Trying to analyze Donald Trump literally made a CRASH speech recognition bot – and the only way to get it to work properly was to remove his grammar and syntax tools
- Researchers awarded speech recognition bot to analyze Donald Trump
- The bot struggled to make sense of its disjointed and fragmentary sentences
- Team hired specialist to remove bot grammar and syntax tools
- Since unlearning grammar, the AI bot has become an expert on Trump and is licensed by Amazon to help Alexa answer users’ questions about the president
The developers of a speech recognition bot to analyze public statements by politicians encountered a major stumbling block when he tried to make sense of Donald Trump.
Built by a tech startup called FactSquared, the AI bot has been responsible for browsing more than 11 million words that Trump has spoken or tweeted since 1976 – in interviews, campaign speeches, media appearances, and publications on social media.
According to FactSquared CEO Bill Frischling, the bot didn’t understand Trump’s speeches until he called in a specialist to remove all the grammatical and syntactic coding from the bot.
Tech startup FactSquared created an AI bot to try to catalog and analyze public appearances and interviews with Donald Trump, but they were so inconsistent and disjointed that the bot actually crashed. It was only after removing the grammar tools that the bot was able to understand Trump
“ He always tried to punctuate him as if he was English, as opposed to trying to punctuate him as if he was Trump, ” said Frischling The LA Times.
A speech had in fact caused the crash of the bot, named Margaret after a character from The West Wing.
Trump delivered the speech on May 4, 2017 to commemorate the 75th anniversary of the Battle of the Coral Sea, which pitted American and Australian troops against the Imperial Japanese Navy.
“ We are privileged to be joined by many incredible veterans from our two countries, and for really, from so many different conflicts, ” said Trump.
“There are so many different conflicts that we have fought and worked on together. And besides, in any case, succeeded. It’s nice to win.
Programmer Bill Frischling founded FactSquared to create analyze and catalog public speeches and apparitions of politicians, which can then be used by journalists, researchers and even companies like Amazon
Without the obstacle of proper grammar and clear syntax, Trump suddenly began to make a lot more sense for Margaret, and today the bot has become an expert on the president.
FactSquared makes Margaret available to journalists and historians as a resource, and it also licenses the technology to Amazon, which uses it to help Alexa answer users’ questions about Trump more precisely.
Margaret not only lists transcripts of speeches, but can also verify Trump’s claims by independently performing requests for documents from public records.
Margaret has also been coded with machine learning tools that allow her to identify the different emotional and psychological states of Trump when he speaks.
According to Margaret, Trump speaks much faster when improvising than when reading a script – about 220 words per minute versus 111 words per minute.
According to Margaret, Trump shows few signs of physical stress when lying, while for most others, lying is accompanied by signs of physical tension.
Margaret discovered that the surest sign that Trump was really angry was when he stopped moving his hands, which he also uses frequently when he talks.
“When he stops gesturing, it’s the – attention,” said Frischling. “Whatever happens, hang in there.”
The overall vision that Margaret developed by studying Trump’s public statements is that there is ultimately very little connection between what he says and does.
“He can say” I think it’s the worst “about a policy proposal, and then literally sign an executive decree 10 hours later that approves it,” said Frischling.