(Image: Paul McDevitt)
Feedback is our weekly column of bizarre stories, implausible advertising claims, confusing instructions and more
The robot reader鈥檚 digest
WORK in the 快猫短视频 office was recently hindered by news from Google Labs. 鈥淧rogress on the path from shallow bag-of-words information retrieval algorithms to machines capable of reading and understanding documents has been slow,鈥� .
Advertisement
This is in part because training a natural-language processing system requires an extensive database of well-annotated writing. Now the researchers have chanced upon such a corpus鈥� in the form of the Daily Mail, which prefaces each online story with a bulletpointed list of its salient details, presumably for the benefit of its human, rather than robotic, readers.
Having digested the literature, the researcher鈥檚 system should be able to work out 鈥淴鈥�, the common thread that links a hi-tech bra, Saccharin and fish oils. However, this is hardly taxing. 鈥淎n ngram language model trained on the Daily Mail would easily correctly predict that (X = cancer), regardless of the contents of the context document,鈥� the researchers admit, 鈥渟imply because this is a very frequently cured entity in the Daily Mail 肠辞谤辫耻蝉.鈥�
Monash University asked Kevin Lee: 鈥淒id you know that Australians use more than 3.5 million tonnes of paper each year? That鈥檚 3,500,000,000 kilograms of chocolate!鈥�
Bittersweet findings
IN OTHER words, if in doubt, guess 鈥渃ancer鈥�. Feedback notes the sad lapsing of the , documenting the paper鈥檚 鈥渙ngoing mission to divide all the inanimate objects in the world into those that cause or cure cancer鈥� (18 September 2010).
The similar, and similarly sleeping, makes no mention of Google鈥檚 reading algorithms, but draws on the Mail corpus to assert that computers in general 鈥渂oth cause and prevent cancer鈥�, a finding which ought to throw any paranoid androids reading the newspaper into existential crisis.
Sowing a truffle kerfuffle
OF COURSE, it鈥檚 not all bad news in the Daily Mail and its fellow British tabloids; they are always keen to bring word of household miracles, too. In March, newspapers worldwide were awash with the glad news that, as the UK Daily Express put it, 鈥溾��. Could this be genuine, and not just another entry for the of 鈥渕iracle cures鈥� promulgated by that august organ (10 May 2014)?
Alas, no. Journalist John Bohannon crowed that he had set the whole thing up; a shallow trial and a 鈧�600 submission fee to was all he had needed to secure the headlines.
Neapolitan dreams
He was planting seeds in fertile ground: claimed in April 2012 that 鈥渁 study by the University of California has found regular chocolate scoffers have less body fat because while it contains more calories than many other foods, it appears to make your metabolism work harder鈥�. And on and on with the wishful thinking.
I should cocoa
BANG on cue, while Feedback was mulling these lines, the Daily Express came up trumps with another front page screamer: 鈥�.鈥�
The rocky road of success
OF COURSE there is also the correlation between nations鈥� chocolate consumption and their Nobel Prize haul (3 November 2012). This has been followed up with 鈥� and, yes, they ate more chocolate than the control group. (Of whom is the control group for Nobelists composed?) We look forward to a further follow-up controlling for subsequent sanity.
Once more with ganache
PURSUING chocolate further, we find a reputable-looking 2012 paper titled 鈥淎ssociation between more frequent chocolate consumption and lower body mass index鈥� (). Our wish to believe is now turned up to 11, as we reach for another square of Belgium鈥檚 finest.
A passing familiarity
IN OUR teens, Feedback puzzled over experiments demonstrating a 鈥渃asual relationship鈥� between this and that. Surely, the whole point of an experiment is to elicit a bit more鈥� commitment?
Our concern persists. Our favourite correction in a long time was published in the British Medical Journal at the end of last year. It regrets that a paper 鈥渃ontains the same typo in figures 2 and 3. In both figures, the middle title should have read: 鈥楥ausal claims from correlation鈥� [not casual as was published].鈥�
We suspect a connection between the error and the paper鈥檚 subject (): 鈥淭he association between exaggeration in health related science news and academic press releases鈥�.
Artisanal light fixtures
FINALLY, energy-saving light bulbs evoke strong feelings in some 鈥� even accusations that they may send you blind or be part of some conspiracy by evil, tree-hugging bureaucrats (19 July 2014). Now, pseudo-traditional bulbs are being produced. Andy Johnson-Laird sends in an advert for an 鈥溾��. Our attention is grabbed by the materials listed for this: 鈥渨ool, leather, carbon鈥�.