In classical modal logic, we typically reason using "possible worlds." A possible world is a total snapshot of reality: for every conceivable proposition, that proposition is either definitively true or definitively false. While mathematically elegant, this "totality" assumption creates significant friction when modeling human language, database gaps, or agents who are simply unaware of certain concepts.
Recent research, led prominently by Wesley H. Holliday, proposes a shift from total worlds to partial possibilities. This framework, known as Possibility Semantics, treats information states as first-class citizens, allowing us to model what we don't know as naturally as what we do.
The Problem with Totality
Standard Kripke semantics requires a model to decide the truth value of every atomic sentence at every point. As Wesley H. Holliday argues in Partiality and Adjointness in Modal Logic, this is poorly suited for scenarios where information is incomplete.
Consider the sentence "It might be raining." In a total-world model, "might" is usually handled by looking at other accessible worlds. But this requires us to have a complete map of all possible totalities. In contrast, human communication often operates in a state of partiality. As Alan Black notes in Embedding DRT in a Situation Theoretic Framework, human language is inherently tied to specific contexts or "situations" where most facts about the universe are simply undefined.
Possibilities as Information States
The core intuition of possibility semantics is to replace the set of worlds with a partially ordered set (poset) of possibilities.
- Partiality: A "possibility" is a state of information that can leave some propositions undecided.
- Refinement: The partial order ($\leq$) represents the growth of information. If state $X \leq Y$, then $Y$ is a "refinement" of $X$—it contains all the information $X$ has, plus potentially more.
This structure allows for a more nuanced definition of truth. In Possibility Semantics, Wesley H. Holliday explains that a disjunction ($A$ or $B$) can be "forced" by an information state even if neither $A$ nor $B$ is individually settled yet. We only require that every further refinement of our current state eventually leads to a state where either $A$ or $B$ is true. This captures the "inquisitive" nature of knowledge progression.
Solving Linguistic Paradoxes
This shift to partiality solves long-standing puzzles in epistemic logic. Matthew Mandelkern and Wesley H. Holliday use this framework in The Orthologic of Epistemic Modals to address "Wittgenstein sentences" (e.g., "It is raining, but it might not be").
In classical logic, these sentences are technically consistent but feel like contradictions. By moving from Boolean algebra to ortholattices—which are better at representing compatibility between partial states—the authors create a logic where these sentences are genuine contradictions. This "Epistemic Orthologic" preserves classical reasoning for facts while correctly modeling the "non-classical" behavior of words like might and must.
Beyond Logic: Databases and Unawareness
The utility of partial states extends into practical computer science and economics:
- Database Theory: When a database is missing values, how do we define a "certain answer"? Leonid Libkin in How to Define Certain Answers and Egor V. Kostylev et al. in A Note on Computing Certain Answers to Queries over Incomplete Databases use an "informativeness preorder." A database state is "less informative" if it has more possible completions. This allows researchers to treat query results as partial objects that can be refined as more data arrives.
- Modeling Unawareness: Traditional models struggle with agents who are unaware that a variable even exists. In A partial-state space model of unawareness, Wesley H. Holliday shows that by using partial specifications, we can represent an agent's limited conceptual framework. If a state doesn't mention a variable, the agent isn't just "uncertain" about it—they are entirely unaware of it.
Why It Works
Possibility semantics works because it bridges the gap between syntax (the formulas we write) and semantics (the models of the world). By treating information states as "first-class citizens," as suggested by Jørgen Villadsen in Information States as First Class Citizens, we no longer need to invent infinite sets of total worlds to explain simple uncertainty. Instead, we can build models directly from the partial information available to us, leading to logics that are both mathematically more robust and more aligned with how humans actually reason.