28-04-2024, 11:23 AM
Agreed with the first part. The text is so large and consistent that weirdoes can be safely ignored for the most part. Unless, of course, an analysis is focused on the weirdoes specifically. They can be valid objects of study, but matter little when it comes to the text as a whole.
Sometimes I think it might help to take this idea even further. For example, 98% of occurrences of EVA-a are provided by [ai, ar, al, am, an, as]. If you leave off [an] and [as], you still get 96%. Something like [ao] accounts for 0.07%. My point is that fretting over the existence of [ao] probably distracts us from what is really going on - a system that predominantly produces [ai, ar, al, am].
Regarding the use of EVA, I think this is a matter of choices made. This will be the issue with all transcription systems: the user has to be aware of the decisions underlying the system.
Sometimes I think it might help to take this idea even further. For example, 98% of occurrences of EVA-a are provided by [ai, ar, al, am, an, as]. If you leave off [an] and [as], you still get 96%. Something like [ao] accounts for 0.07%. My point is that fretting over the existence of [ao] probably distracts us from what is really going on - a system that predominantly produces [ai, ar, al, am].
Regarding the use of EVA, I think this is a matter of choices made. This will be the issue with all transcription systems: the user has to be aware of the decisions underlying the system.