Absence of evidence - grambank/grambank GitHub Wiki

Often, a grammar does not describe a given feature for a language. This could mean that the feature is not relevant in the language because the construction or marker we are looking does not exist. In other cases, this means that the author has not investigated a phenomenon or did not deem it necessary to mention. Absence of evidence does not always serve as evidence of absence.

For some features, we can safely assume that they would be mentioned even in a very basic grammatical sketch. This is the case for high frequency phenomena such as gender agreement, person indexing and core case marking.

Other features deal with less obvious phenomena that may be described in more or less detail for some areas of the world, for some language families, in some traditions, or in some decades. This is the case, for example, for light verb constructions, serial verb constructions and clause chaining, associative plurals, inclusory constructions and comparative constructions.

It is good to get in touch with coders or language experts who have worked on closely related languages to understand whether absence of evidence for a certain feature is meaningful or whether it may indicate a gap in the author's analysis.

Coders are encouraged to add more information about their coding in the comment fields. For example, when a coding decision is not straightforward, when a construction does not fit neatly in our comparative concepts but is still related, or if a grammar uses specific terminology. When the coding decision is based on a lack of evidence, this should be mentioned in the comment field.

The sources you use should describe the language you are coding. You should not use inference from statements about the language family at large or from analyses of related languages. If we rely on comparative data, we run the risk of introducing bias into our data and eventually into the analyses that will be run on our data.

Using inference from related languages is not the same as being familiar with a tradition or author and knowing what to expect from a grammar. If a historical linguist remarks that "languages of family X generally do not reduplicate their nouns" this is not sufficient to code 0 for all languages in that family. However, if you are aware that reduplication is a frequently covered topic in a given language family but an extensive description does not mention it, this is a good reason to code 0.