Content & Context: Parrots aren’t just parroting
I’d like to start this post with a video of an argument between an owner and his pet cockatoo — a species alongside parrots in the order Psittaciformes.
It’s uncanny, right? Pebbles’ swagger is a fine imitation of a belligerent drunk. But besides the body language, he is matching an aggressive English intonation and even conjuring up sequences of fightin’ words, including a versatile use of the expletive “fuck”.
Is it any coincidence that Pebbles is recalling these particular sequences of words? Does he know he’s being antagonistic? What if there was a way to tell if your parrot actually knew what the f*ck he was saying? Well, we actually have the beginnings of an answer. Scientists have recorded and coded hours of African grey parrot (Psittacus erithacus) dialogue and applied a nifty statistical analysis method, the Hyperspace Analog to Language (HAL). This analysis has revealed the surprisingly generative, context-sensitive linguistic ability of parrots (Kaufman, Colbert-White & Burgess, 2013).
Before going any further, let’s break down what HAL actually is. HAL is a high-dimensional clustering model designed by Lund and Burgess (1996). It can be fed a corpus of text in order to determine the presence of higher order semantic representations, or “classes” of words, in the mind of the speaker. It does this by identifying incidents of global co-occurrence. In plain English, this is when words are used in similar contexts across a text. Let’s consider an example of someone giving directions to a driver:
Turn left on that road… Now, turn right on this street.
There is global co-occurrence of the words road and street, as both are “things you can turn on”. This contextually similar usage indicates the probable existence of a higher order conceptual representation of a set of semantically related items in the mind of the speaker.
This concept can be contrasted with local co-occurrence, which is simply when words appear next to or near each other. Returning to our example, left and down exhibit local co-occurrence, and cannot constitute evidence of any semantic relationship in the same way as global co-occurrence. HAL is unique, since most linguistic statistical analyses rely on word order and local co-occurrence.
In their 2013 study, Kaufman, Colbert-White, and Burgess decided to feed several hours of transcribed, coded dialogue between Cosmo, an African grey, and her owner Beatty Jean (BJ).
Of course, besides English words, Cosmo’s vocabulary contained many non-word sounds, which were grouped into categories in the figure below. I can’t imagine how bloody long it must have taken to code this bird-talk…
Additionally, the corpus was split into three separate social conditions: when BJ was in the same room as Cosmo (IN), when BJ was in an adjacent room within earshot of Cosmo (OUT), and when BJ was completely absent and Cosmo was left with their two dogs (AL).
The results were fascinating:
Firstly, Cosmo learned more individual words than stock expressions, as evidenced by the majority of word-level clusters as opposed to phrase level clusters.
The most common global co-occurrence category was “things Cosmo wants”, which included whistle, talk, come, cuddle, and shower. All were preceded by the word “wanna”.
Secondly, Cosmo was not randomly recombining words and non-words. Non-word sounds were mixed in with English words according to globally occurring themes, for example dog sounds were found together with dog-related words, phone sounds with phone-related words.
Thirdly, Cosmo had a social awareness of language use. Cosmo only used requesting clusters in the IN condition (e.g. “wanna grape”). Cosmo ceased requesting and mostly made two-sided contact call clusters in the OUT condition (e.g. “where are you?”), which elicited responses from BJ. In the third condition, Cosmo instead used one-sided contact calls (e.g. “Goodbye”) combined with dog and phone clusters.
This study provides statistical evidence, above and beyond meager anecdotal youtube videos, that parrots (and presumably other Psittaciformes) use language intelligently and have higher order semantic content packed into their peanut brains. In light of this, it seems more likely than not that Pebbles knew exactly how NSFW he was being.
Parrots can be sweet as well. The following video shows a male alexandrine parrot flirting with a potential mate. Apart from being cute, what’s notable are the expressions he uses between kisses. What are you doing? Give me a kiss? This is the exact context sensitive use of language highlighted in this post.
Kaufman, A. B., Colbert-White, E. N., & Burgess, C. (2013). Higher-order semantic structures in an African Grey parrot’s vocalizations: evidence from the hyperspace analog to language (HAL) model. Animal cognition, 16(5), 789–801.
Lund K, Burgess C (1996) Producing high-dimensional semantic spaces from lexical co-occurrence. Behav Res Methods Instrum Comput 28:203–208 Mantel
