To use a software analogy, thoughts are the model. English words are one possible serialization.
(Other possible serializations include other phrasings of English words, Japanese words, mathematical formulas, drawings, musical notation, and so on.)
It's nonsensical to ask me for a more fundamental form of a thought in my brain. It'd be like having a program where you have a reference to a directed graph in memory, and asking "So what form does this graph take? Like XML, or JSON, or what?" It doesn't have any other form. I could serialize it into one of those forms, if need be, but right now it's just a bunch of connections in memory.