Evaluating goodness-of-fit of discrete distribution models in quantitative linguistics
Ján Mačutek, Gejza Wimmer · Journal of Quantitative Linguistics · 2013
The paper questions the use of the Pearson chi-square goodness-of-fit test for discrete models in linguistics. It is argued that the stochastic independence, one of necessary conditions for a correct application of the test, is not realistic for linguistic data. Several alternative possibilities (computational and empirical approaches) are suggested. Advantages and drawbacks of the alternatives are discussed.