An LL test is accepted as more reliable than chi-shquared tests in the corpus field as an LL test, unlike a chi-square test, does not assume normal distribution of data. However, a minimum of 5 for expected cell value also applies as for chi square tests, in which case an exact test is recommended.
Mathematic justification of the reliability of LL in corpus linguistic studies can be found in Paul Rayson's PhD thesis
Rayson, P. (2003). Matrix: A statistical method and software tool for linguistic analysis through corpus comparison. Ph.D. thesis, Lancaster University.