Inaccurate Tone Adjustment in tone_sandi

[apinge]

In python version, tone_sandhi is used to adjust Chinese character's segmentations and tones. This issue is to address the inaccuracies in tones in the original Python version. We can both resolve it directly in our C++ version or submit PRs to the original repo.

• "地" as noun
  e.g. "大地" in “温暖的太阳照耀着大地”
  related code snippet (in _neural_sandhi)

elif len(word) >= 1 and word[-1] in "的地得":
            finals[-1] = finals[-1][:-1] + "5"

Possible Change:

elif len(word) >= 1 and word[-1] in "的地得" and pos[0] != 'n':
            finals[-1] = finals[-1][:-1] + "5"