Is the action history and output buffer embeddings not needed any more?

[zxteloiv]

Hi, first thanks for your kindly sharing of the codes. With the comparison to the original RNNG (in its cpp version), I found that the action history and output buffers are not included in this repo. Only the stack context is used to predict actions and words. Is my understanding correct? If so, do they differ due to the implementation reason or just not necessary?