Ahmed H. Alhuwayshil is a PhD researcher in Linguistics at the University of Edinburgh. His research focuses on scalar implicature, using tools from formal semantics, logic, and cognitive science, alongside experimental methods. His research engages with topics in semantics, pragmatics and philosophy. He holds an MA from UCL and previously lectured at King Faisal University. At The University of Edinburgh, he teaches at the School of Philosophy, Psychology and Language Sciences and is a PGR student representative. He is also an abstract reviewer for the postgraduate student session at the European Summer School in Logic, Language and Information (ESSLLI) 2025 in Germany, and a member of the organizing committee for the PGC 2025 conference. 

When someone says:

“some elephants are mammals”

we often infer from this statement that not all elephants are mammals, even though this contradicts our world knowledge. This inference is called a scalar implicature. It is not part of the literal meaning of some. Scalar implicatures are inferences we make about quantities or degrees when the speaker uses a term like some instead of a stronger alternative like all. So why do we make this inference, how can we formalize it, and how does it emerge in our cognition? In formal semantics, the literal meaning of some can be modelled as:

∃x (elephant (x) ∧ mammal(x))

meaning there exists at least one elephant that is a mammal. The inferred meaning adds:

¬∀x (elephant (x) → mammal(x)) i.e., not all elephants are mammals

So, the full enriched interpretation becomes a conjunction:

(∃x (elephant(x) ∧ mammal(x)) ∧ ¬∀x (elephant(x) → mammal(x)))

“at least one elephant is a mammal, and not all elephants are mammals”

The literal meaning of some is more neutral, while the enriched meaning reflects our tendency to assume that when some is used, the speaker is intentionally not saying all.

There is an ongoing debate in semantics and pragmatics about how and when this inference arises. Default theorists argue that scalar implicatures (the enriched meaning) are automatic. They are built into the way we process scalar terms like some as a default part of language understanding. Contextualist accounts by contrast, argue that scalar implicatures are not automatic but arise only when the context highlights a relevant and stronger alternative such as all.

My research targets a non-standard case where these two theories diverge. Specifically, I examine cases where the not all inference should be shared knowledge. For instance, consider a context where someone says, “John ate some of the cookies.” In particular, situations where both the speaker and listener already know that “it is not the case that John ate all of the cookies” to be true. In such a case, the implicature not all seems unnecessary or even misleading. Yet, if scalar implicatures still arise in these cases, it could indicate that this enrichment is a default cognitive process.

This question goes beyond a technical puzzle in semantics. It has broader implications for understanding the nature of meaning in human cognition. While this research is grounded in questions of meaning and interpretation within semantics and pragmatics, this research could also offer insights into how meaning is represented and processed cognitively. It could reveal whether scalar implicatures behave in the same way even in non-communicative cognition. For instance, if someone is considering how much of a task they have completed and they think, “some of the work is done,” do they automatically interpret this as not all? Are they computing these implicatures even when there is no listener involved?

The research examines whether such inferences are a basic part of cognitive processing and may also offer insights into understanding how they occur even when we are engaging in internal self-talk. This is a crucial step in expanding our understanding of how meaning works not just in spoken language, but also in the language of thought. The relevance to thought lies in the concept of shared knowledge manipulation. In communication, people rely on a common ground, which refers to the set of shared knowledge that both speaker and listener assume to be true. In thought, people are assumed to have common ground with themselves.

By refining our understanding of scalar implicatures in both communication and cognition, this research advances broader efforts to map how humans compute meaning and interpret language. The insights have practical implications for linguistics, philosophy, cognitive science, and even politics, where the way we frame our words can subtly shape interpretation and create ambiguity. Are we aware, when choosing our words, that we may be influencing meaning or leaving it deliberately open?

An experiment was conducted to empirically investigate the real-time processing of scalar implicatures, including cases where the implicature was already part of the common ground. Participants listened to sentences and were asked to select the corresponding image, while their mouse trajectories were continuously recorded to capture the dynamics of the decision-making process.

To examine real-time interpretation, mouse movements were analysed using GAMM, which tracks how paths deviate from a straight line over time. The results showed clear differences between conditions. When the relevant information was made common ground or when the sentence supported an implicature, participants’ mouse paths almost aligned and were much closer to the ideal trajectory (see figure 1).

In contrast, when the implicature was absent, their movements were more indirect and far from the ideal path. These differences were statistically significant, with extremely low p-values (<2e-16). The shape of the mouse paths also changed over time in complex but reliable ways depending on the condition, revealing how the interpretation process unfolds moment by moment (see figure 2).

The findings from this research will contribute to refining theories in formal semantics and pragmatics, particularly on theoretical issues involving scalar implicatures, tense, and aspect, while also shedding light on how meaning is processed in the brain, specifically by revealing the processing patterns and interpretive mechanisms involved. These findings can also be used to test and train AI systems that aim to detect and interpret implicit inferences across multiple languages.