This is more of a partial answer to your question than a complete answer, but I'll post it nonetheless. The book Language MOOCs: Providing Learning, Transcending Boundaries provides a good bit of information about the effectiveness of language MOOCs, but it never directly states any past literature to cite its claims (I'm sure it does later in the book, but only the free preview is available on Google Books).
The consensus the authors of this paper suggest is that there is no clear answer (due to the lack of research in this area), but the authors definitely lean towards the side that MOOCs cannot teach languages effectively.
I've quoted certain excerpts from the preview that I feel to be related to your question (emphasis mine):
However, it must
be acknowledged that, leaving aside the different quality levels in the instructional
design of individual MOOCs, there are different degrees of success with which a given
subject can be expected to be effectively taught in a MOOC per se, as certain reports
have pointed out (Viswanathan, 2012; Bruff et al., 2013).
Before considering the theoretical suitability of MOOCs for learning second languages,
the following aspects need to be taken into account: firstly, language learning
is not only knowledge-based, in the sense that it requires the rather passive assimilation
of vocabulary items and combinatory rules, but is mainly skill-based, in that it
involves putting into practice an intricate array of receptive, productive and interactive
verbal (and non-verbal) functional capabilities, whose role in the overall success
of the communicative act is generally considered to be more prominent than that of
the formal or organizational elements (Halliday, 1993; Whong, 2011).
Secondly, and
linked to the previous point, assuming that the goal of language learning is language
use, it is only common sense to infer that the former should entail considerable practice
of the latter, just like a student must play the piano to become a pianist or take
photographs to become a photographer.
Thirdly, all variables being equal, the mind
that learns (a language) best is the proactive and engaged mind with its high order
skills (relating, contrasting, criticising, inquiring, justifying, deducing, etc.) activated,
rather than just memorization and mechanical reproduction.
Finally, after infancy,
one is generally assumed to gradually lose some of the innate language acquisition
abilities and acquire a more rule-based cognitive profile (e.g., Meltzoff & Prinz, 2002).
Hence, the language learner is likely to benefit from the well-known explicit type of
learning model, something partly based on face-to-face/textual/visual explanations
with illustrative examples followed by some interesting and creative form of practice.
Part of this process will be more effective if undertaken individually, particularly for
the improvement of certain areas of language, such as pronunciation or punctuation,
as it provides the necessary flexibility and adaptation to personal learning styles,
rhythms and circumstances, and enhances metacognitive processes.
Here, the authors draw attention to the lack of research in this field:
As has been noted previously, LMOOCs are in the very early stage of development. It is,
thus, understandable that they have undergone little research up until now, neither of
an empirical nor of a theoretical nature.
...
This procedure returned an extremely low number of hits and showed that there are
no monographic volumes on LMOOCs published to date, no finished dissertations
and only five scholarly articles in refereed journals (one in 2012 and four in 2013).
...
The dates of publication, the lack of books and published dissertations and the scarcity
of papers are, at present, the consequence of the incipient stage of this field of
research.
...
None of the five scholarly publications on LMOOCs were based on empirical
research.
