All of us have to strike the right balance between stubbornly believing that NNs will solve all problems in NLP -which, imho, underestimates what linguistics ca

All of us have to strike the right balance between stubbornly believing that NNs will solve all problems in NLP -which, imho, underestimates what linguistics can do for this field of study- and stubbornly saying that only linguistic knowledge will lead to an improvement of the results -which, imho, means underestimating the outstanding work on deep learning techniques done so far. As you said, creating high-quality input is likely to be costly -and I'm not talking about syntax tree creation tooling only here. But, isn't it worth the while? I believe it is.