Crosslinguistically, evidentials differ along certain parameters (Schenner, 2008), among them embeddability. Focussing on two particles, German wohl and Tagalog yata, and following Rooryck (2001)'s conception of evidentiality, we argue that both particles should be considered evidentials. This is based on the observation that they indicate both source and reliability of information, since adding them to a sentence expresses that the speaker has some piece of evidence to believe the propositional content of the utterance but is uncertain about it. However, while both contribute a similar meaning to their host utterance, their distribution in embedded clauses sets them apart.

Contrasts in embedding have led to a treatment of evidentials as either illocutionary modifiers (Faller, 2002) or epistemic modals (Izvorski, 1997; Matthewson et al., 2007). In this talk, we show that wohl can only be embedded in a subset of the contexts in which embedding yata is possible. This behaviour is expected since wohl, as discourse particle or illocutionary modifier, depends on the presence of illocutionary force, entering a syntactic agreement relation with it (Coniglio and Zegrean, 2012), while yata, being an epistemic modal, does not. Thus, we predict that wohl is found in those embedded clauses that have independently been argued to contain illocutionary force (Haegeman, 2006) while the distribution of yata is less restricted.

Our analysis has two important consequences. First, despite similar meaning contributions, we provide support for the idea that evidential markers need a non-uniform treatment as either illocutionary modifiers or epistemic modals, considering wohl and yata's embeddability. Second, we show that, following Faller (2014), constraints on embeddability are a valid diagnostic to distinguish these types, as the speech act operator type of evidentials are dependent on illocutionary force and are thus much more restricted in their distribution in embedded clauses than the less limited epistemic modals.