Interlanguage is a ‘mental grammar‘, the learner’s own linguistic representation of the target language. It is different from the grammar of the second language (and learner is aware of that?).
It is now the basic assumption behind much of second language research.
The starting point of SLA research is the learner’s own language system…finding out how students learn means starting from the curious rules and structures which they invent for themselves as they go along.
Interlanguage is permeable and constantly changing: it is open to change both from outside (e.g. from input) and inside. It is repeatedly restructured as learner gradually acquires more of the L2. It might also be constrained by the principles and parameters of Universal Language. However, a learner’s IL is systematic.
I would also add into a model of Interlanguage the idea that part of this restructuring involves the learner moving from conscious to automatic processing of language items.
This restructuring is not linear. Rather, it may involve ‘U-shaped’ progress, where a learner replaces a correct form with an incorrect form and only later recorrects: for example, learners might say ‘she ate’, but replace it with ‘she eated’ or ‘she ated’ when aware of the -ed form of the past tense. When a learner utters ‘she ate’ early on it may simply be copying a chunk of language without thinking, whereas ‘she eated’ actually shows some processing, albeit an overgeneralization of the ‘-ed’ past ending. In other words, the error may actually indicate progress!
Interlanguage sees learners as actively involved in trying to make sense of the language they are exposed to, rather than passive recipients who soak in language. [This assumes that the process of learning an L2 is different from an L1: the infant L1 learner soaks it in and learns effortlessly and subconsciously; IL assumes the L2 learner is actively involved…but, of course, some learners are more active than others!]
Below is a diagram I created, representing my understanding (when I created it a couple of years ago – it needs revising!) of interlanguage. It shows where the learner starts: this may be nothing (zero) or it may be universal grammar, the biologically-endowed system. It shows that interlanguage is affected by L2 input, a learner’s L1 and other skills and knowledge. Note that not all L2 input affects interlanguage; only input that becomes intake through noticing affects interlanguage.