MLUm & MLUw: summary of responses

[Elizabeth Bates]

Just a quick observation regarding the difference between MLU in words and
MLU in morphemes:

I agree with the various authors who have concluded that the two measures
are highly correlated.  In fact, MLU of either sort tends to be "co-linear"
(correlate highly) not only with each other but with a whole lot of
complexity measures, including various complicated measures of
propositional complexity.  Many years ago Lynn Snyder and Inge Bretherton
and I took great pains to apply some of the candidate propositional
complexity measures of the time (e.g. Kintsch's procedures; Antinucci &
Parisi's proposals) to our Boulder data (the
same data reported in "From first words to grammar", now housed at least
in part in CHILDES).  In the end we decided not to publish the results,
despite all the efforts involved, because the various indices that we
were comparing were so highly correlated, with each other and with
MLU in morphemes, that they didn't tell us anything that we didn't already
know from old-fashioned MLU.  The matter is treated as a footnote somewhere
in the book.

However, there is another point worth making here, particularly in the
context of cross-linguistic research.  Correlations tell us about
the individual differences between children, i.e. children who are high
on A are also high on B, children who are low on A are also low on B,
and so forth.  That does *NOT* mean, though, that two highly correlated
measures give us the same information or CONTENT.  Weight and height are highly
correlated across the normal population, for example, and yet we would
agree that each one yields quite distinct information.  To illustrate
the point: back in Boulder we also did a study (also unpublished...)
comparing MLU in three free-speech contexts involving the child and
his/her mother: having a snack, playing on the floor with standard toys,
reading a book together.  The MLU measures for these three situations were
highly
correlated, in the sense that individual differences on one correlated
with individual differences on the other.  However, there were interesting
mean differences between situations in the kind of speech that was elicited,
in directions that all child language researchers will recognize (more
pronominal forms in free play; more nouns and adjectives in book reading;
more past and future reference in the snack -- after all, how much is there
to say about the here and now in a snack, i.e. about the cracker and cheese?).

This latter point has recently become important to us again, in the context
of a cross-linguistic project that I am currently carrying out with Cristina
Caselli, Antonella Devescovi, Judy Reilly, and several students in Italy
and San Diego (with sage advice provided now and then by Elena Pizzuto).
We are looking at grammatical development in English vs. Italian
two-year-olds who are matched for vocabulary size (using the MacArthur
CDI).  The point of this exercise is to look at cross-linguistic
differences in grammatical complexity and morphological marking when the
children are matched for levels of lexical development.  We have already
shown WITHIN each of these languages that there is a very powerful
relationship between vocabulary size and grammar (even when variation in
age is regressed out -- recent paper by Caselli et al. in JCL).  However,
we also know (on informal grounds) that the AMOUNT of grammatical
morphological that Italian children have to master at any given point in
development is a lot more than English children have to master.  Our
question was: how can we show this cross-linguistic difference in
quantitative terms?  So we have developed a series of MLU measures ranging
from MLU content words, total MLU in words (including functors), and
several different indices of MLU in morphemes designed to pick of
similarities and differences in the morphological options between these two
languages.  The project is still underway, and I can only cite preliminary
results, but it looks like (1) there continue to be strong correlations
between vocabulary size and any complexity metric, within either language,
but (2) our ability to detect differences between the two languages is
quite dependent on the MLU measure that we choose (e.g. differences in
terms of MLU in words and differences in terms of MLU in morphemes give
very different perspectives on cross-language variation).

I hope this is helpful -- and we would very much appreciate hearing from
anyone who has asked a similar cross-language question.  Advice very
welcome. -liz bates




>Dear info-childes members,
>As requested, here is a summary of the responses to the question I
>posted about a week ago: "Is there a difference between MLU in words and
>MLU in morphemes?"  Responses included references to both published
>studies, as well as  some  anecdotical responses. Thanks to all who
>responded!
>
>1.) Studies of non-English speaking children showed high correlations
>between MLUw and MLUm.
>        Hickey, T. (1991).  Mean length of utterance and the acquisition
>of Irish.  Journal of Child Language, 18, 553-569.
>
>        Aguado, G. (1988).  Appraisal of the morpho-syntactic competence
>in a 2.5 year old child.  Infancia y Aprendizaje, 43, 75-95.
>
>        Thordardottir, E., & Weismer, S. (1998).  Mean length of
>utterance and other language sample measures in early Icelandic.  First
>Language, 18 (52 pt. 1), 1-32.
>
>2.) A study in press by M. Malakoff (to appear in Applied Developmental
>Psych) found high correlation (.97) between the two measures for 24
>month old African American children with low SES.
>
>3.) Responses from researchers who addressed this question either
>directly or indirectly in unpublished studies generally reported they
>recalled a high correlation between MLUm and MLUw.  The one exception
>was a study of Hebrew children, which found a difference between the
>scores of the two measures.  Others speculated that the difference in
>MLUm and MLUw was greater for langauge impaired childern than for normal
>children.
>
>This is all the info I have for now.  If anyone has any more
>information, I'd still be interested.
>Thanks again! -Corinna Butt