Introduction
In many of the world's languages—such as Tibetan, Eastern Pomo, and Cuzco Quechua—speakers are grammatically required to specify the source of their information. These morphemes, known as evidentials, indicate whether a speaker saw an event, heard about it, or inferred it from circumstantial evidence. While descriptive linguistics has long documented these systems, formal semantics and pragmatics have struggled to model their behavior. This paper proposes a unified formal model that treats evidentials as "context-shifters" that dynamically adjust the quality threshold—the level of certainty required to make a felicitous assertion.
The Problem: Beyond Descriptive Labels
Traditional accounts of evidentials face three primary challenges that standard linguistic models fail to address:
- Evidential Hierarchies: Why is visual evidence generally perceived as "stronger" than hearsay, and why does this hierarchy occasionally shift depending on the context?
- Interrogative Behavior: How do evidentials function in questions, and why do they shift the burden of evidence to the respondent?
- Descriptive Ineffability: Why are these morphemes notoriously difficult to translate into languages like English without losing their nuanced pragmatic "feel"?
The authors argue that these issues arise because evidentials do not behave like standard adjectives or adverbs. Instead, they operate on the parameters of the conversational context itself.
The Formal Model: Probability and Thresholds
Building on a probabilistic framework inspired by David Lewis, the authors define two key components of the conversational state:
- Subjective Probability ($C_{A,c}$): This represents an agent’s "credence" or degree of belief in a proposition, mapped as a value between 0 and 1.
- Quality Threshold ($c_\tau$): This is a context-dependent value representing the minimum probability a speaker must possess in a proposition to assert it truthfully and appropriately.
The Evidential Strategy
The paper proposes that using an evidential involves a three-step pragmatic process:
- Commitment: The speaker commits to possessing a specific type of evidence (e.g., "I heard that...").
- The Context Shift: The evidential morpheme triggers a change in the quality threshold ($c_\tau$). It moves the threshold to a value that reflects the typical reliability of that evidence type in the current context.
- Speech Act: The speaker performs the assertion or question relative to this newly adjusted threshold.
Key Insights
Evidentials as Goalpost-Movers
A central metaphor of the paper is that evidentials "move the goalposts" of a conversation. By using a "hearsay" evidential, a speaker effectively lowers the required certainty threshold. This allows them to make an assertion they are not 100% certain of without violating the conversational norms of truthfulness. Conversely, "direct evidence" morphemes often "strengthen" an assertion by raising the threshold, signaling that the speaker is holding themselves to an exceptionally high standard of proof.
Pragmatic vs. Lexical Hierarchies
The model suggests that the "strength" of evidence is not hard-coded into the lexicon but is derived from world knowledge and context. For instance, while vision is usually prioritized, a child might have a higher quality threshold for "Mom says..." than for their own fallible eyesight in a specific scenario. The model’s context-dependent threshold shift naturally accounts for these variations.
Evidentials vs. Epistemic Modals
The authors distinguish evidentials from epistemic modals (e.g., must, might). While modals operate within a fixed threshold by inviting the listener to evaluate the proposition against alternative possibilities, evidentials change the threshold itself. This explains why evidentials feel like they modify the "strength" of the speech act rather than just the content of the claim.
Empirical Scope and Results
The theory is supported by data from several linguistically diverse systems:
- Tibetan: Demonstrating direct vs. indirect contrasts.
- Eastern Pomo: Illustrating a complex four-way distinction (sensory, direct, inferential, and reported).
- Cuzco Quechua and Kashaya: Providing evidence for how evidentials interact with assertion strength and interrogatives.
In questions, the model explains "forward-looking interrogatives": an evidential shifts the threshold for the hearer, signaling that the speaker expects an answer based on a specific evidence type (e.g., "Tell me what you saw, not what you've been told").
Conclusion and Future Directions
The "threshold-shifting" model provides a rigorous explanation for the "ineffability" of evidentials; because they change the parameters of the conversation rather than just the propositional content, they cannot be easily replaced by simple translations.
Future work remains in exploring the syntactic interaction of these morphemes, particularly in embedded clauses, and determining if this model can be extended to other utterance modifiers like "Seriously," which may also function by raising the quality threshold of an assertion.
Summary updated