Last time we introduced the concept of phonemics, that is going from an outsider’s perspective of a language (‘etic’) to an insider’s perspective (’emic’). We need to turn our phonetic text (words written exactly as pronounced) into phonemic text (words written in a way intuitive to a native speaker).
At this stage in the process, we have a few hundred words (for a preliminary analysis anyway; a proper analysis will need thousands) written in phonetics, and we’ve identified our phonetic inventory — which sounds we have.
So how do we get at the implicit understanding a native speaker has of their language? We do it by looking at distribution patterns (interpretation, part 2) and phonetic environment (analysis, part3).
First off we need to acknowledge that the data we’ve worked hard to record and have triple checked for pronunciation has our own biases in it. In part 1 we said that phonetics was all about writing one symbol for one sound. It turns out that this isn’t quite true as some sounds can be written in multiple ways! (Don’t you love subjects where lesson 2 contradicts lesson 1?)
[ɲ] for example (the n sound at the beginning of the English word “new” which is not a pure ‘n’ ) can be written as [ɲ] or [nj]. Both ways are a correct way of writing the sound, but which are you more likely to choose?
As an English speaker, I’m most likely to choose [ɲ] because in English I think of it as a single sound. My ear hears it as a single sound, not two next to each other (unlike [nd], which is two sounds a [n] and a [d] together).
Similarly [d͡ʒ] (the j sound in “judge”) I’m bound to write as the single affricate sound [d͡ʒ] rather than [dʒ]. My brain tells me that’s a single sound. I’ve found it hit or miss trying to get English speakers to see that “j” is actually two sounds and it starts with d 🙂
In writing phonetics for a new language we’ve potentially introduced a bias based on our mother tongues. So how do we examine our data to check for this?
We identify suspicious sounds in our data, which are:
- Segments (single sounds) that could be written as sequences (multiple sounds).
- Vice versa, sequences that could be segments
- Predictable sounds that could function as transitions
- Semivowels that could be written as vowels or vowel glides (diphthongs), or vice versa.
Fancy linguist speak ay? Not to wo6rry, we have a linguistic toolkit on hand that lists them all if that makes no sense. Just check your data against the list.
The next step is to build a distribution pattern for the consonants and vowels in the data. What this means is to count the number of times [p] appears at the start of the word, the middle of the word and the end of the word. Then write down what other consonants it ‘clusters’ with. Sound like tedious, manual work? Indeed it is, but thankfully software exists :0
The idea behind this is that some sounds are unambiguous; they can only be written one way; they are non-suspicious. These non-suspicious sounds show us the distribution pattern, and we should see if it’s reasonable to rewrite our suspicious sounds to fit the pattern.
If, for example, there are only suspicious CC clusters and there are no examples of non-suspicious clusters, it’s possible there aren’t any. The only reason they’re showing up on the CC chart, after all, is because you chose to write them that way and if you had chosen to write them as a single sound they wouldn’t be there.
It’s not a hard and fast rule, of course. You might have written them exactly as a Kovol person would have done and you share the same bias.
In the Kovol example, almost all the CC clusters are suspicious. There are two non-suspicious clusters which might indicate CC clusters are a natural part of the language – but they could also be spelling mistakes. It’s worth checking!
Since there are non-suspicious CC clusters though, this seemed to indicate that there may be more, and perhaps we’d be best leaving things as they are.
Part of the process can be to go ahead and change things so that [mb] becomes [ᵐb] for example. Let’s turn all these double sounds into single sounds and then go on to part 3 and see what happens to both options. Part 2 has big implications for what happens in part 3, so we end up looking ahead and thinking through the implications of multiple sets of changes.
We’re looking for the option that is the most consistent, makes linguistic sense, is neat and symmetric (i.e. consistent) and preferably results in fewer letters for the alphabet (but not at the expense of usability!)
It’s a bit of a mess to be honest, which is why it takes time to pick at the problem until the right solution comes up.