Sequence-to-sequence transduction translates one sequence into another representation, supporting tasks where the input and output may differ in form. section 5.1.2
Typical applications include:
- Machine translation, where one language sequence becomes another. section 5.1.2
- Summarization and code generation, where outputs may use different textual or computational representations. section 5.1.2
- Outputs need not be text; they can also be computer code or images. section 5.1.2
Choosing the modeling objective
A conversational system that stops mid-sentence may be optimized for raw language modeling rather than sequence-to-sequence transduction. Different objectives produce different task strengths. section 5.1.2
