The Queen’s English: What IBM Watson learned about the Queen from all of her Christmas speeches

And brands, similarly, are all too aware of this, nowhere more so than on Twitter where a millennial audience, with its leverage of influence and unique ways of expressing itself, is the holy grail for engagement, and where every keystroke risks them slipping into condescension or outright desperation.

But what if they could know beforehand exactly how the end user would perceive their messages? What if they knew poor word choices were unwittingly conveying anger or fear or sadness or disgust?

Well that’s exactly what IBM Watson’s Tone Analyser aims to do, helping users understand their own emotional, social and language tones, breaking communications down into individual elements and rating them to provide feedback on how brands can improve the effectiveness of their messages and how they are received.

To find out just how effective this tool is we decided to give it a spin, and what better brand to test it on than the Queen, whose every public utterance is meticulously scrutinised. So, to mark her 90th birthday, we have compiled every Christmas address she has ever made, from her accession in 1952 through to 2015, which Watson has dutifully analysed.

So how has her language evolved, what emotions does she convey, and what does word choice tell us about one of the most private of public figures?

Perhaps unsurprisingly, given the Great British stiff upper lip and penchant for a show of strength, overall emotional tone throughout her reign is found to be fairly low – no doubt intentional, considering the heavily scripted nature of her addresses. But it is by no means absent.

In her very first Christmas speech, given at the age of 27 as she was handed power following the death of her father, Watson can identify traces of fear and sadness. But that’s all it is, a trace. Sadness rarely registers in the 35,000+ words analysed, and certainly not to any reliable level.

Indeed, a reading of 50 per cent or more is required to confidently indicate a particular emotion is present, while 75 per cent makes it highly likely.

Surprisingly, then, one of the most ‘reliable’ emotional readings we get from her is one of disgust. And the subject is equally surprising.

In 1966, disgust averages 76 per cent as she directs her address “especially to women”, and jumps as high as 95 per cent when she relays how “in many countries, custom has decreed that women should play a minor part in public affairs”.

The Queen’s feminist credentials may be questionable, but despite her incredible wealth and inherited position, a woman at the upper echelons of society is still a rare sight. And, lest we forget, in 2013 she oversaw changes to the rules of succession that would pave the way for Princess Charlotte to be parachuted into fourth in line to the throne, displacing Prince Harry. Her disgust at the treatment of women, even in an era where it largely went unchecked, perhaps isn’t as unbelievable as it first sounds.

The reason ‘reliable’ has been afforded inverted commas is because a similar reading is found in 1981, when she talks about how “governments now regard it as their duty to try to protect their people, through social services, from the worst effects of illness, bereavement, joblessness and disability”.

This was the year that “three and a half thousand disabled people, with their families, came to tea” at a Buckingham Palace garden party and Watson again shows a spike of 95 per cent for disgust. Understandably, we were thrown as to why. Was the word ‘joblessness’ the cause? Or worse, ‘disabled people’?

Isolating individual terms and running them again through Watson’s Tone Analyser, we found it was indeed ‘disabled people’ causing the spike, which in turn caused us to scratch our heads. Is the Queen really disgusted by disability?

Unconvinced with the reliability of this reading, we tested her speech once more, but replacing the words ‘disabled people’ for ‘people with disability’. This caused the score to drop dramatically.

What happened is that Watson has learned that referencing ‘disabled people’ is often associated with a feeling of disgust – disgust, it should be pointed out, at the struggles this section of society endures rather than at this section of society itself.

The discrepancy here comes from the fact that the Watson Tone Analyser uses a machine learning model, and that these models are never perfect and reflect the training they receive – much like we humans who are influenced by the things we see and hear, and especially by those who teach us. The public meltdown of Microsoft’s AI chatbot Tay earlier this year, whose learning was hijacked by online trolls and troublemakers, shows just how susceptible machine learning can be.

Disgust isn’t the only emotion to rank highly in the Queen’s speeches, however. While few of us would consider her an overly angry person, despite the unfortunate facial expressions she is often caught sporting on camera, in the 60s and 70s it is the most prevalent emotion of the decade, reaching a peak in the 80s where it notches up an average probability of 49 per cent.

In her 1984 address in particular she is found to display most anger, although it is likely just coincidence that this is the year Harry is born. Instead, it is her reminiscing on the 40th anniversary of D-Day that is behind these increased readings as she calls on her subjects to “work to heal old wounds and to abandon prejudice and suspicion”.

The 90s also throw up some high levels of anger before she appears to mellow into old age. It would of course be pure speculation to suggest that her strained relationship with Diana had something to do with this, as she actually returns a sadness reading of 58 per cent in 1997 when telling of her “shock and sorrow” at the princess’s death. The same speech, however, returns a higher sadness ranking (59 per cent) when she comments on how “the old Empire has been laid to rest” as Hong Kong’s sovereignty was transferred to China, perhaps hinting at which she misses most.

But what of more positive emotions? Surely she isn’t all anger and disgust? Well, despite barely registering throughout her reign, other than a very brief spike into the high 90s each year as she wishes her subjects a happy Christmas, joy averages 79 per cent in her 1959 address (1958 by comparison has an average of just six per cent, 1960 seven per cent). We were interested to find out why this might be and so explored Watson’s breakdown of this particular speech.

Alas, it turns out the address was no more joyous than other years, but simply a lot shorter. Her first ever pre-recorded message, it lasted for only a minute and consisted almost entirely of her conveying best wishes, rather than this being a small part of a much longer speech where she reminisces about the highs and lows of the previous year. Hence it has thrown averages out.

Measuring her language style rather than emotional probability also shows up some interesting figures. Confidence is generally high right from the start, scoring 90 per cent or above, and in 1959, eight years into the job, it peaks at 99 per cent.

Sporadically throughout her reign, however, this drops, to 42 per cent in 1991 (her speech that year reflected on the end of the Cold War as well as the fortieth anniversary of her father’s death) and as low as 41 per cent in 1995 (the 50th anniversary of the end of World War II).

Watson’s Tone Analyser is, for the most part, designed to be applied to companies’ and individuals’ modern-day communications, and as a result some of its findings are not quite as applicable to the Queen’s historical speeches. So we can take with a pinch of salt that the language she uses in the 70s highlights her as ‘likely to re-share on social media’ or that in the 60s she is increasingly ‘likely to click on an ad’.

But for no reason other than our own amusement it is interesting to include how Watson finds that her language in the 1950s indicates she is least likely to be considering a change in career. Whether or not she has considered a different job since then, or indeed what this might have been, we can only imagine. But as the analysis of her five most recent Christmas addresses shows job satisfaction again on the increase, as well as revealing an increased likelihood to buy healthy food and decreased likelihood to put her health at risk, it looks like, despite being 90, she has no plans of vacating her throne.

Want to know what you or your company’s communications say about you? Find out more about Watson’s Tone Analyser at ibm.com/watson

This article was first published in The Drum's AI issue, guest edited using IBM Watson technology.