Since Stuart Langridge and I released Which Three Birdies, it has taken the web by storm, and we’ve been inundated with requests from prestigious institutions to give lectures on how we accomplished this paradigm shift in non-arbitrary co-ordinates-to-mnemonic mapping. Unfortunately, the global pandemic and the terms of Stuart’s parole prevent us from travelling, so we’re writing it here instead.
A significant advance on its predecessors was achievable because Bruce has a proper degree (English Language and Literature with Drama) and has trained as an English Language teacher. “Which” is an an interrogative pronoun, used in questions about alternatives. This might sound pedantic, but if a service can’t make the right choice from a very limited set of interrogative pronouns, how can you trust it to choose the correct three mnemonics? Establishing trust is vital when launching a tool that is destined to become an essential part of the very infrastructure of cartography.
The mechanics of how the service locates and maps to three birds is extensively documented. Further documentation has been provided at the request of the Nobel Prize committee and will be published in due course
The Web is for everyone and anyone who makes sites that are inaccessible is, quite simply, not a proper developer and quite possibly a criminal or even a fascist. Therefore, W3B offers users the chance to hear the calls of the most prevalent birds in their location, and also provides a transcript of those calls.
Given that there are 18,043 species of birds worldwide, transcribing each one by hand was impractical, so we decided to utilise –nay, leverage– the power of Machine Learning.
Birdsong to Text via Machine Learning
Stuart is, by choice, a Python programmer. Unfortunately, we learned that pythons eat birds and, out of a sense of solidarity with our feathered friends, we decided not to progress with such a barbaric language so we sought an alternative.
Stuart hit upon the answer, due to a fortuitous coincidence. He has a prison tattoo of a puffin on his left buttock (don’t ask) and we remembered that the trendy “R” language is named after call of a Puffin (usually transcribed as “arr-arr-arr”).
Stuart set about learning R, but the we hit another snag: we couldn’t use the actual sounds of birds to train our AI, for copyright reasons.
Luckily, Bruce is also a musician with an extensive collection of instruments, including the actual kazoo that John Cale used to record the weird bits of Venus In Furs. Here it is in its Sotheby’s presentation case:
Whereas the kazoo is ideal for duplicating the mellifluous squawk of a corncrake, it is less suitable to mimic the euphonious peep of an osprey. Stuart listens to AC/DC and therefore has no musical sense at all, so he wasn’t given an instrument. Instead he took the task of inhaling helium out of childrens’ balloons in order to replicate the higher registers of birdsong. Here’s a photo of the flame-haired Adonis preparing to imitate the melodious lament of the screech owl:
After a few evenings re-creating a representative sample of birdsong, we had enough avian phonemes in the bag to run a rigorous programme of principal components analysis, cluster analysis and (of course) multilinear subspace learning algorithms to learn low-dimensional representations directly from tensor representations for multidimensional data, without reshaping them into higher-dimensional vectors.
All known birds can now reliably be transcribed with 94% accuracy, except for the Crested Anatolian Otter-Catcher. We suspect that the reason for this is the confusion introduced by the Turkish vowel harmony and final-obstruent devoicing. In practice, however, this exception doesn’t affect the utility of the system, because the Crested Anatolian Otter-Catcher is now very rare due to its being extensively hunted in the 19th century. (Fun fact, the bird was once so famous and prevalent that the whole region was known as the Otter-munch Empire.)
Hopefully, this in-depth breakdown of how Which Three Birdies? works will encourage other authors of revolutionary new utilities to open-source their work as we have done for the betterment of all humanity. We’d like to thank the nice people at 51Degrees for commissioning Which Three Birdies?, giving us free rein, and paying us for it. The nutters.