6 Assessing productive and interactive skills Personal refl ection

Using computers to score performance
As computer technology advances, automated scoring of more open-
ended performances is becoming an increasingly practical alternative 
to human ratings for large scale test providers. Automated essay 
scoring has already been in use for many years and with advances in 
speech recognition, automatic scoring of tests of speaking is also 
becoming increasingly popular. 
Automated scoring systems usually depend on fi rst obtaining large 
numbers of human ratings. These are used to ‘train’ the machines by 
building statistical models of what responses at different score levels 
‘look like’. Text analysis software has been developed that can count 
features of a text such as word and sentence lengths, the frequency of 
occurrence of different words, phrases and grammatical structures, 
relationships between parts of a text and so on. Given enough samples 
of assessee writing (or speech captured through automated speech 
recognition tools), the machine comes to associate certain patterns of 
features with scores at different levels. 
For example, if the fi rst essay the system is given to analyse has 18 
words per sentence and is scored as A, but the second has 16 words 

---