Third-way linguistics: generative and usage-based theories are both right

Language research remains largely affected by the generative grammar vs usage-based rivalry. The polarization is so pervasive that one seems to have no choice but to assume that language categories are either entirely innate or fully learned. Nonetheless, it is possible to refrain from taking a side in the generative vs nongenerative debate. This paper highlights the work of authors over the last thirty years who believe that, on the one hand, input and domain-general, cognitive constraints alone are insufficient to learn and represent a language and, on the other, that the faculty of language (FL) – if it exists – must incorporate statistics, i.e., a counting device. The core idea of ‘third-way’ linguistics described in this paper is that languages can work because language users' statistical sensitivity and their innate grammar module interact. For a language to function, language users must implicitly know two things. First, by accumulating experience and memory, language users come to know that some forms are likely to go together in the input. Second, from a frequency-independent device (the FL) they also know in abstract (i.e., prior to input exposure) why those forms and not others can do so.