The main problem with empirical research in language teaching is that language acquisition is not an exact process that can be entirely predicted, and it occurs differently from person to person. In empirical research you base your investigation on personal observations or experiences. Observations are entirely personal to the observant, and may foster bias and partiality. As with any other developmental process, the variables for learning a new behavior are entirely dependant on the background, developmental level, range of exposure, and even on intrinsic and motivating factors that affect the student. Therefore, empiricism can only take you so far when trying to establish a correlation between two variables after applying an intervention; it can only take you as far as what you are able to see, and what you perceive to be the facts.
Any research design performed in the area of linguistics must take into consideration the variables that will help bring about the final outcome of the intervention to be tested. Similarly, the parameters within any investigation should be properly identified. Direct and indirect observations may or may not take every variable into consideration.
This being said, core linguistics have the benefit of being exact and predictable. Applying linguistic principles to the theoretical framework of language teaching research aids in closing the gaps that may surface when applying theoretical frameworks to a direct or indirect observation, or to a collection of experiences. Hence, you can use the empirical research as it applies to the way in which students write words, create linguistic connections, identify morphemes and lexemes, or physically demonstrate the use of language in its pure, raw form. This way the investigation may render itself much more valuable as well as valid. Also, it allows for the experiment to repeat itself under a different context.