How Do I Read Math Formulas in Natural Language Processing When I?

How do I read math formulas in natural language processing when I convert PDF to text formulas will be lost?

There are too many challenges in NLP but based on my 10+ years of experience, the biggest one is combinatorial explosion. Assuming a language has 10,000 vocabulary terms and a sentence would have 10 words, the whole solution space will be as large as 10,000^10, which is so huge that we often cannot effectively compute it at all. You might argue that with grammars or N-gram data, the solution space can be reduced significantly. Both are helpful somehow but t don’t reduce the solution space as much as we expect. Even with the aid of grammar and N-gram data, we will run into combinatorial explosion now and then when processing language, not to the languages are so dynamic and flexible that the grammars or N-gram often bring too many exceptions and finally we have to make a trade-off between the accuracy and time. Today virtually all NLP technologies, from statistics to neural network, can be considered as a certain search algorithm. If we try to use them to solves a real-world problem, sooner or later we run into a combinatorial explosion problem. Why humans don’t have combinatorial explosion problem when handling languages? Humans process the “meaning” of language directly, without needing to search for an answer by comparing the all possible answers in the whole solution space. We have no idea on how humans do that so we cannot use the similar method to avoid combinatorial explosion. Despite that, personally I’m quite optimistic the the future of NLP. I believe soon we will be able to make breakthroughs by inventing a new approach of NLP, which compute with the meanings directly, instead of the symbols of language.