Basically, learners have a tendency to hear the second language through the phonological 'filter' of the first language. They assign incoming speech sounds to categories based on the cues that are used in their first language rather than the second language, so they actually perceive the sounds in a different way from native speakers. This is a concept known as categorical perception. For example, Japanese learners of English tend to perceive English /l/ and /r/ as the same sound and cannot discriminate them reliably in perception experiments. Therefore, we can speculate that they have trouble producing the sounds differentially because they don't actually hear a difference between them.
So, the first step in overcoming pronunciation issues is not necessarily to teach tongue positions, but to train perception.