Jenkins, et al (1980) asked 17 good and 17 poor third grade readers to pronounce nonsense words such as clide, sarwinky, and weapadoot in isolation. They found that the good readers were significantly more flexible in their attempts than were the poor readers, many of whom either continuously repeated a pronunciation or gave one that was entirely unrelated to the key word. Even though given repeated opportunities to correctly pronounce these words, poor readers as a group didn't change each pronunciation or think of reasonable alternatives.
Schell, L. M. (1986). Strategies for Independently Attacking Unrecognized Words. Reading Horizons, 26 (2). Retrieved from http://scholarworks.wmich.edu/reading_horizons/vol26/iss2/5