Robots learn to create language

May 17, 2011 by Katie Gatto weblog
Robots learn to create language
The go-to game allows the agents to test their toponymic lexicon by specifying a target location (B) to meet at. Image credit: Ruth Schulz

(PhysOrg.com) -- Communication is a vital part of any task that has to be done by more than one individual. That is why humans in every corner of the world have created their own complex languages that help us share the goal. As it turns out, we are not alone in that need, or in our ability to create a language of our own.

Researchers at the University of Queensland and Queensland University of Technology have created a pair of robots who are creating their own language. The bots, which are being taught how to speak but not given specific languages, are learning to create a lexicon of their own.

The bots, which have been aptly named Lingodroids, consist of a fairly basic setup when it comes to hardware. The robot consists of a that has been equipped with a camera, a laser range finder, and a sonar setup that allows for the mapping and avoidance of obstacle. In order to allow them to talk, they are also equipped with a and speakers.

This video is not supported by your browser at this time.
Video showing Lingodroids playing a location language game. Schulz, R., Glover, A., Milford, M., Wyeth, G., & Wiles, J. (2011) Lingodroids: Studies in Spatial Cognition and Language, ICRA 2011, The International Conference on Robotics and Automation, Shanghai, China, May 2011

The Lingodroids develop language in the way that most human languages have probably developed, but making up words to name the places that they visit, and then share that name with the other around you. The bots basically find something that they have not seen in the past, create a word based on a random combination of , and then tell the other the word that they have just created. Then the robots memory files will link to the word to that specific location.

Currently the robots are learning new words by , and can only name locations, but the researchers hope that in the future these Lingodroids will be able to create a more complex language.

Explore further: Students turn $250 wheelchair into geo-positioning robot

More information: Research paper: Schulz, R., Wyeth, G., & Wiles, J. (In Press) Are we there yet? Grounding temporal concepts in shared journeys, IEEE Transactions on Autonomous Mental Development

More info: itee.uq.edu.au/~ruth/Lingodroids.htm

Related Stories

Toyota's musical robots (w/ Video)

Feb 25, 2011

(PhysOrg.com) -- The odds are that when you talk to the average person on the street they will have an opinion on what type of music is good and what type is bad. We tend to think of music as a human thing, ...

Thanks to RoboEarth the bots can learn on their own

Feb 10, 2011

(PhysOrg.com) -- We tend to assume that robots need human input in order to understand the world around them. In the near future humans may not even be a part of the robotic-learning equation. Soon, robots ...

Psychologist to study how we put stress into words

Dec 06, 2006

How does a child learn that the stress is on the second syllable of giraffe, and on the first of zebra? Is it memory, the structure of the word itself or clues provided by the sounds in the word?

Recommended for you

Simplicity is key to co-operative robots

19 hours ago

A way of making hundreds—or even thousands—of tiny robots cluster to carry out tasks without using any memory or processing power has been developed by engineers at the University of Sheffield, UK.

Students turn $250 wheelchair into geo-positioning robot

Apr 16, 2014

Talk about your Craigslist finds! A team of student employees at The University of Alabama in Huntsville's Systems Management and Production Center (SMAP) combined inspiration with innovation to make a $250 ...

Using robots to study evolution

Apr 14, 2014

A new paper by OIST's Neural Computation Unit has demonstrated the usefulness of robots in studying evolution. Published in PLOS ONE, Stefan Elfwing, a researcher in Professor Kenji Doya's Unit, has succes ...

User comments : 17

Adjust slider to filter visible comments by rank

Display comments: newest first

atom_jarvis
4.8 / 5 (6) May 17, 2011
It's not a creating a language. It's randomly generating syllables to mark previously unidentified areas that the other robot stores and uses as it's marker name.
The concept of places and place marker names were already programmed into each bot, as were the general objectives. But the objective needs be expanded further to simulate any usefulness of language. 1) The language must be useful in explaining a new concept, one thing one bot figures out before the other and identifying nouns doesn't do that, and 2), even if robots have perfect recall the language needs a structure.
hush1
4.5 / 5 (2) May 17, 2011
The 'structure'(for language)used by the researchers:
"...the representation of spatial and temporal concepts."

At the most fundamental,oversimplified level, all language(for all forms of life as well as for all inanimate objects such as bots) is 'shape'.

So your requirement "language needs structure" is fulfilled.

You object:
"Shape and structure are two words which both have only one meaning"

I agree:
"Yes. And the word "meaning" is called "language."
You object:
"The definition for the word "meaning" is NOT "language".
I agree:
"Yes. "Meaning" is 'what happens' when you literally look at shape."
You object:
"What happens" is too vague. What is happening?
I agree:
"Yes. What is happening is when you look at shape, the act of looking at this shape, changes your shape (state), mutual shape change."

Any event, in which any state of that event changes mutually interactively with the state of any other event is language.

The mutual exchange of information.
New information. Created
maxcypher
5 / 5 (1) May 17, 2011
Hush, that was fun to follow the flow of your logic... of course, you realize that we have individual neurons that detect the abrupt change of the intensity of a particular sort of visual input, i.e., an intensity shift that forms a diagonal line of a particular degree. This is the basis of our ability to visually perceive any shape at all. I'm just trying to say that -- if your logic holds -- then this basic flow of information implies a structured language instigated at the cellular level
that_guy
5 / 5 (2) May 17, 2011
I don't like atom's comment, but I grudgingly agree somewhat with him.

Although you could say that because they have terms for places or things, that this is a language, it is only a language in the most basic sense.

By nature of how it is created and evolves, it is a very limited and boxed form of communication. Human 'language' consists of multiple relations and syntax. It has past present and future. Words are almost never created from "random syllables", they are created from relationships to other words - which is why you can have families of words that have similar parts.

We'll call this a baby step for now, and see how things progress. The research is definitely clever.
macsglen
5 / 5 (2) May 17, 2011
I'm not altogether convinced that the bots are creating language. They name new locations (not objects) with a random combination of syllables. It doesn't sound as if they even name objects at all. They could just be making up names for sets of map coordinates.
hush1
5 / 5 (1) May 17, 2011
The corset Atom proposes we wear...ugh. Max. Thks.I selected this logic to loosen the straps.

If I embed a map with a RANDOM grid and points onto a plane, these 'bots' are the way to go. If they arbitrarily labeled each point with "random syllables", fine. Of course "syntax" is not present here. How do I learn their language? The points on the Grid have been assign "random syllables", (or to satisfy purists, yes, elements of sets for sets of map coordinates)by them.

So here I am, in space, in our universe, with bots that have the ability to map and label all points in space. Fine.
I know there is a structure. A Plane with Grid and Points. I listen to them carefully. I hear random noise!! Their "random labeling" makes perfect sense to them! and no sense to me!! There is no way to "decipher" their true random language!

Bots will do what any self respecting human will do.
Assign an ARBITRARY unit of measure and mutually exchange units of that measure and call it communication!
RobertKarlStonjek
5 / 5 (1) May 17, 2011
Yeah, that's clever ~ let the robots invent their own language so that when one says to the other "let's get the humans tonight, while their sleeping" we'll have no idea what they're saying...
Physmet
5 / 5 (2) May 18, 2011
So, let me get this straight. The robot goes into a room and performs image processing to decide whether or not it has been there before. If it has not, it runs a randomizer against a database of sound bytes to come up with a random combination. It stores that sound in a "word" table in its database. It then tells the other robot, which stores the "word" in its database table. And, they're calling that creating a new language? Other than the image processing, that is very basic programming. I think I've missed something.

I'm sure language building for people involved a few more things. Since more than one person sees the same thing, which word wins out? The strongest? The most quickly accepted?

Or what about the ease of remembering it? For something dangerous, a short loud sounding word is best.

I don't think putting together random syllables like tagging a photo on facebook can be called creating language.
Physmet
5 / 5 (1) May 18, 2011
And, what about the "telephone game"? A person comes up with a word, but by the time it moves around camp, it has morphed. Ultimately, the most commonly used version gets used.

I guess what I'm saying is that language is so much more complex. I'm betting the designers of these robots have a much grander goal and approach than this article is letting on. I think the author dumbed it down way too much. At least, I hope that's the case.
that_guy
not rated yet May 18, 2011
Yeah, that's clever ~ let the robots invent their own language so that when one says to the other "let's get the humans tonight, while their sleeping" we'll have no idea what they're saying...

You deserve 10 stars for this quote. And remember folks, kill all humans.
hush1
not rated yet May 18, 2011
Nice humor. :)
And, what about the "telephone game"?


For me, the 'brick wall' is the word 'meaning'.
We all know USE (mutual exchange of information) morphs meaning.
We didn't anchor and sugar coat the meaning of the word 'meaning' with nice things like invariance.

Perhaps if the authors had asserted:
If information has structure, uniqueness and language are sure to follow, as a an immutable consequence.
spectator
5 / 5 (2) May 18, 2011
In order to make a "real" robotic language which functions in the same way as a human language you need a sufficiently designed virtual engine to allow a self-emergent logic structure which can assign data to strings of symbols, and then later "read" them by hashing and parse data to extract meaning.

I also agree that symbols and words should not be "random," as words in real languages obey rules regarding syntax, grammar, prefixes, suffixes and other modifiers which mostly follow the same rules across the board, with the exception of a few "irregular" words, such as "deer" being both the singular and the plural form.

Words like "deer" are actually problematic, since they violate plurality rules, and therefore the exact meaning of the word can only be obtained from other words in the same sentence or following sentence. The ideal would be to produce a language in which this, "I don't know till the next phrase or sentence" problem is avoided entirely.
spectator
not rated yet May 18, 2011
The proposed word "deers" would be superior to the phrases "some deer" and "multiple deer".

Example:

Multiple deer crossed the road.

Some deer crossed the road.

Deers crossed the road.

Sentence 2 is ambigous even with the use of the word "Some" since the word "some" is itself ambigous and could be refering to a single object or to multiple objects, i.e.

"Some guy..."

and

"Some guys..."

are both correct.

While ambiguous words are often required when knowledge itself is incomplete, or when complete knowledge is not necessary, one should still have the language designed to be as specific and efficient as possible.

Words are not simply labels for things.

Some words are clearly "function names" for a function which modifies the meaning, function, and usage of other words or phrases: articles, adjectives, adverbs, pronouns, conjunctions, prepositions, gerunds, infinitives, etc.

Written language is nested in a way similar to the C family.
spectator
not rated yet May 18, 2011
Now what I just said about the C family, I'm sure some may disagree with, but consider this.

In C, we have punctuation and grouping symbols which are not common to ordinary written language, ":,;,<,>,{,},(,)".

Ok, we use parenthesis in ordinary writing, but not very often as compared to C. However, we do use quotes and punctuation at least once per sentence, and as often as every phrase or even every word.

",',.,?,!, etc.

In written spanish, questions and exclaimation marks are even doubled with the preceding upside down symbol, which is very, very reminiscent of the markup of C, HTML, and PHP.

?Como Estas?

<? php some code ?>

some text

Additionally, in English even though we don't write or speak such grouping symbols, they are "understood groups," such as a prepositional phrase.

Our written language is a lot more like a computer programming language than might be at first realized.
hush1
5 / 5 (2) May 19, 2011
...I'm sure some may disagree...


:)...but consider this:

A college professor wrote the words: "A woman without her man is nothing" on the chalkboard and asked his students to punctuate it correctly.

All of the males in the class wrote: "A woman, without her man, is nothing."

All the females in the class wrote: "A woman: without her, man is nothing."
Recovering_Human
not rated yet May 22, 2011
^Though, upon closer inspection, the second sentence is actually nonsense unless you remove the "A ". :)
Eikka
5 / 5 (1) May 23, 2011

Our written language is a lot more like a computer programming language than might be at first realized.


That's because they were made that way, by people who wanted to have the computer understand written instructions in normal language instead of having to think everything in registers and moves.

More news stories

Turning off depression in the brain

Scientists have traced vulnerability to depression-like behaviors in mice to out-of-balance electrical activity inside neurons of the brain's reward circuit and experimentally reversed it – but there's ...

There's something ancient in the icebox

Glaciers are commonly thought to work like a belt sander. As they move over the land they scrape off everything—vegetation, soil, and even the top layer of bedrock. So scientists were greatly surprised ...