The comments in this thread, specifically the questions about deaf readers prompted me to Google "deaf readers subvocalization", which lead me to the book "The Psychology of Reading" by Keith Rayner and Alexander Pollatsek.
I'll excerpt a few paragraphs.
[Page 191 - http://goo.gl/VLaaM]
Electromyographic Recording (EMG) records have been widely used to study the role of subvocalization in reading. [...] Generally, EMGs are also recorded from some other part of the body, such as the forearm, where muscle activity should have no relevance for reading. In fact, normal skilled readers show little forearm activity during reading, while EMG activity in the speech tract increases markedly during reading in comparison to the baseline condition (where the subject sits quietly). In contrast, deaf readers show a considerable amount of forearm activity during reading (we shall see the reason for that later in this chapter).
[Page 211 - http://goo.gl/R3NPT]
That profoundly deaf people can read at all is sometimes taken as evidence that speech recoding in normal readers is optional. However, as we have seen, they do not read very well, and the best available evidence indicates that when reading English text deaf readers recode the printed information into their native language (ASL) for comprehension purposes. Since they have not experienced speech sounds, recoding into a phonological code is precluded from deaf readers. Instead, the information is recoded into a manual form to aid comprehension. The poor reading of deaf people is thus probably due in part to inner speech being a more efficient system of recoding than overt manual gestures.
Readers of logographic systems such as Chinese can probably access the meaning of many printed characters directly from the visual representation. However, associations between the printed word and the appropriate pronunciation are activated during reading and appear to be important in comprehending text.