Apparatus and method for an adaptive neural machine translation system
An apparatus has a network interface circuit to receive a source sentence from a network connected client device. A processor is connected to the network interface circuit. A memory is connected to the processor. The memory stores parameters of a neural machine translation system and instructions executed by the processor to operate the neural machine translation system trained on a corpus of source sentences and corresponding target sentences. The neural machine translation system is trained to predict a target sentence from a source sentence. A translation hypothesis is formed for the source sentence to form the target sentence. The target sentence is supplied to the network connected client device. A correct translation is received from the network connected client device. Parameters of the neural machine translation system are updated based upon the correct translation.
1. An apparatus, comprising:
a network interface circuit to receive a source sentence from a network connected client device;
a processor connected to the network interface circuit; and
a memory connected to the processor, the memory storing parameters of a neural machine translation system and instructions executed by the processor to:
operate the neural machine translation system trained on a corpus of source sentences and corresponding target sentences, wherein the neural machine translation system is trained to predict a target sentence from a source sentence,
form a translation hypothesis for the source sentence to form the target sentence,
supply the target sentence to the network connected client device,
receive from the network connected client device a correct translation, and
update the parameters of the neural machine translation system based upon the correct translation by computing a gradient of a loss function for the source sentence and the target sentence and using the gradient of the loss function to update the parameters of the neural machine translation system.
2. The apparatus of claim 1 wherein the memory stores instructions executed by the processor to compute the perplexity for a set of updated parameters of the neural machine translation system based upon the source sentence and the target sentence.
3. The apparatus of claim 2 wherein the memory stores instructions executed by the processor to determine a number of updates for the source sentence and the target sentence by comparing the perplexity with a threshold value.
4. The apparatus of claim 1 wherein the memory stores instructions executed by the processor to compute a context vector as a linear mixture of vectors representing source word positions.
5. The apparatus of claim 1 wherein the memory stores instructions executed by the processor to form a graphical user interface with a source sentence text entry area, a corrected translation text entry area and a target sentence display area.