Papers in linguistics
This page collects my papers in two groups. The first is a short series of measurement papers on YouTube comment data: what such data can and cannot support, and what to check before relying on it. The second gathers small empirical observations of language online: cheers, names and renamings.
Each paper comes with its data, code and a record of corrections, so that it can be checked. Refutations, reviews and comments are most welcome.
YouTube comment data · measurement series
Three papers, one question: what can a YouTube comment corpus support, from the day it is collected to the years after it is deposited?
What the Corpus Does Not Say
Reply structure, time and coverage in YouTube comment data — what linguists and other researchers should check before treating comments as conversation
Where the delivered fields mislead about who answered whom and when, by how much, and ten checks that find it in minutes. Its premises are re-checked every six months. Version 1.7 (30 October 2026); premises last checked 24 October 2026, all unchanged; next check May 2027.
What the Corpus Can No Longer Find
Legacy YouTube comment IDs in a 2017 research deposit no longer resolve, although most of the comments are still there
None of the 703,339 comment IDs of a published 2017 deposit can be looked up today, although most of the comments appear to be still visible under IDs in a different format. Before relying on deposited IDs, test a sample one by one.
Research note · 2026 AbstractPDFZenodoVerification package
What the Corpus Still Holds
Testing keys, survival and result stability in deposited YouTube comment corpora
Three deposits from 2020 to 2024: the IDs still work, and 74% to 95% of the comments can still be retrieved. The loss comes in whole videos and threads, and the more hostile the corpus, the more of its hostile comments are lost. Yet every deposited distribution moves by less than one percentage point.
Preprint · 2026 AbstractPDFZenodoVerification package
Small empirical observations in linguistics
Small, descriptive studies that follow my own curiosity. I claim no academic standing for them; I hope the observations are valid and of some use. Refutations, reviews and comments are most welcome.
Official renamings and incidental mentions
Revision-level evidence from English Wikipedia
How do articles that mention a place only in passing change after an official renaming? Four renamings, followed revision by revision for 600 days. The new name tends to follow the title of the place's own article, and many insertions mention the name rather than use it.
Used or discussed? Lake Ontario and Lake America in YouTube comments
Separating name use from name talk in a frozen first snapshot
Is a newly introduced place name used to refer to the lake, or mainly talked about as a name? In the first weeks, ten of eleven comments with the new name discussed it; one used it.
Preprint · 2026 AbstractPDFZenodoReplication
Storm names as comparative benchmarks
Lay referential uptake of institutional weather names in German YouTube comments
What do ordinary speakers do with storm names years afterwards? Mostly they use them to measure: worse than Kyrill, as bad as Lothar. The 2021 Ahr valley flood was officially named Bernd, yet almost nobody uses that name. Being named and being catastrophic is not enough for a name to catch on.
Preprint · 2026 Zenodo
“Allez, Venga, Vamos”
From lexical hits to validated cheers and metapragmatic commentary in YouTube climbing comments
Do climbers and spectators borrow foreign cheers? Names, venues and ordinary uses produce many plausible-looking hits. Of 146 candidate comments, two blind coders retained 21 as cheers in a different-language comment; in most of them the cheer closes the comment, like a detachable formula rather than a borrowing.
Preprint · 2026 ZenodoData and coding
Timmy or Hope?
Competing names for a humpback whale in German YouTube comment threads
One whale, two names. In the first 24 hours, commenters mostly use the name their video supplies (74.2% Hope under Hope videos, 84.7% Timmy under Timmy videos). Timmy becomes dominant elsewhere, but the Hope threads do not switch: local naming persists side by side.
Preprint · 2026 SSRN
All papers, earlier work and superseded versions: Google Scholar · ORCID · SSRN. Older versions stay available on Zenodo, each with its version history.
Current work
One renaming, two platforms. YouTube comments and Wikipedia revision histories on the same event, the US federal renaming of Lake Ontario as Lake America, observed on the same day, 22 December 2026, and again in December 2027, under rules fixed in advance.
Reprimands in the Bundestag. How explicit reprimands in German plenary records are taken up, with a design registered in advance (OSF).
What the corpus still holds, again. A repeat measurement of the three deposits in May 2027, with the same scripts.
About
I am not affiliated with a university or any other institution. I graduated in English philology (Staatsexamen) from the University of Regensburg in 2008 and have kept an interest in linguistics ever since.
I do this research as an amateur, in the old sense of the word: out of interest. My interest is primarily linguistic. The topics are small and follow my own curiosity, mostly things that are happening now and can be studied while they happen. Some questions kept leading back to the data themselves, and so the measurement papers came about. I do not try to build new theories. What matters to me is that a study is empirical, transparent and checkable, so that others can test it and, if necessary, refute it.
I fix designs and decision rules before looking at the data where I can, and deposit data and code on Zenodo. My papers name their weaknesses openly and are revised with each version; the correction registers are part of the record.
Philosophy has remained a second interest. Essays that are not research are collected on the Philosophy page.
I have worked as carefully as I could. If I have made a mistake, or overlooked or misrepresented relevant work, I apologise and would be glad to be told.
Use of AI
I design the studies, collect the data, and set the analytic decisions and decision rules. AI does much of the execution: Claude (Anthropic) is used under my direction to write the analysis code, compute the figures and draft the text, which I then revise and edit. ChatGPT (OpenAI), Gemini (Google) and Consensus are used for literature searches and critical review. All numbers come from scripts run on the deposited data, so that anyone can check them. Each paper states the division of labour; I am responsible for every claim.
Contact
Max Sedlmair · research@woidwide.net · ORCID 0009-0007-2471-5595 · Google Scholar