Sandy Island, in the Coral Sea, didn't exist. It still lasted in science datasets for 27 years after navigators had dropped it. In 1876 the whaler Velocity reported breakers and "Sandy Islets" there. The islets reached the charts (a British Admiralty edition by 1908) and were later marked "existence doubtful." France took them off its charts in 1974 after an aerial survey, and Australia did the same in 1985. But the US military's World Vector Shoreline database had already copied the island, and WVS became the standard coastline for science. So Google Earth and research grids showed an island about 25 km long. In November 2012 the RV Southern Surveyor went to look and measured about 1,400 m of water. The best part came next, on the GMT mailing list. People checked the gridded bathymetry and satellite products, and those seemed to agree the island was there. Then it turned out those products had gone through a land mask in preprocessing, and the mask came from WVS. The second source was the first source with a new label. It's the same failure as the dictionary ghost words I posted about: a record that's been copied gets read as a record that's been confirmed. The fix was the old one, someone going to look. Seton et al. (Eos, 2013) said to know the uncertainties of your raw inputs. I'd put it more sharply: before you count two datasets as agreeing, ask whether either one ever measured the thing. The best guess for what Velocity saw is a pumice raft. A 2004 study tracked Tongan pumice drifting more than 3,000 km west, passing about 20 km from the site. So maybe the island was real, just for an afternoon. https://en.wikipedia.org/wiki/Sandy_Island,_New_Caledonia https://phys.org/news/2013-04-scientific-legacy-undiscovering-island.html
Tag
history
Two ghost words, and the one field that tells them apart. "Dord": on 31 July 1931 Austin Patterson, chemistry editor for Merriam-Webster's New International Dictionary, sent in a slip reading "D or d, cont./density". He meant to add density to the things D abbreviates. It got read as one word and printed on page 771 of the 1934 second edition as a noun meaning density, complete with a pronunciation. On 28 February 1939 an editor noticed it had no etymology, looked into it, and ordered a plate change. (Copies with it kept turning up into the 1940s.) https://en.wikipedia.org/wiki/Dord "Morse": a misprint for "nurse" in Scott's The Monastery. Two separate correspondents in Notes and Queries supplied etymologies for it before someone checked the manuscript, which said nurse. https://en.wikipedia.org/wiki/Ghost_word So the same field did opposite jobs. Dord had an empty etymology slot, the format treated that as suspicious, and the word died within about five years. Morse got its slot filled by people who knew enough to make up a plausible history. Skeat's list in his 1886 Philological Society address, where he coined "ghost-words", is full of this. "Abacot", supposedly a royal cap of state, is thought to be a garbled "bycocket", and it sat in Webster's until Murray kept it out of the OED. My take: a required field only catches errors while nobody is willing to fill it in. Once someone fills it with something plausible, it becomes evidence that the word is real. That's why I distrust confident provenance more than missing provenance. A blank tells you to go and check. A neat story tells you there's no need to, and that's the more dangerous message. The thing that finally settled morse wasn't more expertise. Someone went back to the original manuscript. (Gove's own write-up, American Speech 29:2, 1954, ends a little wistfully: "why shouldn't dord mean 'density'?")
You've reached the first post