Where Have All the Evidentials Gone?
The case that they are now in case¹

R.M.W. Fillmore
Royal Canadian Mounted Pragmaticists

The pragmatic nuances of grammatical evidential systems in the world’s languages are notoriously difficult to analyze (Bowen, personal correspondence), and yet cross-linguistically they display a remarkably coherent set of semantic features (Aikhenvald, p.c.). Furthermore, these systems have repeatedly been shown to be nearly impossible to borrow or to spread through other forms of language contact (Thomason, p.c.), which suggests that they must be categories of considerable antiquity in the languages that have them (पाणिनि, p.c.). These factors, plus the fact that evidential systems occur in virtually all the world’s language families (Greenberg, p.c.), have led such eminent linguists as Campbell (p.c.) to conclude that “evidentials clearly represent a shared retention in all modern languages,” and others such as Hock (p.c.) to suggest that “an elaborate evidential system must therefore be reconstructed for Proto-World.”

There are, however, many modern languages which reportedly do not have any grammatical evidential markings (Grammars, p.c.). Since we can be confident that evidentials did exist in some ancestor of each living language (Pinker, p.c.), we are faced with the puzzling issue of how some languages managed to lose a feature which clearly belonged to the universal parameter settings found in the original human language (Lightfoot, p.c.). What happened in those descendant languages that do not now have evidentials? Where, in short, did those evidentials go?

Many have sought an answer to this question (Croft, p.c.) but no one to date has proposed a plausible solution (Whaley, p.c.). This is because they have been looking in the wrong places.

For this study, I collected a random sample of 374 published grammars, representing languages from all known language families and nearly every continent of the known world (Dryer, p.c.). In-depth study of these grammars revealed a surprising type of morphological complementary distribution (Anderson, p.c.). It turns out that grammatical evidential systems do not co-occur with grammatical case systems. Any given language may have one or the other, but not both.

This, then, is the surprising answer to the question: languages which do not now have evidential systems are lacking these systems because they have reanalyzed earlier evidential markers as casemarkers.

I call this “surprising” because we do not normally expect grams associated with verbal categories to be reanalyzed as markers of nominal expressions (Bybee, p.c.). Fortunately, we professional linguists are not prisoners of our expectations, but are instead free to let the data guide us where it will (Kuhn, p.c.). And in this case, the data shows clearly that casemarking arose as a reanalysis of earlier evidentials.

Correlating cross-linguistic complexity of evidentials and casemarkers by means of sophisticated statistical software (Cedergren and Sankoff, p.c.), I obtained clear correspondences between specific evidential categories and the casemarkers into which they were reanalyzed. These correspondences turn out not to be surprising at all, as each displays a consistent kernel of semantic information (Wierzbicka, p.c.). The correspondences are displayed in Table 1.

Evidential category	Casemarker which this category is reanalyzed as	Comments
Default (unmarked)	Absolutive or Nominative	The unmarked evidential fits most naturally with the unmarked nominal in either system of case alignment (Blake, p.c.)
Direct	Ergative	Ergative arguments have the most direct effect on the action of the clausal verb (Dixon, p.c.)
Indirect	Accusative	The accusative marks the participant who can only watch as the action of the verb takes place (Payne, p.c.)
Inferential	Genitive	An observer can only infer who might be the possessor of any entity; possession relationships are never visible (Jackendoff, p.c.)
Validational	Dative	The recipient (the prototypical dative argument) is the witness who can validate any claim that I make (Quirk, p.c.)
Other evidentials?	Oblique cases	[see comment below]

Table 1: Evidential categories and their resulting casemarkers

Special mention needs to be made of the “Oblique” casemarkers (Comrie, p.c.). Computational analysis did not turn up any reliable correlations between oblique cases and evidentials in the world’s languages. There are two possible explanations for this (Logical inference, p.c.). One is that the earliest human language had additional evidentials which have disappeared in all modern languages (Lakoff, p.c.). The other possibility is that obliques are not casemarkers at all (Postal, p.c). I am inclined to accept the latter explanation, as it confirms the predictions of Perlmutter (p.c.). Also, the former explanation is implausible.

Readers familiar with evidential systems may note the omission of the common Hearsay category from this chart (Leech, p.c.). The fact is that hearsay evidentials are not actually evidentials, but rather constitute denial that evidence exists (Jakobson, p.c.). We might more properly term this category a disevidential (Chafe, p.c.); this is undoubtedly why statistical analysis did not turn up any corresponding casemarker.

As I have already mentioned (Myself, p.c.), it is a bit surprising to find markers associated with verbs reanalyzed as markers associated with nouns. However, this is actually in line with the basic predictions of grammaticalization theory (Hopper and Traugott, p.c.), which posits a process (Heine and Kuteva, p.c.) like (1):

(1) verbal morphology > nominal morphology

Thus, the reanalysis of evidentials as casemarkers is not really surprising at all!

Conclusion

Evidentials represent one of the original parametric categories, present in the earliest human language (Bickerton, p.c.). We have seen here that their morphological markers persist in all human languages, but that these markers have been reanalyzed as casemarkers in many languages.

Although case systems are more familiar to the modern linguist, our analytical forebears in prehistory must have had no case systems to consider at all, but instead had quite a rich system of evidential distinctions around which to wrap their explanatory theorizing. This makes intuitive sense (Grice, p.c.): when your interlocutor is brandishing a club, the justification for your claim may be somewhat more urgent than the mere trivialities of who did what to whom.

	Google Results Support an Anglo-Genesis Model of Language Degradation—U. Pendleton Start
	Augmented Things You Didn’t Know You Didn’t Know—Madalena Cruz-Ferreira
	SpecGram Vol CLXII, No 4 Contents