Nosy algorithms and secret sauces

Posted on

Videos, despite being mightily powerful business tools for engagement and persuasion are, frankly, dumb beasts.

For all their walkie-talkie pfazz, they are black boxes.

Worse: they suffer from their own form of Locked-In Syndrome. Once viewed. They then close up. You know their name. But not what’s going on inside.


Videos, despite being mightily powerful business tools for engagement and persuasion are, frankly,

dumb beasts.

For all their walkie-talkie pfazz, they are black boxes.

Worse: they suffer from their own form of Locked-In Syndrome.

Once viewed. They then close up. You know their name. But not what’s going on inside.

So what we’ve been doing for a while now is taking consumer-made videos (though they can be any video), transcribing them in our own secret-sauce-hand-cooked way, then submitting them to an inquisition of nosy algorithms.

There’s a few other companies out there hard on our heels, pushing us, so we’re moving fast.

We’re heading down the road, turning onto the freeway/motorway/highway (delete as culturally appropriate). Full beam, full throttle, windows down. Spinal Tap on the radio… Ok Ok, maybe that’s too much… but hopefully we’re keeping the competition in our rear view mirror.

Where the others are – generally – is that they are simply transcribing (and sometimes translating) your videos, then sometimes submitting them to a (usually) standard search engine.

Simple. Works. In a way.

Yet sometimes doesn’t work. Especially if you submit the videos for machine transcription and translation.

Now if you ever switched on the auto-transcriptions services on YouTube, you’ll know how bad auto-transcription can get. Perfection will come one day. Sure as eggs is eggs. But not quite this day…

And even if you had the perfect transcription, running that through a machine-translator gets you into Monty Python territory…  (sorry, but it can and does happen). Grab a load of this real world example that came up in one of our tests in machine translations:

Machine Translation from Mandarin

(Female) This is her eat now, is what? Ice cream is a Magnum of vanilla ice cream.

(Female) I do not want to eat.

(Female) eat a lot of the right, the line, you put your little ass wipe it.

(We’re serious, this is what came out.)

Human Translation from Mandarin 

(Mother:) The little princess is called Yu Bei. Her face is like a dirty cat’s face. This is what she is eating…what’s this?

(Daughter) Ice cream.

(Mother) Ah it is a vanilla flavour ice cream of Magnum.

(Daughter) I don’t want to eat it any more.

(Mother) Ah, you have had a lot, don’t you? Now clean your little lip.

Ok – so that’s easy to fix. If if a company wanted to.  And some do offer a human alternative.

But remember that inquisition of nosy algorithms?

Because what we then do to our clients’ videos is send them for immediate and automatic semantic and sentiment analysis – so we’re looking at not just what they say, and what they do, but what they mean, and if the people in these videos are happy/sad/outraged etc at the time of utterance.  

This also means building a system to take account of how people speak.

Not what they’d write.

Because they are 2 different languages.

What this means for you – if you’re a client of ours – is that nuggets of partially remembered gold be re-found, AND accidental nuggets will also be discovered.  

Whole damn seams. 

So even though folks may not be saying the same thing that you typed in your search field… if it means the same thing, we show it to you.

For simplicity’s sake, let’s take bread AND cheese where the sentiment is positive.

I love the way this [brand name inserted here by our transcriber, even though it isn’t mentioned] cheese tastes on my toast

To us means the same as

I adore how this Gruyere goes with this baguette

It doesn’t stop there.

We then apply yet another secret sauce to our vocabularies.

This sauce involves tailoring our entire ontologies to FMCG/CPG vocabularies. 

So if someone is looking for Surf soap powder, they ain’t gonna get someone riding a board on some blue ocean wave.

And there’s more… we can tailor these vocabularies and semantic analyses not just to an individual company,  but to individual people and even to their Project X’s

And there’s more…  You can then filter your videos by socio-demographic data, think

Show me clips from women in Brazil who have kids but aren’t well off.

And there’s more… you can then (as with a few competitors of ours, this we’ll admit) extract individual soundbites from what you’ve found and share with your colleagues and external agencies.

But here’s what’s clever – we Snapchat-style the sharing links so what they contain isn’t there after 48 hours. This means if you share a video outside of your company, be it with an agency, a freelancer, etc, they have 48hrs to view, then it’s gone. No sign-in, no hassle, no lose strings to clear up later. Your nugget of gold remains safe in your video vault.

(c) https://www.flickr.com/photos/orinrobertjohn/

And we’re not stopping there. Of course we’re not… we’re turning up the volume on that Spinal Tap track to 11… but more on that later.