| Introduction |
| ============ |
| Different from offline translation system, the evaluation of simultaneous translation requires incremental decoding with an streaming input. |
| The simultaneous introduce the a front-end / back-end setup, shown as follow. |
|
|
| The back-end contains one or multiple user-defined agents which make decisions of whether to generate prediction at a certain point. |
| The agent can also considered as queue, where the input are keep pushed in and policy decides the timing to pop the output. |
|
|
| The front-end on the other side, represent the source of input and recipient of the system prediction. |
| In deployment, the front-end can be web page or cell phone app. |
| In SimulEval, the front-end is the evaluator , which feeds streaming input to back-end, receive prediction and track the delays. |
| The front-end and back-end can run separately for different purpose. |
|
|
| The evaluation process can summarized as follow pseudocode |
|
|
| .. code-block:: python |
|
|
| for instance in evaluator.instances: |
| while not instance.finished: |
| input_segment = instance.send_source() |
| prediction = agent.pushpop(input_segment) |
| if prediction is not None: |
| instance.receive_prediction(prediction) |
|
|
| results = [scorer.score() for scorer in evaluate.scorers] |
|
|
|
|
|
|
| The common usage of SimulEval is as follow |
|
|
| .. code-block:: bash |
|
|
| simuleval DATALOADER_OPTIONS EVALUATOR_OPTIONS |
|
|
| We will introduce the usage of the toolkit based on these three major components: Agent, Dataloader and Evaluator. |
