I am new to Natural Language Processing, I think NLP is a challenging field, the syntax and semantic ambiguities could cause a lot of problems. For example I think for these problems machine translation is a hart task.

Therefore probably many approaches and methods has been applied to the field, I would like to know what are the latest and most promising approaches and methods in the field of NLP?

Does these techniques are dependent highly to the target language?

  • Most of the latest and most promising approaches and methods in NLP depend heavily on previous work, going back several decades. The latest stuff really can't be easily understood without understanding its basis, too. It's not just software on the shelf, yet -- you gotta make a lot of assumptions and do a lot of massaging.
    – jlawler
    Oct 13, 2014 at 17:59
  • 1
    Woah this is way too broad. It's like asking "what is state of the art in the field of science?" You need to narrow this down to one field of NLP. The P just stands for "processing", which covers tons of utterly different stuff. Oct 14, 2014 at 1:30
  • @hippietrail the objective of most NLP tasks is parsing or understanding the language. also I brought the example of machine translation.
    – Ahmad
    Oct 14, 2014 at 5:58
  • As this website has a shared area (NLP) with DataScience.StackExchange.Com website, after not receiving an answer here I asked the question there and I got one answer datascience.stackexchange.com/questions/2268/…
    – Ahmad
    Oct 14, 2014 at 6:00
  • Survey of the state of the art in human language technology. Vol. 13. Cambridge: Cambridge University Press, 1997. is nice but old, anyone has a newer version? Too bad this question got closed. I re-posted on quora.com/… Nov 27, 2014 at 3:59

1 Answer 1


Most NLP problems are language- and resource-dependent, therefore comparing different approaches is not that simple. There are two types of sources which can tell you what approaches are the best in the sub-fields of NLP:

  1. Papers with surveys of methods in the specific subfield (I provide random papers from my field as examples). For example: http://wwwusers.di.uniroma1.it/~ponzetto/pubs/poesio10a.pdf , http://www.sfu.ca/~mtaboada/lot/readings/Mitkov_1999.pdf. Papers like this describe main approaches and methods in the subfield.
  2. Evaluation forums. There is a tradition of evaluation forums, something like competitions between systems which performs specific task. Some well-known examples are SemEval, EVALITA, ROMIP and RU-EVAL. These evaluations show what techniques can achieve state-of-the-art results and what is the level of state-of-the-art in the sub-field.

Not the answer you're looking for? Browse other questions tagged or ask your own question.