There is some evidence that thought disorder (also called loose association) arises, at least partially, from increased spreading activation; schizophrenics, for example, often show a greater increase in activation to indirectly related words compared to unrelated words, than do non-thought disordered controls. This is primarily a cognitive mechanism, not a biological one, of course, and I have found very little biologically-based explanation of spreading activation itself; it comes from the computational cognitive science end of things, and is often discussed as part of a neural network model of cognition.
To address Alex Stone's comment above, I believe what he is talking about is either Broca's area (the language-dominant hemisphere's inferior frontal gyrus), generally understood to be important for producing language as well as comprehending complex sentences, or Wernicke's area (the language-dominant hemisphere's posterior superior temporal gyrus), which is generally thought to be important for processing the dominant meaning of words as well as understanding nonliteral language such as jokes, sarcasm, etc. There is some evidence for a correlation between volume reduction in both Broca's and Wernicke's areas, and thought disorder symptoms. Again, however, no mechanism has been suggested; and as with much imaging research, it is at this stage impossible to tell whether the neural abnormalities are the cause of the symptom or themselves a symptom of another cause.
Schizophr Res. 2013 May;146(1-3):308-13. doi: 10.1016/j.schres.2013.02.032.
Schizophr Bull (2008) 34 (3): 473-482. doi: 10.1093/schbul/sbm108
Neural Basis of Semantic Memory, ed. Kraut & Hart (2007). Cambridge University Press.