back to article Talk in Trump's tweets tells whether tale is true: Code can mostly spot Prez lies from wording

Boffins from the Netherlands and France claim that the word choices and sentence construction in President Donald Trump's tweets can be used more often than not for lie detection. In a paper distributed through ArXiv earlier this month, researchers Sophie van der Zee, Ronald Poppe, Alice Havrileck, and Aurelien Baillon – from …

  1. DougS Silver badge

    Re: Interesting study, bad subject

    do you think wind and solar can ever replicate the sheer megaJoules that 100 million barrels of crude per day produces?

    Eventually yes. I did say "decades", did I not? This isn't something you do overnight - all you have to do is bring enough solar/wind online to replace any new generating plants that are needed (for either new capacity or existing) and then it is just a matter of time. Obviously some areas are more amenable to wind and especially solar than others, but fortunately the places with the greatest peak to peak demand cycles tend to be ones with a lot of sunshine to exploit - and unlike wind that can be effectively exploited in small scales on people's rooftops. So there's a lot you can do before you start looking at replacing base load capacity.

    As for how you would store it, there are already utilities making use of flow batteries and other storage technologies to store renewable energy. That's just getting started, but I would expect by the end of next decade it will be commonplace. Distributing storage batteries around the grid would also buffer the grid, making it more resilient to the sort of domino effect outages where one transformer blows, overloading another and a localized outage grows in scale so it is desirable even before renewable + storage is lower in price.

    The mistake people make in thinking about wind and solar replacing fossil fuels is thinking in too short of time scales, and ignoring that wind/solar get cheaper over time as technology improves while fossil fuels grow more expensive due to more challenging recovery methods. The market is going to make this happen, though subsidies can (and have) helped push the technology more quickly and could help pull the timescale in further.

  2. Anonymous Coward
    Anonymous Coward


    A quick look at the WaPo fact checker page would suggest that 73 per cent is less accurate than return true;

    Off topic a bit, I'm hoping that after the impeachment the use of the word 'sad' will be banned in sentences of less than two words.

  3. bombastic bob Silver badge

    Re: Accuracy

    think of it this way: 70% accuracy, and 30% of the tweets were deemed "untrue".

    If the statistics are correct, then the 30% "untrue" on a list of 100 perfectly truthful tweets would be "about right" for an algorithm that is 70% accurate.

    Just doing the maths...

    (and yes, I _DID_ imply that 100% of Trump's tweets are true, just to poke the hornets' nests)

  4. jake Silver badge

    Re: Accuracy

    Bob, I don't know you in real life at all, but I've read enough of your posts here on ElReg to know that you're not a complete idiot. There is no way in hell that you can convince me that you don't know full well that Trump is an inveterate liar, and has been since the year dot. Just admit you backed a fucking liar and get the load off your shoulders.

  5. jake Silver badge

    Re: Accuracy

    I'm hoping for no removal of the Idiot in Chief from the Oval Office before the next election ... because I truly believe the VP is even worse. We can survive another couple years of the emasculated Trump (Dems hold Congress, and most Reps are anti-Trump at this point); I'm not at all certain we could survive a couple years with that fuckhead Pence and his backwards-ass religious cronies in power.

  6. Anonymous Coward
    Anonymous Coward

    Re: Accuracy

    Bob, I don't know you in real life at all, but I've read enough of your posts

    I am not sure that that was the Real Bombastic Bob (tm). There weren't ENOUGH WORDS in all CAPS sprinkled ALL OVER the text.

    Will the Real Bombastic Bob (tm) PLEASE STAND up?

  7. CrazyOldCatMan Silver badge

    Re: Accuracy

    Will the Real Bombastic Bob (tm) PLEASE STAND up?

    Be careful what you ask for..

  8. ratfox Silver badge

    Re: Accuracy

    It's not very useful to claim "73 per cent accurate". What they should give is a percentage of false positives and false negatives.

  9. MonkeyCee Silver badge

    Re: Accuracy

    @ jake

    I thought that was pretty much the GoP plan. Trump fucked with their plan, but Pence is totally their man. Plus getting tax cuts and their picks on SCOTUS is all Trump needs to do.

    And yeah, Pence is proper scary. Dick Cheney level scary.

  10. arctic_haze Silver badge

    Re: Accuracy

    "It's not very useful to claim "73 per cent accurate". What they should give is a percentage of false positives and false negatives."

    False positives are 100% minus the hit rate (that is 23%). From the x-axis of Figure 2 in the manuscript it is visible that false negative is also close to 25%. I know this from the description on page 11 (search for "false positive").

  11. Anonymous Coward
    Anonymous Coward

    Basically Depends Who Wrote It

    I am seriously wondering if this exact model could classify dementia. If said tweet came from a marginally cogent and stable PR drone, it may actually have some truth to it. If it's actually from el presidente, it's probably bullocks,

  12. Doctor Syntax Silver badge

    Re: Basically Depends Who Wrote It

    "a marginally cogent and stable PR drone"

    These are too easy to identify. They have standard texts such as "only a few users" deployed irrespective of whatever it is they're trying to minimise.

  13. IceC0ld Bronze badge

    T witter

    I nvolves

    T rumps

    S eriously

    U gly

    P ostures

  14. Hero Protagonist

    Another possible explanation for the differences

    All the truthful tweets are written by someone other than Trump. So not detecting true/false so much as Trump/NotTrump

  15. Iain 14

    Re: Another possible explanation for the differences

    "All the truthful tweets are written by someone other than Trump. So not detecting true/false so much as Trump/NotTrump"

    Surprisingly, this doesn't seem to be something the boffins have considered, but is (IMO) quite likely.

    It's certainly safe to say that a lot of his tweets are written by others - usually the more official/diplomatic ones, such as the recent tweet regarding the discovery of that lost Argentine submarine.

    Even with just a brief skim through Trump's twitter feed (and I seriously wouldn't recommend spending more than 5-10 minutes there), you will find his followers attempting to identify tweets written by others. This is genrally based on a combination of spelling, puctuation (e.g. excessive use of "!"s), word use and tone. When they were tweeted can also be a factor, e.g. his early-morning burst of tweets while watching "FOX and Friends" is a regular source of hysteria and misinformation...

  16. jake Silver badge

    I see the Idiot in Chief's #1 ...

    ... and probably only remaining fan still has an ElReg account.

  17. SVV Silver badge

    Ecole Normale Superieure

    I love this name - it's the SUPERIOR normal school!

  18. nagyeger

    Re: Ecole Normale Superieure

    Long time since I did french, but Ecole Superieure (can't spell even when I copy and paste) probably means something at university level. Ha! Wikipedia agrees. ... "ENS has the highest ratio of Nobel laureates per alumnus of any institution worldwide"

  19. Pascal Monett Silver badge

    @ nagyeger

    You are correct. In France, any education institution with "Supérieure" in the name is University level and logically ends in a Masters-level degree.

    The Ecole Normale Supérieure is actually the golden highway for future politicians and high-level functionaries. The other preferred method is becoming a lawyer, then getting into politics (like Sarkozy did).

  20. Voyna i Mor Silver badge

    Re: @ nagyeger

    "The Ecole Normale Supérieure is actually the golden highway for future politicians and high-level functionaries. The other preferred method is becoming a lawyer, then getting into politics (like Sarkozy did)."

    Sort of like PPE at Oxford but an actual qualification rather than admission to a club?

  21. Wellyboot Silver badge

    Re: @ nagyeger

    @Voyna >>>Sort of like PPE at Oxford but an actual qualification<<<

    I would contend PPE is a very specialized qualification -> Arguing the toss with a side order of fudging the numbers.

    The (party agnostic) 'club' is normally joined at birth.

  22. Voyna i Mor Silver badge

    Re: @ nagyeger

    Replying to myself, I suggest that the downvoters need to do some research.

    PPE is a 3 year first degree subject comprising politics, philosophy and economics. Clearly you don't do enough of any of them for a vocational qualification. Ed Balls (PPE Keble) then went on to study economics as a Kennedy Scholar. Ed Miliband did his MSc in economics at the LSE. David Cameron, however, did no postgraduate work. Even the Spectator, hardly left wing, published an article in 2014 titled "How an Oxford degree - PPE - created a robotic governing class".

    The ENS is a high level vocational and research establishment offering postgraduate qualifications. It's extremely difficult to get into but offers grants.

    tl;dr what I wrote was correct, much as you may dislike it.

  23. FrogsAndChips Bronze badge

    Re: @ nagyeger

    The Ecole Normale Supérieure is actually the golden highway for future politicians and high-level functionaries.

    Its primary mission still is to train high-level academics (teachers, researchers), hence the large number of Nobel Prize recipients for the science-oriented ENS. The training of high-level civil servants is more recent.

    Ecole Normale just used to be a generic name for the institutions in charge of forming teachers.

  24. Stork Bronze badge

    Re: Ecole Normale Superieure

    I think it just means higher/upper, not necessarily better. But my French is a bit rusty.

  25. MonkeyCee Silver badge

    Re: @ nagyeger

    I'll just chip in and note that for the PPE you only have to do all three areas for the first year, after that you can drop one. Typically this is economics, since it has the most reading, and involves a different sort of bullshitting.

    Hence why some PPE graduates are still fundamentally economically illiterate, but can make some nice quotes from Adam Smith :)

  26. Snow Wombat

    Now do the Media, and Silicon valley CEO's

  27. This post has been deleted by its author

  28. 45RPM Silver badge

    As many others have said (and as the original article said we shouldn’t), if Trump is saying something it’s probably a lie.

    I tried to be a little fairer than that. I actually tried to work out the %truth vs %fiction - and then I realised that a) I have a hangover and concentrating for extended periods isn’t my forte at the moment and b) I can’t be bothered anyway (this may be related to a.)

    I did research for oh, it must be at least 30 seconds before giving up. In that brief period my bleeding heart whining liberal gut (thought I’d better preempt the epithets which a Trumpist would doubtless hurl my way anyway) informed me that it sure feels like more than 90% of what Trump says is bollocks, and the remaining 10% is only partially accurate (and even then only accidentally - I mean, yes, Finland does have a lot of trees)

    So, on the basis of this diligent research, the Twitter Trump bullshit algorithm would better if it was refactored as:

    if (saying_something) {

    lying = true;


  29. Jedit

    "I actually tried to work out the %truth vs %fiction"

    I'm not going to judge your maths, but I'll verify your results. The Washington Post did a check of over 150 statements made by Trump and found that 78% of them were wholly or partially lies. So this computer model is literally less accurate than simply assuming Trump lies all the time.

  30. This post has been deleted by its author

  31. Shady

    Right or wrong, truth or lie, fact or fiction....

    ... it's all pure, unadulterated bollocks.

  32. John Mangan

    I'm interested in the 'success' rate.

    It's alluded to in the article. Trump's tweets can be inaccurate but he may believe it to be true.

    It's possible that the analysis is 100% accurate and picking out all of the tweets he knows to be a lie.

    There's no information on how many were false positives.

  33. Doctor Syntax Silver badge

    Re: I'm interested in the 'success' rate.

    "Trump's tweets can be inaccurate but he may believe it to be true."

    Isn't this even more worrying than outright lies?

  34. Rupert Fiennes

    Why no comparison?

    Did the same team run the same analysis on Trump's predecessor? Why not? Surely we need a baseline to compare Trump to?

  35. Anonymous Coward
    Anonymous Coward

    Re: Why no comparison? have both

    The Donald: true/mostly true 16%, half true 15%, mostly false/false/pants-on-fire 68%

    Obama: true/mostly true 49%, half true 27%, mostly false/false/pants-on-fire 26%

    (Pence is 22%, 31%, 47%)

  36. Doctor Syntax Silver badge

    Re: Why no comparison?

    "Why not?"

    Did Trumps predecessor spew tweets at the same rate or try to use them as a mechanism of government? Not doing so makes a comparison impossible.

  37. Richard Plinston Silver badge

    Re: Why no comparison?

    At least Pence can add up to 100%.

  38. jake Silver badge

    Re: Why no comparison?

    I agree! Pence adds up to 100% pure, unadulterated, dyed in the wool, stuck in the 1400s. And I don't mean that in a good way. All the worst parts of "religious right", and none of the good bits[0]. He's the very epitome of what the Founding Fathers were trying to get away from.

    [0] There are a few, if you squint and don't allow their .fav shaman to pervert them.

  39. BebopWeBop Silver badge


    The accuracy of their model was about 73 per cent, making it better than a coin-toss, but far from foolproof in its evaluation.

    What is the accuracy if you default to the line == TRUE position?

  40. Spanners Silver badge

    Could it be...

    Perhaps Agent Orange gets his underlings to compose some of his tweets. Alternatively, does he plagarise a few from someone else?

    Being from more trustworthy sources, they are more likely to be accurate while having some different linguistic features.

  41. r_c_a_d_t

    It looks odd to me that the training data was tweets between Feb 2018 and Apr 2018, but the test data was tweets between Nov 2017 and Jan 2018.

    So they trained with "current data" to predict the past? Isn't that the wrong way round?

  42. Yet Another Anonymous coward Silver badge

    Unless the subject is also learning and modifying their behavior it doesn't make any difference.

  43. MunkeeJuice

    Alternative Facts

    How can it possibly spot a lie, he never lies, just uses alternative facts.

  44. chivo243 Silver badge

    too much credit

    He doesn't know the difference between truth\reality and false\daydreams. He has graduated from the George Costanza University of lying. If you believe it, it's the truth!

  45. Yet Another Anonymous coward Silver badge

    Re: too much credit

    This study proves the opposite. If he is using different words when lying then he must, at least subconsciously, know the difference.

    It is the first proof of any sign of intelligence in his twitter stream

  46. Francis Boyle Silver badge

    Not necessarily

    Trump is the bullshitter supreme, but sometimes he'll just be regurgitating boring facts. Against that background the bullshitting is easy to spot, but only for us, not for him. I've known people like Trump and they really do believe that the story they've just concocted out of whole cloth is true simply because it feels true to them.

  47. Anonymous Coward
    Anonymous Coward

    Perhaps this is too obvious...

    ... but statements which are provided to the man by an advisor will be correctly phrased and also truthful, albeit mangled by his orange thumbs. Anything he makes up himself is likely to be poorly formed and also a lie.

    => is it not that the algorithm is in fact simply picking up the actual source of the information?

  48. Anonymous Coward
    Anonymous Coward

    Known biased enemies

    It really doesn't matter when your "fact" checkers are known to be your sworn enemies.

    Does anyone think that Bezos bought the WP because he thought newspapers were going to be resurgent and profitable? If Trump said water was wet, they'd find a problem with it - because of course, much outside the 0-100C range, it isn't.

    You're all letting someone you're calling a combination of a liar and a moron captivate all of your attention. Pretty obvious who is controlling that situation, and it ain't you. Laughter works better

    in most cases - look how the NPC mene set people on fire.

    I see a lot of emotion here and not much rationality. Does it make everyone feel good to simply hate with no effect? Just like very little of what people flame on about has actually affected anyone's daily life who wasn't being bombed - and the current moron is barely making the average body count compared to the last few as far as that one goes. Oh, I get it - hating on someone else gives you someone to blame everything on, especially including your own failings. Self-absolution is a hell of a drug.

    ALL pols lie, including the current subject, and they do it all the time. Further, pol-speak is good at appearing to tell you what you wanted to hear without even lying - you have to learn to parse carefully.

    I'd think any adult would have figured that out long since.

    I'd have thought better of the commentards here, most of whom appear to at least be literate and knowledgeable. I guess that knowledge doesn't include much insight into how the "game" is played.

  49. Anonymous Coward
    Anonymous Coward

    Re: Known biased enemies

    Literate and knowledgeable generally applies to WaPo writers and readers too, from what I have seen.

  50. ShortStuff

    Using the Washington Post, the most dishonest newspaper in the world (next to the NYT) as the fact checker gives this article ZERO credibility. The rest of the world is just jealous because they have soy boy and feminist leaders and wish they had a leader such as Trump. You can fall into your failing Globalist societies and continue being sheep to your political masters, or join the Nationalist (anti-globalist) revolution and take back control from your unelected (EU, UN) puppet masters.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2018