All credit to Mitch for this answer, but a very similar question was posted to Linguistics.SE:
There are qualitative and quantitative measures for 'distance'.
Qualitatively, many languages are easier to compare simply by know
something about their family tree, which is implicitly recognizable
(if one is lucky enough to know so many languages so well), by
comparing vocabulary lists, syntax rules/systems, and phonology. The
more common elements the closer they are. Mandarin and Cantonese are
not mutually intelligible, but both have rich tonal systems, similar
isolating grammars, and fairly simple phonological rules to map
cognate words. Likewise French and Italian have a set of comparisons
(similar conjugation/gender rules) and fairly regular rules mapping
cognates. And the two sets of differences, those between
Mandarin/Cantonese, and those between French/Spanish are of the same
scale. The difference between French and German is so much larger than
between Mandarin/Cantonese that one would be hard pressed to say they
are of similar distance.
Quantitatively, one can define distance mathematically, by collecting
a set of quantitative features in each language, and then defining a
distance function that calculates an exact positive real number out of
a combination of all the differences of these features. Then one can
do a simple numerical comparison, the distance between X and Y and the
distance between Z and W. One can go further and use a clustering
algorithm to create a formal family tree of all languages of concern.
Presumably, the utility of a distance function would be to predict the
difficulty in language learning or translation (machine or human); the
further the distance, the more changes would need to be navigated. At
some point, a distance will be meaningless; the difference between
French and Swahili is just too much to settle qualitatively or quantitatively to then bother comparing with some other distance.
Alternatively, this scientific report suggests that:
The distance between two languages may also depend on whether it is in the
written or spoken form. For example, the written form of Chinese does not vary among
the regions of China, but the spoken languages differ sharply. Alternatively, two
languages that may be close in the spoken form may differ more sharply in the written
form (for example, if they use different alphabets, as in the case of German and Yiddish).
Perhaps the way to address the distance between languages is not through
language trees which trace the evolution of languages, but by asking a simpler question:
How difficult is it for individuals who know language A to learn languages B1 through
Bi, where there are i other languages.
If it is more difficult to learn language B1, than it is to learn
language B2, it can be said that language B1 is more “distant” from A
than language B2. Language B3 may be as difficult to learn as is
language B1 for a language A speaker, but that does not mean that
language B3 is close to language B1. Indeed, it may be further from B1
than it is from A. Alternatively, if the issue is the adjustment of
immigrants speaking languages B1 through Bi in the linguistic
destination A, one would want to know how difficult it is for speakers
of B1 through Bi to learn language A.
Here are two images of the table that shows the linguistic proximity between languages according to the study.