Wittgenstein said that “about what one cannot speak, one must remain silent.”
I can’t prove to you that I have non-linguistic thoughts (because proof is a form of essentially linguistic rigor), but I assert that I do nonetheless. When I make a snap judgment whether to speed through a yellow light or slow down abruptly, I don’t phrase it as a question before an answer arrives. Usually, my head is full of song lyrics whenever I’m driving. I suspect this kind of calculation, if it’s even profitable to think about in terms of sensoria, to be performed visually and spatially, a quick-compare against memorized impressions of speed and inertia.
Wittgenstein might argue that this doesn’t constitute a form of thought, but surely it’s an interaction between the contents of my mind. In order to know whether my van will fishtail if I slam the brakes, I have to know something about (a) how heavy objects behave when acted upon, (b) how slick the road feels, and (c) how much distance there is, deceleration-adjusted, to the stoplight. I don’t know any of these things in a form that could be easily communicated – that is, I have no precise words for the knowledge that I act upon. If asked, I am left to describe the thought, and with effort.
I’m actually a pretty bad driver. Let me refer instead to music, something I’m alright at. The usual role of bass in a rock band is counterpoint. In order to know what notes to walk, to bridge the chords in the song, I pick around intuitively until I produce sound with an emotional character that I like. I can’t tell you why I prefer certain modes, except using an invented vocabulary (e.g. “burnt,” “bluer,” “darker,” “pointier,” etc.) which is the reason most people hate musicians. To me, these characters are expressed as shapes and colors, but since I can’t share my synaesthesia directly with others, I can only hope my metaphor produces “favorable coincidence” in the mind of the person to whom I am speaking.
Now, I’m not refuting Wittgenstein here. When I say that I prefer F# because it’s orange, I’m still speaking about my private impression of F#. But I’m not communicating it, particularly. There is no language model by which I can assume that my bandmate will hear “orange” and know why I think that fits. However, neither could Wittgenstein demonstrate that I don’t think about its fitting as such, unless we’re defining thought circularly as langauge.
Almost every form of art has developed a means of using impressions to elicit responses in the recipient – particularly the manipulation of emotion. For example, one emotion I have frequently when watching motives is hate (i.e. “unfavorable coincidence”). When I was watching The Watchmen, I had no idea why certain action beats made me want to groan, until somebody told me Snyder was doing a bunch of shit to conceal wirework. Now, I know why I hated half a dozen other movies. At the time, I only knew that the response it produced in me was negative. I have forgotten what the scene even looked like. How would Wittgenstein explain the immediacy of such impressions from art? With a “hidden language” that operates beneath or outside the space of conscious speech? It’s too bad I can’t ask him; I somehow doubt this would be his reply.
If so, however, then it brings up the question I ask of linguistic determinists in general: given the hypothesis that our other forms of thought are conditioned or even constituted by language, why is everyday speech so frequently useless for explaining them?

