I don't know of actual research, but from my personal experience:
I agree that word-by-word translations will on the one hand lack features that are not available in the target language and on the other hand introduce features that are not present in the source language, which is why I wouldn't rely too much on such strict word-by-word translations either. They are an approximation, but it is only to a very limited extend possible to depict in a completely different natural language how a particular other natural lanugage conveys the meaning in terms of the grammatical and semantic subtleties.
However, what I always found very helpful in second language learning were interlinear glosses with detailled morphological annotation (for the grammatical features) and morpheme-by-morpheme correspondence (for the semantic structure), and then in the third line the natural translation in the target language.
To pick up your example, it would be something like:
Qu' est-ce que ça veut dire?
what be.3SG-this that that want.3SG say.INF
"What does this mean"?
Pour-quoi est-ce qu' il n' y a pas des crayon-s rouge-s?
for-what be.3SG-this that he NEG there have.3SG not a.PL pen-PL red-PL
"Why are there no red pens?"
(BTW, if anyone has a better suggestion how to gloss ne ... pas and des, or how to differentiate between that as a pronoun and as a conjunction while preserving maximal informativity in the interlinear gloss, feel free to edit - I didn't find an official guideline for French. The same holds of course if I made any mistakes in the translation.)
As you can see, I am able to get both a thorough understanding of how the source language is structured grammatically AND semantically (by the interlinear gloss) and what it is supposed to mean (by a rather free, but maximally meaning-depicting translation in how I would express it in my mother tongue), but at the same time didn't face the problems with a word-to-word translation into a sentence that ought to be possibly close to the source language but still grammatical in the target language which is often simply not possible.
An informative morphological glossing requires some lingustic knowledge though (even my annotation is probably not perfect), and also takes some time to do throroughly.
If you want to take a look at it, the Leipzig Glossing Rules are some generic guidelines for linguistic glossing.
But I'd like to emphasize again, this is just subjective experience and not scientific, generally applicable or proven research.