Outline 1. Introduction 1. Motivation 2. Portuguese Sign Language 3. Goals 2. Architecture 1. Gloss Translate 2. Lookup 3. Animate 3. Implementation 4. Evaluation 5. Conclusions 6. Demo
Motivation the system would be helpful in improving communication and reducing the isolation of the deaf community higher rate of illiteracy and learning problems hard to access content, aggravated by the rise and speed of new technologies
Sign Languages and LGP
Knowledge Sources
Different Sentence Structure OSV (Object-Subject-Verb) vs. SVO (Subject-Verb-Object) strong topicalization
Different Grammar classes No distinction between nouns and adjectives Articles? (o, os,.. um, uns) inflexion: gender, number, aspect, time... spatial agreement
Gesture Subunits gestures are built from parameters hand configuration hand orientation hand placement hand movement non-manual
Writing Systems
Writing Systems
Goals basic architecture and implementation that others can build upon study existing research and list open problems
Architecture
Gloss Translate Statistical Machine Translation language model bilingual lexicon alignment model Rule-Based Machine Translation lexicon and bilingual dictionary rules
Gloss Translate Rule-Based Machine Translation pipeline
Lookup
Animate - Rig Inverse Kinematics Definition of spatial marks Facial Controls and Mouth Shapes
Animate - Procedural Gestures
Animate - Avatar variance
Animate - Procedural Gestures inter- and intra-gesture blending transitions and timing acquiring data
Implementation
Blender
Blender Python Addon Interface Rendering
Translation and Natural Language Processing Python + NLTK - Natural Language Toolkit
Lookup JSON database
Rigify
Spatial marks - mesh
Spatial marks - grid
Base configurations 57 base configurations
Action Layering and Blending
Gesture building and Transitions
Gesture building and Transitions Distinction between: regular glosses numerals fingerspelling
Evaluation 6 deaf persons / interpreters + collecting feedback, observation in informal environment
Evaluation usefulness user interface translation quality avatar perceptibility
Conclusions definitely possible, definitely a challenge Sign Languages and LGP have particularities that pose additional problems more work from a multidisciplinary team is needed more linguistic studies
Contributions 1. summary of related work and background 2. overview of pipeline and architecture 3. open implementation as proof of concept 4. list of problems and future work
Contributions
Demo
Recommend
More recommend