In the previous chapter I addressed the problem of how to analyse free speech. I proposed that it is important to analyse naturalistic speech data from the point of view of complexity, accuracy and fluency. That is, each utterance should be coded for its lexical, morphological, morphosyntactic and syntactic complexity and accuracy, each disfluency and hesitation should be marked, and overall lexical diversity should be assessed.

You can imagine how quickly the wrong coding system can make an entirely unreadable mess of your data. However, do not do what I did when I wrote my Ph.D., which was to print out all of my data (comprising some 175,000 words) multiple times, and to go through them with different coloured text markers to code features such as singular and plural noun phrases, main clauses, subordinate clauses, different tenses, different genders of noun phrases, different cases, and so on, subsequently counting them all by hand. Instead, I recommend that you use the system of coding and transcriptions which has been developed by Brian MacWhinney in the CHILDES (Child Language Data Exchange System) project (http://childes.psy.cmu.edu; see also MacWhinney, 2000).