Signs in sign languages have been mainly analyzed as composed of three formational elements: hand configuration, location, and movement. Researchers compare and contrast lexical differences and similarities among different signs and languages based on these formal elements. Such measurement requires extensive manual annotation of each feature based on a predefined process and can be time consuming because it is based on abstract representations that usually do not take into account the individual traits of different signers. This study showcases a newly developed tool named DistSign, used here to measure and visualize variation based on the wrist trajectory in the lexica of two sign languages, namely American Sign Language (ASL) and Ghanaian Sign Language (GSL), which are assumed to be historically related (Edward 2014). The tool utilizes the pretrained pose estimation framework OpenPose to track the body joints of different signers. Subsequently, the Dynamic Time Warping (DTW) algorithm, which measures the similarity between two temporal sequences, is used to quantify variation in the paths of the dominant hand’s wrist across signs. This enables one to efficiently identify cognates across languages, as well as false cognates. The results show that the DistSign tool can recognize cognates with a 60 percent accuracy, using a semiautomated method that utilizes the Levenshtein distance metric as a baseline.

手语中的符号主要被分析为由三个构成要素组成：手的配置、位置和动作。研究人员根据这些形式元素比较和对比不同符号和语言之间的词汇差异和相似之处。这种测量需要基于预定义的过程对每个特征进行大量手动注释，并且可能非常耗时，因为它基于通常不考虑不同签名者的个人特征的抽象表示。本研究展示了一种名为 DistSign 的新开发工具，用于测量和可视化基于两种手语词汇中手腕轨迹的变化，即美国手语 (ASL) 和加纳手语 (GSL)，假设为历史相关（Edward 2014）。该工具利用预训练的姿势估计框架 OpenPose 来跟踪不同签名者的身体关节。随后，动态时间规整 (DTW) 算法测量两个时间序列之间的相似性，用于量化优势手手腕跨符号路径的变化。这使人们能够有效地识别跨语言的同源词以及错误的同源词。结果表明，使用利用 Levenshtein 距离度量作为基线的半自动化方法，DistSign 工具可以以 60% 的准确率识别同源词。用于量化惯用手手腕跨标志的路径变化。这使人们能够有效地识别跨语言的同源词以及错误的同源词。结果表明，使用利用 Levenshtein 距离度量作为基线的半自动化方法，DistSign 工具可以以 60% 的准确率识别同源词。用于量化惯用手手腕跨标志的路径变化。这使人们能够有效地识别跨语言的同源词以及错误的同源词。结果表明，使用利用 Levenshtein 距离度量作为基线的半自动化方法，DistSign 工具可以以 60% 的准确率识别同源词。