more thoughts on MLU

[Brian MacWhinney]

On 2/19/04 8:03 PM, "Joseph Stemberger" <stemberg at interchange.ubc.ca> wrote:

> Is there any way to take the differences in MLU in adult speech in
> different languages, and create a way to adjust child MLU's to equate
> for inherent differences between the languages?
>
Joe,

How about this method:  You take the Bible in its various translations,
morphemicize it, devise a counting system, compute Bible MLU and then
compute corrected child MLU as the ratio of raw child MLU over Bible MLU.

Since you would have a complete morphemicization of both your child corpora
and the Bible, you could just vary strings in CLAN or run global
replacements, as I suggested in my earlier message.  Using this method you
can compute Brown MLU, Peters MLU, Stemberger-cautious MLU,
Stemberger-Radical MLU, dual-system MLU, and whatever.  For each, you would
correct by the Bible using the matching system.

However, the real proof of the pudding is whether MLU predicts anything.  I
think that, despite many papers to the contrary in books on developmental
methodology, the obvious candidate is age (across some large sample, of
course).  For me, the MLU measure that correlates best with age is the best
language-internal MLU measure.  What would be really neat is getting a
Bible-corrected MLU that ended up not only predicting age within the
language, but which yields corrected values that actually look similar
cross-linguistically.

However, as Yonata noted in an offline message to me, all of this becomes
impossible if you have to do hand computation of morphemes.  It only works
if you have automatic morphological analysis as we now have for English,
Japanese, and Spanish.  (Cantonese, Italian, and French are still not good
enough for this).  Oh, yes, one other thing is that you would have to add a
bunch of rather strange words for Bible, such as "gnash" and "disciple",
probably sticking with the New Testament.

--Brian