Все ссылки в воду: Как Филолог Всея Руси цитирует научные работы

Analysis of a video on Cyrillic readability and scientific studies questioning claims about Cyrillic vs Latin fonts.

Key Takeaways

Cyrillic script is not inherently less readable than Latin; market and design factors play a bigger role.
Scientific studies cited to prove Cyrillic superiority have methodological and statistical issues.
Phonological ambiguity in Cyrillic can slow word recognition, but this was not properly controlled in the study.
The original video’s conclusions about Cyrillic readability are not fully supported by the cited research.
Better font design and more rigorous research are needed to fairly assess Cyrillic readability.

What the video covers

The video critiques a popular video by the 'Philologist of All Russia' channel about Cyrillic script and font design.
It discusses the historical development of Cyrillic and the market limitations affecting font quality.
Focus is on a cited scientific study testing word readability in Cyrillic and Latin alphabets using Serbian participants.
The study claims Cyrillic words are recognized faster, but the analysis reveals methodological flaws and weak effect sizes.
Phonological ambiguity in Cyrillic words may affect recognition speed, but the study does not provide the word list to verify this.
The study's statistical reporting is questioned, particularly the improbably high degrees of freedom reported.
The experimental design lacks ecological validity due to use of large fonts and limited word types.
The video argues the cited study does not support its own conclusions and that the original video's author did not critically assess the paper.
Overall, the problem with Cyrillic readability is attributed more to economic and market factors than inherent script qualities.
The video calls for more rigorous scientific scrutiny and better font design tailored specifically for Cyrillic.

Chapters

What is the main argument about Cyrillic readability in the video?

The video argues that Cyrillic is not inherently less readable than Latin; rather, market limitations and poor font design are the main issues.

What methodological problems were found in the cited scientific study?

The study had limited ecological validity, lacked transparency about the word list, and reported implausibly high degrees of freedom, undermining its conclusions.

Why is phonological ambiguity important in this context?

Phonological ambiguity can slow word recognition in Cyrillic, but the study did not clarify if this factor was controlled, casting doubt on its results.

Full Transcript — Download SRT & Markdown

Speaker A

The "Philologist of All Russia" channel released a video on what’s wrong with Cyrillic, and why Russified logos look like cheap market knock-offs. And the answer given is this: it’s not about the letters. Cyrillic grew from the Greek alphabet, not the Latin one. Its

Speaker A

development was broken several times by top-down reforms, and in the nineties, no decent fonts were made for it; they just drew Latin ones by hand and passed them off as new Cyrillic ones. All this is interesting, but that’s not

Speaker A

today’s topic. Today’s topic is about the 8 minutes toward the end of the video, where she moves from history to science. Most likely, the Philologist of All Russia is right.

Speaker A

Cyrillic is not uglier than Latin. The problem is money. The market is ten times smaller, fonts were made as an afterthought for decades, and we’ve simply gotten used to cheap work. I like this thesis. I’m ready to believe in it, and I won’t be arguing

Speaker A

with it today. The problem lies elsewhere. This conclusion is propped up by three scientific papers, and all three exist. All three are open-access.

Speaker A

I have read them. Let’s start with the first scientific paper. Here is what Julia, the channel’s author, says about it.

Speaker A

This experiment tested word readability and ways to improve it, such as making the text bold or increasing the spacing. Participants were shown a word. It could be real or made-up, in Latin or Cyrillic, and additionally in a

Speaker A

standard font, bold, or with increased spacing. The faster you press the button to say if it’s real or made-up, the faster you supposedly recognized the word, and therefore, the more readable the text is. The experiment involved Serbs, and the Serbian

Speaker A

language is written equally often in both Cyrillic and Latin. So, it’s all fair. And was Cyrillic recognized faster than Latin? What is this?

Speaker A

Cyrillic is more readable. The article is somewhat obscure, let’s say, although that, in itself, is not a problem, of course. Here, Julia paraphrases it quite closely. The authors do indeed write that Cyrillic words are processed faster. My complaint is not with the summary. My

Speaker A

complaint is, firstly, with the subtle leap from identifying a word on a screen by a fraction of a second faster to claiming that Cyrillic text is generally more readable, and secondly, with the article itself, which doesn't support its own conclusions. Let’s

Speaker A

start with the numbers directly from the study. The effect size of the alphabet in the first experiment is 1.6% of the explained variance, and in the second, it is 0.1%. 0.1%. And the fastest cell in the entire study is,

Speaker A

wait for it, Latin with expanded spacing at 772 milliseconds, which is faster than any Cyrillic variant. This is their Table number three. This is on their Figure four. It’s visible to the naked eye. Based on the internal logic of this article, the headline

Speaker A

should have read: "Spaced-out Latin, the fastest." But the truth is, you can’t say that either. As we are about to see, there is very little that can be said about this study, and for several reasons at once. First, the

Speaker A

most trivial one. The research methodology is extremely limited. Do you know how exactly the words were shown to the subjects? Roughly like this. This is 48-point font, a bit larger than standard printed text, isn't it? Moreover, only six-letter

Speaker A

nouns were considered. It is easy to guess that this is quite different from real reading, meaning the experimental design itself raises questions. In nerd-speak, this is called the problem of ecological validity. Unfortunately, this is not the only problem with the

Speaker A

cited study. I have reread it several times. And do you know what I didn't find in there? The list of the words used in the testing. There were 60 real words, plus an equal number of pseudowords. Hence, by the way, the 120

Speaker A

trials per subject. We will find this figure useful later. Keep it in mind. And this is very important. The fact is that phonological ambiguity affects word recognition speed. What is that?

Speaker A

Well, let's look at an example. Look at what is written here. Betap, Vetar. That is exactly what phonological ambiguity is. The word here has five letters. In the experiment, they were six-letter words, but the principle is the same. Now, look at the same Serbian

Speaker A

word written in the Latin alphabet. As you can see, the reading is unambiguous here. The letter V clearly tells us the word is in Latin, whereas in Cyrillic, all the letters are shared.

Speaker A

Phonologically ambiguous words take longer for the brain to recognize. Did the study authors take this into account? We don't know, because the list of those sixty words used in the experiment simply isn't there. We are left only to trust the authors and rely

Speaker A

on their competence. But do they deserve such trust? No, they do not. And to realize this, you don't even need to read the whole article. It is enough to read the introduction. I will show you four fragments of text. Notice

Speaker A

anything? Yeah. The quotes on the right. They are just a paraphrase. Of the quotes on the left. And if you think that only a couple of sentences are repeated... No, of course not. The authors lazily wrote half the

Speaker A

introduction twice, barely changing the wording. The editing was clearly top-notch, right? But that's only half the trouble. A study with poor editing can still be valuable to science. The main thing is that there shouldn't be problems with the methodology. Let's

Speaker A

just assume the authors didn't screw up the word list and accounted for phonological ambiguity. Is everything fine now? You believed that, did you?

Speaker A

Well, how should I put this politely? To figure this out, I had to perform a little reverse engineering. Let's take the first row from Table 2 in this study. They give us two values there.

Speaker A

These are the F and the partial eta squared. Sounds intimidating, but in reality, it's nothing special. F is a criterion that shows how many times the variance between groups is greater than the variance within groups. The higher the F, the more confident we can be

Speaker A

that the effect we investigated is not accidental. Partial eta squared is a measure of the effect size. It shows what proportion of data variance is accounted for by this factor, excluding all other factors and their influence.

Speaker A

In our case, it turns out that lexicality, that is, whether it's a real word or not, explains 7.3% of the variance in reaction time. So, using these two indicators, F and eta squared, we can find the df, that is, the

Speaker A

degrees of freedom. What are degrees of freedom? It is a numerical reflection of the number of independent observations. Yes, it sounds complicated, but it's not actually that difficult. DF is divided into DF effect, meaning how many groups we are

Speaker A

comparing minus one, and DF error, meaning how many of those independent data points we have left to separate a meaningful result from random noise.

Speaker A

The authors did not specify the error DF. That happens quite often, by the way. For such cases, there is this formula. If we run the data from the study through this formula, we'll find the error DF value and be blown away.

Speaker A

The error DF turns out to be 8,053. I repeated the calculation for each row of the table. And the figures there are of the same order. One might rightly wonder, what is so scary about the number 8,053? Look at their design. The

Speaker A

authors describe it in plain text. Alphabets and typefaces are between-subject factors. Each participant fell into exactly one of four groups and saw only their alphabet and only their typeface. And lexicality is a within-subject factor. Everyone saw the words and pseudowords. Those

Speaker A

are the very 120 trials. And for such a mixed design, DF is calculated simply.

Speaker A

For between-subject effects, it is the number of subjects minus the number of groups. For a within-subject effect, the formula is slightly different, but the number comes out the same, 71.

Speaker A

Remember, with any honest calculation, the error DF in this article is about 71. And now remember what error DF we calculated from their own table. 8,053.

Speaker A

Mor

Speaker A

120 is 9,000. We filter out the erroneous, bad trials, and about 8,053 remain. There is your answer. They counted the word as the unit of independent observation, not the person . The authors artificially inflated the sample size by passing off 120 attempts

Speaker A

by one person as the results of 120 different people. I don't know if it was done on purpose or not, but a study with such errors is absolutely invalid.

Speaker A

The authors committed one of the most banal errors: pseudoreplication. But could these just be typos in the table?

Speaker A

No, I repeated the exact same calculation for the second experiment. There are 68 subjects and 8,160 trials.

Speaker A

And the reconstructed degrees of freedom obediently approached 8,100 for the new sample. Typos aren't consistent across seven rows of two tables and don't scale with the number of trials.

Speaker A

And that’s not all. The authors provide confidence intervals for differences of plus or minus 22 milliseconds. With their own variance and 75 subjects, the interval would be seven times wider. Such precision is only possible with thousands of observations. Their own confidence

Speaker A

intervals reveal they calculated based on trials. Intersubjectivity hits the paper from another angle too. Let's analyze the example of the effect of the alphabet and letter spacing—that is, increasing the intervals between letters. Letter spacing and standard typeface were not tested on the same

Speaker A

people, but on different groups. About 17 people per group. With bold typeface in the first experiment. Same story.

Speaker A

There were groups of 19 people there. And people, actually, read at very different speeds. Think of yourself and that friend who finished reading in one evening. The authors got a non-significant result and wrote: " Letter spacing does not help Cyrillic."

Speaker A

So, I opened their own table three and sat down to calculate. They don't provide the standard error, but it, like the degrees of freedom, can be reconstructed. And I got a number after which this "letter spacing does not

Speaker A

help Cyrillic" claim. Looks, well, er, bold. And I won’t show it here. That’s 3 minutes of formulas on screen. Plus, people already wrote under my last video that it’s getting too stuffy. The entire calculation, the table, and this magic number itself are

Speaker A

on Boosty. Link in the description. That’s my arithmetic, not some secret data, the table itself, and the article . All of this is publicly available.

Speaker A

And you can verify it even without me. I will say this, the magic number is greater than 90. Greater than 90 of what? You’ll find out on Boosty. But don't worry, there are plenty of problems in the article for this video.

Speaker A

I’ll just remind you that the methodology was already strange. To test readability and reading speed, they showed subjects words, not text.

Speaker A

And in size 48 font, for crying out loud. That already sounds questionable. Except they don't even report the angular size of the words—that is, how many degrees they occupy in the human field of view—or at least the distance from the person to the screen.

Speaker A

It might seem like extra information, but in fact, the letter spacing tested in the experiment counteracts the crowding effect, which depends precisely on the angular size of the text. They don't tell us how much the text was spaced out. Maybe the authors

Speaker A

added half a pixel between letters, or maybe they tripled the interval. We don't know, which means their experiment is fundamentally irreproducible. We don't know how often the subjects made mistakes. The authors say they filtered out some data due to

Speaker A

certain extreme values. Which ones? We don't know. I already mentioned the phonological ambiguity at the beginning . Furthermore, the text contradicts its own tables. In the second experiment, the authors write that pseudowords are processed faster than words. Let’s

Speaker A

open their third table. Pseudowords are slower in every single cell. And it doesn’t match their first experiment either. No one proofread the results.

Speaker A

The p-values don't match the f-values. The table lists an f of 134 with a p of 0.70. With that f and those degrees of freedom, p would have had about 20 zeros after the decimal. These two numbers could not have come from the

Speaker A

same calculation. The study isn't just weak, it's invalid, and it should have raised red flags in the introduction alone. And the cherry on top. Even if this paper were flawless—which it clearly is not—it's about Serbs, people who read in both alphabets daily

Speaker A

from first grade. Applying this to a Russian reader, who has a completely different experience with two scripts, is a leap that no one justifies in the video. Although it could generally be justified, Yulia references this study regardless. And note, she summarized it

Speaker A

honestly. The problem lies elsewhere. She didn't check if the paper actually supports its own conclusions. If she read it, I have a question: why wasn't she bothered by everything described above? If she didn't read the study and only looked at the abstract, that in

Speaker A

itself raises questions. Well, maybe it's just one poorly chosen study. Well , things happen. After all, we are presented with another one later.

Speaker A

In this experiment, the author wanted to find out if the letters really look similar to a native speaker, and if there's the same problem in Latin or other writing systems?

Speaker A

Uh, no. First, the author wanted to test their own method for evaluating similarity relationships between symbols. This wasn't an alphabet test; it was a validation of a research method. It is a purely methodological paper.

Speaker A

They placed three letters side by side and asked to remove the odd one out. If everyone has the same results, it means the similarity is obvious and striking.

Speaker A

Well no, if everyone has the same results, it only means that the participants applied the same criterion . Here is a clear example for you. Most would say that the circle is the most different. Verified in my Telegram

Speaker A

channel. Subscribe. By the way, does that mean that a triangle and a square are generally similar? No, it's just that the criterion of whether there are angles is the most obvious one.

Speaker A

Moreover, the study's author understands this perfectly well and writes that there are many equally valid criteria for assessing similarity . And if the participants 'votes are divided, it only means that there are multiple competing criteria for evaluation, or, for example, that there

Speaker A

aren't enough of them. That is, none of the outcomes indicate to us that the similarity between the letters is obvious or vice versa. Moreover, based on the context, I can say that Yulia shouldn't be interested in the similarity of letters at all. If the

Speaker A

question is how confidently a letter is recognized during reading and whether it is confused with others, then this is not a question of similarity, but of distinctiveness, which the article calls perceivability. And the study specifically separates these concepts.

Speaker A

This is one of the fundamental methodological concepts of the article. Well, let's see what else Yulia has to say about this study.

Speaker A

It didn't work. What, excuse me, didn't work? After all , the method described in the article turned out to be more than successful.

Speaker A

Yes, those with weak language skills answered identically, while those who were fluent answered differently. For one, it's one extra letter, for another , it's a different one. This means that for those who know them, the letters are not identical, not similar. But

Speaker A

this thesis is in very bad shape. Well, first of all, this question is presented to us in a very polarized way . Some answer identically, others differently, but the numbers in the study itself show a more balanced picture. Let's take Cyrillic. To assess

Speaker A

consistency within the group, the study used Gwet's AC1 coefficient. For the fluent speakers it was 0.55, and for those with weak skills it was 0. And that, as you can see, is not a very contrasting picture. Even the study

Speaker A

itself already contains statistical analysis on this topic. We won't go deep into how the Mann-Whitney test works right now. Let's just look at a more intuitive sign: the sample size.

Speaker A

In the Cyrillic group, there were over 400 fluent speakers and 27 with weak skills. And in the Latin group, there were only two people with weak skills in total. Based on such a sample, it is impossible to draw these kinds of

Speaker A

conclusions. The chosen font had a strong influence. Well, here we have a classic case of journalistic interpretation. The influence of the font was indeed found in the article. And the method's sensitivity to font design is actually the main positive result of the work.

Speaker A

It’s just that the scale is very different. For the Calibri and Century Schoolbook pair, the font effect appeared in five out of fifty-six letter triplets. For the Futura and PT Sans pair, it appeared in twenty-two; the most significant cases are very

Speaker A

specific, like the Latin letter' g ', which is single-story in one font and double-story in another. And now to the main thesis of this excerpt.

Speaker A

And in the end, Cyrillic does not blur more than Latin. A wonderful statement, and we have already mostly sorted out why such conclusions cannot be drawn from this study. Letter blurring is a problem of distinctiveness. Abregezin consciously minimized the influence of

Speaker A

distinctiveness. Comparing Latin and Cyrillic based on this study is absurd in general, because they didn't test the entire alphabet, but eight lowercase letters. And yes, the Latin and Cyrillic were evaluated by different people with very different backgrounds. But here I do have a

Speaker A

couple of kind words. Though not for the philologist of all Russia, but for David Březina. He clearly articulated all the limitations of his work, and the idea of the task using triplets is quite good. Yes, the study is not

Speaker A

without flaws, but it is solid, decent work, even if somewhat overstated in its conclusions, but we are talking about the philologist of all Russia here. So let's listen to what Yulia says about the third work she references. And finally, the third

Speaker A

article concludes that the problem is not Cyrillic at all, but the fact that international fonts are designed through a Latin lens. The aesthetics of Cyrillic are judged by foreign criteria instead of being viewed as an independent system.

Speaker A

Whoa, that’s quite a bold claim. Let’s start with the genre. This is not a study, but a practical report at a Brazilian info-design conference.

Speaker A

Seven pages on how it’s easier for a designer with native Latin skills to draw Cyrillic. No experiment, no methodology. And that, in itself, is not a crime. It’s just that you can’t use such a text to support an

Speaker A

empirical thesis. But the problem goes deeper. The article does not criticize Latinocentrism. It is permeated by it.

Speaker A

It offers a practical recipe. Take Latin forms and reuse them for Cyrillic . It’s more productive that way. That is that very Latin lens, only with a positive spin. The only place where the author even touches on the subject of

Speaker A

who judges whom and by what standards is one paragraph where they contrast a Eurocentric view (well, that is, Peter’s Latinization) and an open view; and they side with market opportunity—meaning exactly at the point where the article could have said

Speaker A

what Yulia just attributed to it, it says the opposite. To me, it looks like they didn’t read the source beyond the abstract. In short, three links and two different problems. With the first article, Yulia honestly recounted the conclusions, but didn’t check that

Speaker A

there was simply nothing under those conclusions. With the second and third articles, she attributed things to the works that simply aren’t written in them. And the most obvious explanation for both problems is the same. They were reviewed based only on their

Speaker A

abstracts. I know many people do this, but that’s a very poor excuse. And this is where the most unpleasant part comes in. Remember where we started? I said that the philologist of all Russia is likely right. Cyrillic is not ugly

Speaker A

Latin. The problem is money and a foreign lens. I still think so. But after these three articles, I am forced to admit a simple thing. I didn't learn this from the video. I just took it on faith. The first work claims that

Speaker A

Cyrillic is recognized faster, but its statistics fall apart after a single recalculation. The second does not show that the letters don't blend together.

Speaker A

The third is written from exactly that Latin lens that we are encouraged to criticize using its own quote. The conclusion seems correct, but it lacks any of the foundations we were shown.

Speaker A

We are forced to trust educators because, as content consumers, we often simply don't have the time to check every link. And one would hope that educators would justify this trust. I really like the philologist of all Rus' and I really enjoy the content on this

Speaker A

channel. But this situation with the research, of course, raises some questions. I haven't checked the links in other videos and I really want to believe this one is an exception, but now I will check them. And honestly, I

Speaker A

advise you to do the same, including with me. Stay safe and goodbye to everyone. M.

Topics:CyrillicLatin alphabetfont designreadabilityscientific studyphonological ambiguitySerbian languagetypographystatistical analysislinguistics

Get More with the SozAI App

Transcribe recordings, audio files, and YouTube videos — with AI summaries and speaker detection. 30 minutes free.

App Store Google Play

Or transcribe another YouTube video here →

Free tools: TXT to SRT · SRT Validator · Merge SRT · Subtitle to Text · All tools