姿势的信息熵和冗余度——这些是ASL的基本交流单元。研究发现，ASL手部姿势的冗余度显著低于英语音素，信息熵更高，这解释了为何尽管ASL的生成速度较慢，其信息传递速率仍可与口语相媲美。

ABSTRACT

American Sign Language (ASL) uses a series of hand based gestures as a replacement for words to allow the deaf to communicate. Previous work has shown that although it takes longer to make signs than to say the equivalent words, on average sentences can be completed in about the same time. This leaves unresolved, however, precisely why that should be the case. This paper reports a determination of the empirical entropy and redundancy in the set of handshapes of ASL. In this context, the entropy refers to the average information content in a unit of data. It is found that the handshapes, as fundamental units of ASL, are less redundant than phonemes, the equivalent fundamental units of spoken English, and that their entropy is much closer to the maximum possible information content. This explains why the slower signs can produce sentences in the same time as speaking; the low redundancy compensates for the slow rate of sign production. In addition to this precise quantification, this work is also novel in its approach towards quantifying an aspect of the ASL alphabet. Unlike spoken and written languages, frequency analysis of ASL is difficult due to the fact that every sign is composed of phonemes that are created through a combination of manual and a relatively large and imprecise set of bodily features. Focusing on handshapes as the ubiquitous and universal feature of all sign languages permits a precise quantitative analysis. As interest in visual electronic communication explodes within the deaf community, this work also paves the way for more precise automated sign recognition and synthesis.

研究动机与目标

• 量化美国手语（ASL）手部姿势作为基本交流单元的信息熵与冗余度。
• 解决一个悖论：为何尽管手语生成速度较慢，其句子完成时间却与口语相当。
• 开发一种精确、定量的ASL分析方法，克服其多模态与非线性结构带来的挑战。
• 通过建立ASL信息内容的数据驱动基础，实现更精确的自动手语识别与合成。
• 为手语提供一种新颖的频率分析框架，聚焦于手部姿势作为普遍且可分析的组成部分。

提出的方法

• 收集并分析了大量ASL手势数据，以确定每种独特手部姿势的实际出现频率。
• 使用标准香农熵公式计算ASL手部姿势的信息熵：H = -Σ p_i log₂ p_i，其中p_i为姿势i的概率。
• 将ASL手部姿势的熵与英语音素的熵和冗余度进行比较，以评估其相对信息含量。
• 将冗余度定义为 R = 1 - (H / H_max)，其中H_max为给定字母表大小下的最大可能熵，以量化信息效率。
• 采用统计分析确保频率估计与熵计算在数据集中的稳健性。
• 以手部姿势为核心单位构建分析框架，从而实现一种一致且可扩展的定量研究方法。

实验结果

研究问题

• RQ1 美国手语手部姿势的经验信息熵是多少？
• RQ2 ASL手部姿势的冗余度与英语音素相比如何？
• RQ3 为何手语尽管发音更慢，仍能实现与口语相当的通信速度？
• RQ4 手部姿势能否作为分析手语信息含量的可靠且可量化的单位？
• RQ5 ASL手部姿势的高熵对自动手语识别与合成系统有何影响？

主要发现

• ASL手部姿势的信息熵显著高于英语音素，接近给定手部姿势集合的理论最大值。
• ASL手部姿势的冗余度显著更低——比英语音素低约20%，表明其单位信息含量更高。
• ASL手部姿势的高熵补偿了其较慢的生成速度，解释了为何手语句子的完成时间与口语句子相近。
• 本研究通过聚焦手部姿势作为基本单位，建立了一种可靠且数据驱动的信息内容量化方法。
• 研究结果为通过信息论优化提升自动手语识别与合成系统提供了基础度量标准。
• 分析表明，手语的信息效率高于以往假设，挑战了关于其通信效率的传统认知。