Usually, we think of grammar as just the invisible seams in a sentence. It’s just the list of some arbitrary rules that keep nouns and verbs functioning so we can convey meaning.
But what if grammar actually has a social side, a side that alters how we judge human character? What if we are to determine the level of a person’s agency just from the way they speak and not the declarative and descriptive layers of the language?
In fact, linguistic agency operates on both levels. First, the declarative. Here you have the language that focuses on the actual events that convey agency. It dictates how we project power, how we influence social judgement, and how we shape the perception of our competence.
Beneath that is the second level, which is structural. It refers to the formal logical rules of how actions exist in time and space within the language itself. This level also determines if the action even allows a person to be an independent doer or if they’re just a passive undergoer of an event.
However, to understand how they interact, we probably need to first establish the baseline of that social side first – to see why the language is even the medium to decipher agency in the first place. And here comes the language of the big two.
The Big Two: Agency and Communion
Following Bakan in 1966, Abele and Brückmuller defined ‘agency’ and ‘communion’ as the two main lenses through which human beings “see” every single person they encounter. The research defines “agency” as a person’s striving for success — their autonomy, competence, and goal orientation. (Abele & Bruckmüller, 2013) It’s the desire to prove yourself, to take charge of your environment. If someone is ambitious, highly competent, or relentlessly self-confident, they are radiating high agency.
‘Communion’, by contrast, means warmth, trustworthiness, friendliness, and the general maintenance of social connections. It is all about cooperation and group harmony. Interestingly, the research places morality under the umbrella of communion as well, because being moral usually means acting in ways that preserve the trust and safety of the group. In short, agency is the drive to get ahead, while communion is the drive to get along.
An interesting thing happens, though, when Abel and Brückmuller introduce the Dual Perspective Model (DPM). The DPM reveals that while agency is the dynamic dimension that actually drives action in the world, communion is the primary perspective in human judgement — primary in the sense that it is the absolute first thing we filter for when a new person enters our environment.
When you are the one looking at or listening to someone else, you process information about their communion much faster, and you assign it significantly more weight than any information you process about their agency.
This naturally links to an evolutionary threat assessment. First, you scan for communion: is this person a friend or a foe? Are they safe to approach? Only then do you scan for agency, because if they fail the communion test and they are a foe, you quickly need to know how capable they are of hurting you. And if they pass the communion test and they are a friend, you want to know how capable they are of helping you.
What is also interesting is that this, despite what it may look like, is a global and not culture-dependent phenomenon.
Hofstede’s cultural dimensions theory links cultural individualism to agency while linking cultural collectivism to communion, i.e., the West being highly individualistic and agentic versus the East being more collective and communal. But the findings from researchers like Abele and Brückmuller are that the primary scanning for communion operates universally and was documented across English, French, German, Italian, Polish, and various Asian languages.
Regardless of the cultural baseline of individualism versus collectivism, human beings naturally categorise each other along these two axes, with communion being the first.
The DPM also explains everyday social misunderstandings. Receivers focus primarily on scanning for communion, but message senders – the people speaking and doing things – focus mostly on projecting agency because most people want to be seen as competent and autonomous.
Consider a classic scenario: you walk into a performance review or a job interview. You lean into assertive, highly agentic language because you want the room to think you are a strong, capable leader.
But the entire room is operating from the observer perspective. Their brains are running the communion scanner, looking for warmth and trustworthiness first — and you are starving them of that data. You might be successfully communicating competence, but for the listeners, that agency reads as coldness, uncooperativeness, or outright unfriendliness. Quite a mismatch, isn’t it.
But here comes a question – how do we project the communion or the agency when in these kinds of situations? Would smiling more make a difference? To some extent it might, but the delivery mechanism relies a lot on the language.
The Verb-Agency Link
To be more specific, the mechanism relies on one speech element – verbs. This comes from research done by Formanowicz et al. (2017) that discusses what they call the verb-agency link.
The core argument is that verbs are the inherently dynamic grammatical category. Unlike adjectives and nouns, which are designed to describe static or long-term qualities, verbs can act as linguistic markers of movement irrespective of their actual semantic meaning. The grammatical categorisation of a word alone makes the brain perceive it as more agentic, even if the word itself is complete gibberish.
The Polish pseudo-word experiments conducted by Formanowicz and her team proved exactly this. To demonstrate a pure grammatical effect, the researchers had to strip away semantics, since natural language carries historical baggage and cultural associations.
They generated completely fake words, ensuring these words followed a phonetically natural Polish structure but meant absolutely nothing — testing the impact of syntax on the human brain.
Taking a fake root word like ‘frall’ (to make it a suitable English-language example), they then attached specific morphological suffixes to force it into different grammatical categories: a pseudo-verb (to frall), a pseudo-adjective (frallative), and a pseudo-noun (a frall). The semantic meaning of ‘frall’ was none; the only variable was the grammatical bucket the word was placed in.
What turned out was that participants consistently rated the pseudo-verbs as significantly more agentic than the pseudo-adjectives and pseudo-nouns. Apparently, the brain identifies verb structure and automatically links it to action, competence, and striving, even when the word itself doesn’t exist in the dictionary. The grammar itself carries the weight of the meaning.
To test whether this lab finding held in real life, researchers moved from controlled experiments to analyse massive databases of actual, naturally produced language in both German and Polish and how those structures are associated with the demographics.
We know that society typically views men and young people as highly agentic, while stereotyping women and older people as less agentic and more communal. And the archival research confirmed that our grammar naturally mirrors those biases. In the Polish corpus, the stereotypically agentic target — young people — was almost three times more likely (2.89 times) to be placed immediately next to a verb than the non-agentic target older people. The pattern held for men versus women across both German and Polish.
People naturally and likely entirely unconsciously pair age and gender with verbs or adjectives and nouns to match the expected agency.
But there is more here – not all verbs are equal, and the type of the verb is also at play here.
Aktionsart: How Actions Exist and Unfold in Time
Let’s go beyond social psychology into the role of the verbs themselves. The core concept here is Aktionsart — a German term translating as “types of action”. This is how verbs are structured in relation to the time of the action they describe. Building on the foundational work of Vendler (1967), Dowty (1979), and Hana Filip’s overview, Aktionsart breaks all verbs into four distinct aspectual classes: states, activities, accomplishments, and achievements. (Filip, 2020)
States are simply conditions of being. Verbs like ‘know’, ‘own’, ‘love’, or ‘exist’ are inherently non-agentive and unchanging. If you know how to speak English today, you knew it yesterday — there is no ongoing dynamic action required to maintain that state. A state has no inherent duration and no logical endpoint. It just is.
Activities last over time, but they do not have a natural logical endpoint built into them. Verbs like ‘run’, ‘walk’, or ‘push’: you can run for ten minutes or ten days. The action of running itself does not contain its own finish line. In other words, an activity is a never-ending GIF animation: the character is breathing or jogging in place, expending energy but not progressing toward a specific goal.
The shift comes with the introduction of telicity — the concept of an action having a natural goal, a climax, or an end state. Accomplishments are telic: a process that logically must lead to a definite end. Examples include ‘build a house’, ‘draw a circle’, or ‘eat an apple’. Once the apple is eaten, the action is over; the object is completely affected, and the event reaches its conclusion. An accomplishment, therefore, is a progress bar: you are forging a win, the bar slowly fills, and there is a definite end – the action becomes the win.
Finally, achievements are also telic, but they also change the subject’s state in the blink of an eye. Verbs like ‘realise’, ‘win a race’, or ‘notice’ occur in an instant. You do not spend an hour progressively noticing a mistake. So an achievement is a rapid change: it happens in one single frame in which the character’s status is permanently altered.
What is interesting here, though, is that human language structures offer diagnostics to test for agency level – such as the accomplishments or activities, not just states. They will literally reject non-agentive objects if we attempt to force them into agentive tasks.
A classic diagnostic test in English is checking a verb’s compatibility with the progressive aspect and the agentive ‘be’. Consider: ‘John is being noisy.’ The brain accepts this effortlessly because John is a human with his own control — he is choosing to act noisily, and that’s what “being” suggests: a choice. But apply the same structure to an inanimate object: ‘the motor is being noisy’. A native English speaker finds it odd or treats it as an act of animation. You would simply say ‘the motor is noisy’. The progressive ‘is being’ requires an agent with autonomous control. Because a motor lacks its own agency, it fails the diagnostic test. The language literally refuses to allow a machine to act like an agent.
Another example of it is loosely linked to the agentic shift. Consider the difference between ‘I was asleep’ versus ‘I went to sleep’. The first uses ‘to be’ and is entirely static — it describes a passive condition where the person is doing nothing to affect their environment, avoiding the language of decision-making. But ‘went to sleep’ is an active construction, utilising a movement verb to imply an active autonomous transition — a change of state initiated by an independent agent. Even though the biological reality is identical (a person was lying down asleep), the linguistic framing shifts them from a static object to a dynamic actor. Linguists who study Role and Reference Grammar (RRG) map this visually. (Martín García, 1996; Osswald & Düsseldorf, 2021).
Human agency is not just a psychological concept; it is built into the logic of our syntax. You need a doer to do.
Weaponizing Passivity
But the language can build an agent just as it can be weaponised to intentionally erase one.
When analysing the lack of agency, we are looking at passivity, linguistic externalisation, and the consequences of changing causality framing.
The Fausey and Boroditsky study on the 2004 Super Bowl halftime show wardrobe malfunction is the exact study to use when trying to understand this linguistic framing. The researchers wanted to see if altering the syntax could actually alter the perception of guilt. They had participants read two slightly different descriptions of the exact same event. One group read an agentive, active description: ‘Timberlake tore the bodice.’ The second group read a non-agentive passive framing: ‘the bodice tore.’
The results of that simple syntactic shift were pretty significant. Participants who read the agentive framing — the sentence that linguistically mapped Timberlake to the action — suggested financial fines that were 53% higher than the group that read the passive framing. (Fausey & Boroditsky, 2010)
Just removing the agent from the subject position of the sentence reduced imaginary fines by over half, showing how the brain processes cause and effect based on sentence structure and the explicit placing of a subject.
So when public figures use passive language to deflect blame, they are not just being strategically evasive — they are actively exploiting a cognitive glitch in how the human brain processes cause and effect.
Linguistically, that distortion also translates into a diminished sense of agency and subjective responsibility. When someone uses passive language – when they say ‘mistakes were made’ – it is often an accurate linguistic reflection of their distorted cognitive reality. They are using syntax to distance themselves from the event itself but also from their own control of the situation and therefore also the guilt. (Nook et al., 2017)
So passive voice is a language of both – the victims and those who want to be perceived as victims of the external circumstances they had no control over. By handing over the control, they can remain morally righteous.
Interestingly, the language also decides what we focus attention onto and what we remember better. (Fausey et al., 2010) Agency (and therefore perpetration) is not an objective, universal feature of how humans process events. Instead, the grammatical structures we use act as an attentional framework. So when someone uses passive voice, removing themselves from the action, they also unconsciously hope that the listeners will not remember their participation in those (unfortunate) events.
The Architecture of Linguistic Agency
All of this theory merges into practical rules for analysing speech — your own and others’.
Consider the classic corporate performance review. The manager opens by saying, ‘You are a very collaborative team member, and we value your presence, but the Q3 deliverables were delayed, and we need to see more impact.’
First, the manager called you ‘collaborative’ and valued your ‘presence’ – heavily emphasising semantic communion and static being, defining you by your warmth and relationships rather than your competence.
Second, the syntactic vehicle ‘you are a team member’ uses a static noun phrase and the verb ‘are’ to define your identity, completely avoiding any dynamic verbs that would highlight your actual actions.
Third, the framing ‘The Q3 deliverables were delayed’ is a passive construction – the agent is removed entirely. The deliverables just magically delayed themselves. The ‘do’ operator has been completely unmapped. This is no-blame and likely the only positive in the whole statement.
Yet, it already looks like carving a path towards letting you go or forcing another decision onto you. You’ve clearly been turned into an object rather than an actor (which also tells you a lot about the speaker, does it not).
So as the employee, you should reclaim your agency but not by saying ‘I am a very hard worker’ — that just reinforces the static noun framing. Instead, you should deploy an accomplishment verb: ‘I engineered the new database architecture, which secured the client account.’
Let’s break down why it works: ‘engineered’ is an agentic verb, placing you as the animate subject mapped to the ‘do’ operator; ‘the new database architecture’ is a specific affected object serving as an opposition to you being the agent; ‘which secured the client account’ frames the activity as a telic achievement with a definite punctual end state.
The syntax forces the manager to acknowledge your agency. It has a chance to work, as long as you don’t overdo it by throwing more verbs blindly into it. Context is key here.
Why You Can’t Just Verbify Everything
If verbs inherently project competence and agency, the obvious conclusion would be to speak almost entirely in verbs.
In fact, the most successful marketing agencies, politicians, and PR firms often lean heavily into them. The most memorable political and marketing slogans of the last few decades are almost entirely propelled by verbs: ‘Yes we can’ and ‘Make America Great Again’. ‘Vote “Leave”.’ ‘Just do it.’ Verbs inspire action because they are action. (Formanowicz et al., 2021)
However, context is the ultimate moderator. Leaning too heavily into agentic language can actually backfire for some entities. When an observer looks at a corporation, a politician, or a leader, the default societal assumption is that they are already highly agentic – they generate profit, achieve strategic goals, and dominate competitors. No one doubts their agency.
So if they spend their marketing budget shouting about how agentic they are — ‘We build, we drive, we dominate’ — the audience gets a lack of balance. The observer’s brain is still looking for communion, asking, ‘Do you care about the community? Are you ethical? Are you trustworthy?’ Because if you have all that power but no warmth, you are a threat. So sometimes you may actually benefit far more by utilising communal messaging.
Conversely, NGOs and charities are assumed to be highly communal and caring by default — but people may worry they’re ineffective. So NGOs often benefit from injecting more agentic verbs into their appeals to demonstrate they can actually execute on their goals. This is a linguistic compensation effect: you use your grammar to supply the specific dimension that your brand or identity is naturally lacking in the eyes of the observer.
The contextual catch applies here too. In a highly collaborative NGO environment, screaming in first-person agentic verbs might read as tone-deaf. Communal organisations respond far better to collective agency: ‘We engineered the architecture together.’ In a hyper-competitive corporate environment, personal agency dominates. The rule is to read the room before selecting your approach – knowing precisely when to pass a verb and when to soften the syntax.
And it becomes even more difficult when you are a woman.
Society places heavy communal expectations on women, so when a woman utilises highly agentic language – when she adopts the exact same verb-heavy assertive linguistic framing as her male counterparts – she frequently faces severe backlash. She is judged more harshly than a man using the exact same syntax, told she is too tough or abrasive because her grammar is agentic. But if she retreats to adjectives and nouns to sound more communal, she is viewed as too soft or lacking leadership potential.
It is a highly restrictive linguistic double bind: penalised for the verbs but passed over for the nouns. It is known as the ‘Goldilocks Dilemma’ — trying to find the linguistic approach that is just right while the zone of acceptable expression remains deliberately narrow.
Translated to a modern workplace context: when listening to a young male executive, the brain is statistically more likely to accept a verb-based structure: ‘I lead’, ‘I innovate’, and ‘I change’. But for an older female executive, the more acceptable way to describe oneself might be by defaulting to a noun or adjective – ‘I’m a leader’ or ‘I’m innovative’.
Sometimes the best way to win is not to fight, but to accept the rules and bend them to your advantage.
The language we use is not there just to communicate the facts. There is much more than that in it, and – as the research shows – one doesn’t need to be into linguistic analysis to understand the less explicit meaning and even derive conclusions about someone’s traits based on the language. So the question that follows is likely unavoidable: as we use our own language to teach AI, the implications of which we still do not fully understand ourselves, how will the hidden social grammar shape how these systems understand their own responsibility? If human beings naturally use passive syntax to deflect blame and externalise coercion, what kind of language will an autonomous AI use when it makes a catastrophic error? Will it speak in telic accomplishments and own the ‘do’ operator – or will it tell us ‘mistakes were made’? It will be interesting to see how our own linguistic structures will be thrown right back onto us.
—
Sources:
Abele, A., & Bruckmüller, S. (2013). The Big Two of Agency and Communion in Language and Communication (pp. 173–184).
Bakan, D. (1966). The test of significance in psychological research. Psychological Bulletin, 66(6), 423–437. https://doi.org/10.1037/h0020412
Dowty, D. R. (1979). Word Meaning and Montague Grammar. The Semantics of Verbs and Times in Generative Semantics and in Montague’s Ptq. Journal of Symbolic Logic, (2), 501–502. https://doi.org/10.2307/2273576
Fausey, C. M., & Boroditsky, L. (2010). Subtle linguistic cues influence perceived blame and financial liability. Psychonomic Bulletin & Review, 17(5), 644–650. https://doi.org/10.3758/PBR.17.5.644
Fausey, C. M., Long, B. L., Inamori, A., & Boroditsky, L. (2010). Constructing Agency: The Role of Language. Frontiers in Psychology, 1, 162. https://doi.org/10.3389/fpsyg.2010.00162
Filip, H. (2020). Lexical Aspect (Aktionsart) (pp. 1–61). https://doi.org/10.1002/9781118788516.sem038
Formanowicz, M., Pietraszkiewicz, A., Roessel, J., Suitner, C., Witkowska, M., & Maass, A. (2021). “Make it Happen!” Social Psychology, 52, 1–15. https://doi.org/10.1027/1864-9335/a000435
Formanowicz, M., Roessel, J., Suitner, C., & Maass, A. (2017). Verbs as linguistic markers of agency: The social side of grammar. European Journal of Social Psychology, 47(5), 566–579. https://doi.org/10.1002/ejsp.2231
Hartshorne, J. K., O’Donnell, T. J., Sudo, Y., Uruwashi, M., Lee, M., & Snedeker, J. (2016). Psych verbs, the Linking Problem, and the Acquisition of Language. Cognition, 157, 268–288. https://doi.org/10.1016/j.cognition.2016.08.008
Martín García, J. (1996). Aktionsart in Word Formation: Verbal Prefixation. Catalan Working Papers in Linguistics, ISSN 1132-256X, No 5, 1, 1996, Pags. 113-126.
Nook, E. C., Schleider, J. L., & Somerville, L. H. (2017). A linguistic signature of psychological distancing in emotion regulation. Journal of Experimental Psychology: General, 146(3), 337–346. https://doi.org/10.1037/xge0000263
Osswald, R., & Düsseldorf, H.-H.-U. (2021). Activities, Accomplishments and Causation.
Vendler, Z. (1967). Linguistics in Philosophy. Philosophy, 45(171), 71–72. https://doi.org/10.1017/S0031819100009657