Do scientists really have to talk like that?
It's wordy, but is it worth it?
Isn’t the scientific way of speaking just an aesthetic preference?
Answer: it’s not just pretty. Scientific inquiry is often criticized for being too “wordy” or inaccessible. But that’s like watching a game of chess and thinking they’re moving the pieces “excessively.” If you don’t see why they’re making those moves, it’s only because you don’t share that body of experiences.
Sure, there’s bound to be one example here or there where a more accessible word could be used. Desiderata comes to mind. We could just use “Design objectives” instead. But as far as I’ve seen, the information gain of increased precision vastly outweighs the rare cases where its value is less obvious.
But why? Why do I have to be precise?
Good question: Of course, I’m going to be quoting Feynman. "The first principle is that you must not fool yourself – and you are the easiest person to fool." If science is partially about not fooling yourself, then using words to clearly define your prediction helps you let go of a hypothesis after its useful lifetime has expired.
The very moment we have conclusive proof that our understanding didn’t work, we can move on to a different understanding. It has not died, no, its spirit lives on for the lifetime of our inquiry. Think of it like the soreness you feel after a good workout, it’s a sign you’re headed in the right direction. Being precise means you are honest with yourself about what the evidence actually means.
Good lacks specific criteria for evaluation. Instead of using this vague term, reach for "measurably effective" or "statistically significant" whenever available. For example, rather than writing "good results," write "The treatment showed a 45% improvement in patient outcomes."
High Quality suffers from undefined quality metrics. Replace this with specific standards. Instead of "high-quality data," write "Data with 99.9% accuracy rate and <0.1% missing values."
Correct implies absolute truth without criteria. Better alternatives include "consistent with evidence" or "matches established criteria." Rather than "correct methodology," write "Methodology following IEEE standards ABC section 7.3."
Ethical is problematic because moral frameworks vary. Instead, specify compliance with specific guidelines. Replace "ethical research" with "Research approved by IRB #12345 and following Helsinki Declaration principles."
Moral is subjective across cultures and contexts. Use "aligned with stated principles IJK" instead. Rather than "moral behavior," write "Behavior consistent with predetermined ethical framework X."
Right implies absolute correctness without justification. "Validated" or "verified" might suit your use-case better. "Right approach" becomes "Approach validated through peer review and three independent replications."
Should implies a universal obligation without justification. Instead, recommend based on specific criteria. "Researchers should" becomes "Based on a meta-analysis of 50 studies, we recommend researchers..."
Do you see the same pattern I see? These are words that are aesthetically regarded as “subjective,” and “unconditional.”
What this doesn’t mean
This doesn’t mean “never ever use these terms, ever” You absolutely may use them when
They are explicitly defined early in the work
The definition includes measurable criteria
The criteria are consistently applied throughout
The definition is relevant to the research context
For example
“For this study, we define ‘high quality’ sleep as meeting three criteria:
Sleep efficiency >85% as measured by polysomnography
<2 awakenings per night lasting >5 minutes
Subjective sleep quality rating >7 on the Pittsburgh Sleep Quality Index”
The issues arise when instead of saying “I prefer X,” you say “X is the best.” The issue arises when you prescribe that something ought to be a specific way. Think of it like some form of invented authority, leaving no room for anyone else to have another preference.
This also doesn’t mean that “there is no useful work to be done when defining good, high quality, or correct.” In fact, operationalizing these terms is immensely important, and in my view, one of the most important steps in research. This only means “to communicate effectively, it’s necessary to also show how the writer intends the reader to understand the terms.”
Where it’s less clear
You’ll have to exercise your own judgment for these ambiguous cases
massive, tiny, huge, and minimal all depend heavily on context — this is dependent on the established precedent in the research discipline (I’m looking at you, “Large Language Models”).
robust, reliable, stable, and resilient — Robust to what perturbations? Reliable in what sense? In different disciplines, these have different, but useful definitions, so depending on your research domain, this may be perfectly acceptable.
often, rarely, frequently, sometimes, and occasionally create unclear thresholds, but may serve as valuable qualitative descriptors if used consistently. If you’re using some convention, stating it will help your readers understand it.
normal, typical, standard, usual, and conventional — unless your research area already uses these words, consider replacing them with statistical descriptors such as “mode”
obvious, trivial, clear, evident, certainly, and undoubtedly — these are not only hard to quantify, but can also be fairly exclusionary and make you sound completely elitist
better, worse, improved, degraded, and enhanced — these prompt the next questions: compared to what? Under what conditions? For whom?
elegant, beautiful, clean — It’s probably fair for academics, especially mathematicians, to talk about subjective aesthetics. In fact, this is a strong motivator for a lot of people! There’s nothing wrong with these preferences per se, but saying that your aesthetic sensibilities are superior is where the elitism seeps in again, causing you to lose trust.
This is too many words, can I memorize some principles instead?
When something is highly multifactorial, or subjective, we keep the factors separate for as long as we can to maximize communication
We state our definitions when there’s a chance we may be understood
See also
Extra: Do Large Language Models Reason?
The experts disagree, but they are rarely using the same working definitions of the terms. Compare the blog post by https://openai.com/index/learning-to-reason-with-llms/ with the following post and podcast. These are all legitimate working definitions, and confusion arises when a conclusion is moved between contexts.
But the term reasoning has come to refer to such a diverse set of behaviors that we might have to replace it altogether with new words (and prevent the same from happening to the new words, too). This in a word, is a tragedy. It marks the end of the same word being used consistently for a while. That is useful because it keeps us on the same page. When definitions shift around, we have to do more work to communicate the same information.
Future Work
Procedurally extracting terms from text corpora likely to create unproductive or contentious debates
Find word embedding correlates of these “subjective terms” and compare if those words have a correlation to the “unproductive debates” found above
Second-order effects: what words do people replace it with if these words are banned?
Does misinformation, propaganda, and manipulative speech have an overrepresentation of these terms?