Q: How to integrate  syntactic structure and semantic analysis(similarity of words) in n-gram language model? (MW)
A: n-gram models only describe the probability of symbol  sequences (words or characters). They are not able to accommodate any syntactic or semantic structures.

Q: Explain applicability of Chain Rule in N-gram.(LE)    
A Language model tries to compute the probability of a symbol sequence. Because for longer sequences such a probability cannot be estimated directly from data it needs to be broken down into the conditional probabilities of the individual symbols,  given their preceding context. How this can be done is described by the chain rule. 

Q: In N gram model, as the window size(size of N) increases the demand for computing resource increase but increasing the size of N leads to a better accuracy. how to decide the best window size?  (YM)
A: The optimal size can only be determined experimentally by trying out different combinations of n-gram sizes and smoothing techniques.

Q: To deal with unknown words or out of vocabulary OOV, using open vocabulary system is recommended, but people make spelling errors and inncorrct words would be put into the dictionary. how to handle such situations?  (YM)
A: Adding misspelled observations to the dictionary is not a recommendable solution. Depending on the application different alternatives can be used: Replacing all OOV words with a special symbol, or breaking down the OOV word into substrings, e.g. morphs, n-grams, or elementary units determined by means of unsupervised ML approaches. Byte-pair encoding is one of them.

Q: What are the applications of n-gram model? What is special about this model as compared to other such as neural language model for NLP? [DM]
A: N-gram models can only be used in combination with other kinds of model, e.g. for speech recognition, machine translation or text completion. They have the advantage that they can be trained on large amounts of monolingual data, even though the target application might require more expensive data sources, e.g. acoustic recordings or translated texts. N-gram models can be either probabilistic or neural ones.