Are words an instrument of knowledge? And, if so, what sort of? Are they an instance of inference insofar as one infers the meaning on the basis of the words used? Or are they are an independent instrument of knowledge, since the connection between words and meanings is not of inferential nature?

Kumārila discusses the topic in the Śabdapariccheda ‘Chapter on Words’ (or ‘Chapter on Linguistic Communication’) of his Ślokavārttika (henceforth ŚV). The chapter is the first one focusing specifically on language of the ŚV and in this sense it needs to hint to various arguments which will be elaborated upon in various successive chapters. Accordingly, in the same chapter Kumārila addresses Sāṅkhya, Naiyāyika, other Mīmāṃsaka, and Vaiśeṣika and Buddhist opponents.

I have dealt with the first group of opponents (of the Sāṅkhya, Nyāya and Mīmāṃsā schools) in this post. Against the Buddhist opponents, Kumārila’s strategy is to first show that no inference (in the technical sense of the Sanskrit anumāna) can be formalised in the case of word and meaning. Instead, words are described as being an independent instrument of knowledge. Then, however, two dramatic turns occur:

1. Kumārila candidly admits that he has only provisionally accepted that the words are instruments of knowledge. Instead, words are in fact not an instrument of knowledge. Only sentences are. Words, rather, repeat what has been acquired through another instrument of knowledge or function as recollectors (smāraka).
2. Next, Kumārila even more candidly admits that words can indeed be an instance of inference. This is nonetheless no problem for the advocates of Linguistic Communication as an instrument of knowledge since sentences, not words are the vehicles for knowledge.

This is not the first case in which Kumārila is ready to abandon a thesis in order to save a more general point (see John Taber’s Kumārila On Perception regarding perception and Erich Frauwallner’s Bhāvanā und Vidhiḥ bei Maṇḍanamiśra regarding the role of root and verbal ending in communicating the bhāvanā).

Does it mean that Kumārila is often chiefly a polemist? Or that he focuses more on other issues (in this case, more on the authority of the Veda)?

More information on the workshop on the Śabdapariccheda out of which this post originated can be read here.
Here you can read my analysis of the first part of the chapter.