Blog

Here are some responses to the word cloud and corpus I released earlier…

John B. submits the above image, created from just a list of episode titles, using Tagxedo. (I used Wordle for mine.) He also points out that Tagxedo contains more customization filters and tools for creating word clouds, which is good to know. Thanks, John!

Rubrick correctly points out that “It should, I feel, be a crime to refer to ‘words that show up once and only once’ without using the awesome linguistic term for such words, hapax legomenon.”

Apparently there are ~6100 such words in the corpus, and they have helpfully been extracted by Jonathan B. here. How boring a writer I am to have only used ‘breakdancing’ once in nine years!

Elytsvil accurately notes that the corpus I released is missing a lot of punctuation, which Oh No Robot uses as meta-markers. I did streamline the textfile somewhat (eliminating URLs, strings of punctuation, and the ubiquitous ‘In which’) to try and force the word cloud into something approaching relevance, but the point is duly noted. Here is a completely unredacted ONR data export.

Finally, Shmibs extracted a list of the longest words in the corpus, and some of them are pretty great:

nervousenergynervousenergynervousenergy (from the alt-text on #016)

abcdefghijklmnopqrstuvwxyz — from #598

mmmellllltiiiinnnnngggggg — from #247

procrastihibernation — title of #614

radiogrammephonimat — a transcriber-added detail to #401

biepinzingerunting — from #715

biepbiepbieperzung — also from #715

telegrameutophium — a transcriber-added detail to #336

relationshaaaooww — from #588

yardeyardeyaryar — from the alt-text (which I wrote) on #199 (a guest comic)

superdorkasaurus — from the alt-text on #662

dunderschnauzen — from #515

glondxhatzoljlg — from #680

What an erudite collection of completely invented words and/or sounds!

On the occasion of Wondermark’s eight hundredth episode, I thought I would celebrate by looking at a complete corpus of words used in Wondermark, and creating a cloud from them (similar to my existing tag cloud of subject matter):

“Huh,” I thought to myself, “I suppose it is unsurprising that the most common words used in a large sample of comics probably closely resembles a list of common words found in the language in general.”

So no great discoveries here, unfortunately. It’s further complicated by the fact that the text I’m using as a corpus is an export of my Oh No Robot database, which contains user-submitted transcriptions of all my comics, which themselves often contain transcriber-invented character names and extensive scene descriptions — both of which are great, but which somewhat muddy the dataset. The heavy incidence of the words “man” and “woman” in the cloud, for example, are probably due to transcriptions reading something like:

Man: I have started a bean farm.
Woman: We’ll be millionaires!
Man: Not if flies eat the crops first.
Woman: Time to invest heavily in pesticides.

In that sample transcription, the words “man” and “woman” both appear twice as frequently as any other word, despite not occurring in the dialogue at all.

It’d be neat to see, instead of a brute word-frequency cloud, something like a collection of statistically improbable phrases, or words that show up in Wondermark once and only once…things like that. I wonder what interesting things could be mined from the data? If you’d like to play around with the corpus yourself, dirty as the set is, here’s the text file I used. If you derive anything neat, let us know!