Corpus Linguistics: How do I compare date from two corpora correctly?

Question

I want to count the occurrences of the English “wanna” structure diachronically, in order to create a frequency per century chart.

My first step was to use the Corpus of Historical American English (COHA). I found 210 occurrences in the 19th century and 1590 in the 20th century. However, I also want to count the data from the 21th century and for that I need to use another corpus, since COHA doesn’t have data from this century.

Fine, so I went to another corpus (a synchronic one) and found 1274 occurrences. The result is shown in the picture.

My problem is: how can I quantify this data correctly? I am only getting more results in the 20th century because of the size of the corpus (COHA has 400 million words and the synchronic current corpus has only 2 billion). Is there any math calculation I may use in order to make such frequency index correct?

Answer 1

You could simply divide by the size, so you’re comparing relative frequency. You could also express the log relative frequency, given how small the resulting decimal number is.

It’s common to express relative corpus frequencies using a standard denominator (per million, for example), as suggested here. CELEX, for another example, provides both the per-million and the log frequency.