When we read a sentence, our brains do not just extract raw facts; we construct a mental model of causality, intent, and blame. Consider two ways of reporting the exact same physical event:
- “A driver hit a cyclist.”
- “A collision occurred involving a cyclist.”
Both statements can be factually accurate, yet they leave readers with fundamentally different impressions. The first assigns clear agency to an actor, while the second frames the event as a spontaneous phenomenon. This phenomenon is known as linguistic perspectivization—the strategic use of grammatical and lexical structures to shape how readers attribute responsibility, focus, and blame.
Recent computational linguistics research demonstrates that bias in media and discourse frequently resides not in the inclusion of false information, but in these subtle constructions of causality. By altering the syntactic scaffolding of a sentence, writers can systematically manipulate accountability without ever telling a literal lie.
The Syntactic Scaffolding of Blame
Why does changing the voice of a sentence alter our perception of responsibility? The answer lies in how grammatical roles map to cognitive focus.
In an active transitive structure ("The perpetrator attacked the victim"), the agent occupies the subject position. This structural prominence naturally maximizes the reader's focus on the actor, increasing the perceived blame attributed to them. As shown by Sara Gemelli et al., human perception of responsibility is highly sensitive to this distinction; active verbs anchor the blame firmly onto the perpetrator.
Conversely, when a sentence is flipped into the passive voice ("The victim was attacked"), the agent is demoted or removed entirely. This structural shift moves the focus away from the actor and toward the recipient of the action. Caleb Ziems et al. highlight that the "agentless passive" is a powerful tool for minimizing accountability in reports of violence, as it completely erases the party responsible for the action from the grammatical structure.
Beyond the passive voice, writers can use even more extreme structural erasures:
- Nominalization: Turning a dynamic action into a static noun (e.g., changing "X shot Y" into "The shooting occurred"). As Ghizlane Bourahouat et al. observe, nominalization strips away the explicit actor, making it difficult for the reader to assign accountability and often introducing an unintended bias against the victim.
- Unaccusative Verbs: Verbs that describe an event without an inherent agent (e.g., "The window broke" or "A collision occurred"). Sara Gemelli et al. note that these verbs structurally necessitate a focus on the affected party, naturally obscuring agency because the grammar itself prevents the mention of a responsible actor.
Lexical Lenses: Reporting Verbs and Mind Attribution
Perspectivization does not stop at syntax; the specific classes of words chosen to describe actions inject powerful presuppositions into a text.
Reporting and Attitude Verbs
When journalists introduce a quote or a claim, their choice of reporting verb acts as a primary driver for how readers judge its factuality. Writing that a politician "said" something is neutral. Writing that they "insisted" or "boasted" introduces what Tiancheng Hu et al. call "quotative bias," signaling a speaker's emotional state or hidden intent.
Remarkably, Sandeep Soni found that these lexical framing choices outweigh the actual content of the claim or the credibility of the source when readers judge factuality. Similarly, the choice of propositional attitude verbs—such as "knowing" versus "believing"—significantly alters an argument's perceived strength by toggling between factual certainty and mere consensus, as explored by Mark-Matthias Zymla et al..
Anthropomorphism and Agency
The choice of verbs can also artificially inflate or deflate the perceived agency of non-human entities. In technical writing about artificial intelligence, researchers frequently use mentalizing verbs like "thinks," "knows," or "considers."
Susanne Hindennach et al. demonstrate that this language creates a "responsibility-concealing" effect. By attributing a mind and awareness to the software, the language shields human developers from accountability when systems cause harm. Myra Cheng et al. track this trend computationally, showing that cognitive or affective verbs (like "struggling") drive a highly anthropomorphic perception of agency, whereas neutral verbs (like "performing") maintain objective framing.
Cognitive Smuggling via Pragmatic Framing
Why are readers so susceptible to these framing strategies? The answer lies in how our brains process background information versus direct claims.
When a statement makes a direct assertion ("The government failed to act"), the reader can consciously evaluate and challenge it. However, writers can use pragmatic framing to smuggle assumptions into the background text where they are harder to notice. Qi Yu demonstrates this using "presupposition triggers." Adding a single word like "again" ("Again, refugees drowned") forces the reader to accept the underlying narrative of a recurring, systemic failure just to make sense of the sentence. Because this information is presented as a taken-for-granted background assumption, it bypasses the reader's immediate critical defenses.
This selective construction of causality allows third parties to systematically entrench biases and set agendas. As Wenyu Zhang notes, controlling the narrative frame dictates how an individual interprets the character and actions of others without requiring direct evidence.
Quantifying Perspective at Scale
Understanding the mechanics of linguistic perspectivization allows researchers to move beyond simple sentiment analysis (which merely measures positive or negative words) and build tools that map the deeper social dynamics of text.
For instance, the strategic use of pronouns can be tracked to see how individuals establish distance or claim authority. Minh Nhat Vu shows how survivors of violence use first-person pronouns ("I," "me") to reclaim identity and center their experiences, while using direct second- or third-person pronouns to enforce accountability on perpetrators. Conversely, Gabriel Lima et al. demonstrate that shifting text from rights-based language to duty-based language changes public acceptance of controversial concepts by redirecting focus from an agent's entitlements to an observer's obligations.
Even seemingly minor linguistic elements, like prepositions, reflect these deep cognitive shifts. Hatte Blejer et al. observed that when the perceived animacy or agency of a subject drops, speakers naturally substitute prepositions (such as using "with" instead of "by" in passive constructions), proving that our grammatical choices are continuously tuned to the underlying power dynamics we perceive.
By formalizing these insights into computational frameworks, modern NLP systems can automatically detect point-of-view differences across documents (Khalid Al Khatib et al.) and track how power and social dynamics evolve in narratives (Maria Antoniak et al.). As the field moves forward, creating automated methods to evaluate these semantic structures remains vital for uncovering how media narratives subtly shape public perception of accountability and justice (Eleonora Cappuccio et al.).
Key Foundational Work
On Representing Governed Prepositions and Handling "Incorrect" and Novel Prepositions (1989) — contributed 1 concept elements to this direction.
Structure of Perspectivity: A Case of Japanese Reflexive Pronoun "zibun" (1990) — contributed 1 concept elements to this direction.