A very common assumption (virtually any work on ergativity could be cited as a source), usually made in categorical form: No language is purely ergative, while it is possible, or supposedly indeed frequent, for languages to be purely accusative. On the other hand, there are patterns, especially ones reflecting the semantic cohesion between verb and arguments (closer with patients than with agents when both are present), in word formation and syntax, which would universally seem to align ergatively, suggesting that alignment mixture is universal. Presumably, for purposes of predicting typological variation as this universal attempts to, only such rules and regularities should be taken into account whose alignment is crosslinguistically variable in the first place.
A very common assumption (virtually any work on ergativity could be cited as a source), usually made in categorical form: No language is purely ergative, while it is possible, or supposedly indeed frequent, for languages to be purely accusative.
On the other hand, there are patterns, especially ones reflecting the semantic cohesion between verb and arguments (closer with patients than with agents when both are present), in word formation and syntax, which would universally seem to align ergatively, suggesting that alignment mixture is universal.
Presumably, for purposes of predicting typological variation as this universal attempts to, only such rules and regularities should be taken into account whose alignment is crosslinguistically variable in the first place.