Recognition of Non-Manual Content in Continuous Japanese Sign Language.

The quality of recognition systems for continuous utterances in signed languages could be largely advanced within the last years. However, research efforts often do not address specific linguistic features of signed languages, as e.g., non-manual expressions. In this work, we evaluate the potential of a single video camera-based recognition system with respect to the latter. For this, we introduce a two-stage pipeline based on two-dimensional body joint positions extracted from RGB camera data. The system first separates the data flow of a signed expression into meaningful word segments on the base of a frame-wise binary Random Forest. Next, every segment is transformed into image-like shape and classified with a Convolutional Neural Network. The proposed system is then evaluated on a data set of continuous sentence expressions in Japanese Sign Language with a variation of non-manual expressions. Exploring multiple variations of data representations and network parameters, we are able to distinguish word segments of specific non-manual intonations with 86% accuracy from the underlying body joint movement data. Full sentence predictions achieve a total Word Error Rate of 15.75%. This marks an improvement of 13.22% as compared to ground truth predictions obtained from labeling insensitive towards non-manual content. Consequently, our analysis constitutes an important contribution for a better understanding of mixed manual and non-manual content in signed communication.

Auditory coding of human movement kinematics.

Although visual perception is dominant on motor perception, control and learning, auditory information can enhance and modulate perceptual as well as motor processes in a multifaceted manner. During last decades new methods of auditory augmentation had been developed with movement sonification as one of the most recent approaches expanding auditory movement information also to usually mute phases of movement. Despite general evidence on the effectiveness of movement sonification in different fields of applied research there is nearly no empirical proof on how sonification of gross motor human movement should be configured to achieve information rich sound sequences. Such lack of empirical proof is given for (a) the selection of suitable movement features as well as for (b) effective kinetic-acoustical mapping patterns and for (c) the number of regarded dimensions of sonification. In this study we explore the informational content of artificial acoustical kinematics in terms of a kinematic movement sonification using an intermodal discrimination paradigm. In a repeated measure design we analysed discrimination rates of six everyday upper limb actions to evaluate the effectiveness of seven different kinds of kinematic-acoustical mappings as well as short-term learning effects. The kinematics of the upper limb actions were calculated based on inertial motion sensor data and transformed into seven different sonifications. Sound sequences were randomly presented to participants and discrimination rates as well as confidence of choice were analysed. Data indicate an instantaneous comprehensibility of the artificial movement acoustics as well as short-term learning effects. No differences between different dimensional encodings became evident thus indicating a high efficiency for intermodal pattern discrimination for the acoustically coded velocity distribution of the actions. Taken together movement information related to continuous kinematic parameters can be transformed into the auditory domain. Additionally, pattern based action discrimination is obviously not restricted to the visual modality. Artificial acoustical kinematics might be used to supplement and/or substitute visual motion perception in sports and motor rehabilitation.