--- # Who are you? --- ### Information I'd like - Year - Major - What languages do we speak? --- ### Two changes - Will's office hours are now Wednesdays 3-5pm - Section A01 (8am Friday) will now meet in HSS 1305 --- # What is Language? ### Will Styler - LIGN 101 --- ### Today's Plan - Who are you? - What is Language? - What are some characteristics of Language? - How Universal is Language? - What *isn't* language? - Does our language shape our thoughts? --- # What is Language? --- ### Language - Language is a system for linking expression to meaning - ... but we're probably better off thinking about what makes language, language-like --- # What characteristics does Language have? --- ### Three important characteristics of Language - 1 - Arbitrary relationships between signs and the signified - 2 - Fully Productive and Creative - 3 - Speakers have knowledge of grammar and the 'rules' of language --- ### Characteristic #1 ## Arbitrary relationships between signs and the signified --- ### Language has 'signs' and 'signified' concepts - The 'sign' is something a human does in communication - The 'signified' is something in the world (conceptual or concrete) - *What is the link between these things?* --- ### Non-arbitrary signs - *The sign is causally linked to the signified* - Smoke is a sign of fire - Bleeding from your finger is a sign of a papercut - Will's presence is a sign of bad puns --- ### Iconic (or representational) signs - *A referential abstraction from the signified* - # ☂ ☃ ☀ 💩 🤦 - Some hand gestures - Some sounds meant to 'mimic' the world ---
--- ### Arbitrary Signs - *Referential only by agreement and widespread knowledge* - Non-representational symbols (or uses of symbols) (♄, 🍆, 🍑) - Non-iconic gestures --- ![](img/scotland_speedsign.jpg) - "Go whatever speed is legal here" --- ### 🔥 in 'omg he started a 🔥 in his trashcan' is ... A) Non-Arbitrary, Iconic B) Arbitrary
--- ### 🔥 in 'omg he started a 🔥 in his trashcan' is ... A)
Non-Arbitrary, Iconic
B) Arbitrary --- ### 🔥 in 'omg that party was 🔥 🔥 🔥 ' is ... A) Non-Arbitrary, Iconic B) Arbitrary
--- ### 🔥 in 'omg that party was 🔥 🔥 🔥 ' is ... A) Non-Arbitrary, Iconic B)
Arbitrary
--- ### The most important arbitrary signifiers are... --- ## Words! - The connection between words and meanings in language is (mostly) arbitrary --- ### The connection between sound and meaning is mostly arbitrary in spoken Language
--- ### Signed languages have arbitrary signs too! (These ASL examples are from [Sign with Robert on GIPHY](https://giphy.com/signwithrobert/))
--- ### Signed languages have arbitrary signs too!
--- ### Signed languages have arbitrary signs too!
--- ### ... but there are often elements of iconicity in signs
--- ### ... but there are often elements of iconicity in signs
--- ### ... but there are often elements of iconicity in signs
--- ### Spoken language has moments of iconicity too! --- ### Onomatopoeia *Words which, when spoken, sound kind of like the things they represent* * (See also the idea of an [Ideophone](http://en.wikipedia.org/wiki/Ideophone)) --- ### Onomatopoetic Words in English Boom Gurgle Woof Snip Whoosh Splash --- ### Onomatopoeia differs across languages --- ### A rooster says... cock-a-doodle-doo (English) kukko kiekuu (Finnish) chicchirichí (Italian) kuklooku (Urdu) kukuriku (Hungarian) Sourced from [this awesome site](http://www.eleceng.adelaide.edu.au/personal/dabbott/animal.html) --- (So, different languages can't even map sounds onto sounds uniformly) --- ### The best evidence for arbitrariness: Words differ across languages - If the link between signifier and signified were non-arbitrary, *everybody would have substantially similar words for concepts* --- ### Spoiler Alert: We do not
--- ### Characteristic #1 ## Arbitrary relationships between signs and the signified --- ### Characteristic #2 ## Languages are fully productive and creative --- ### Let's test that out - Let's describe an image that you've never seen before ---
--- ### This was not created by a human, so it doesn't follow human categories - This was generated by [Stable Diffusion](https://en.wikipedia.org/wiki/Stable_Diffusion) - The prompt was 'beach san diego kitten spaceship silverware motorcycle dinosaur by Thomas Kinkade' - This is one of the weirdest natural language processing examples around! - **Yet we can still describe the output!** --- ### Languages are capable of describing anything (eventually) - Might take time - Might require some words to be 'circumlocuted' or invented - "I don't know a word for this, but I'll use other words to describe what I'm after' - Might require incredible skill - "They should have sent a poet" - But they'll get the job done --- ### Languages are also capable of producing an infinite set of unique sentences - Any human speaker of a language can produce a sentence that nobody has ever spoken before - "I plan to buy a half-empty barrel of Doritos as soon as my penguin gets home with the Crown Jewels." - "I sure am glad that Michael Bay has agreed to direct a movie adaptation of my favorite childhood cartoon" --- ### ... and sentences can be infinitely grown and expanded - I saw Anna. - I saw Anna yesterday. - I saw Anna yesterday before I refilled the hummingbird feeder. - I saw Anna yesterday afternoon right after writing a bad pun but before I refilled the hummingbird feeder. - I saw my neighbor Anna who likes to hug polar bears yesterday afternoon right after writing a bad pun but before I refilled the hummingbird feeder. --- ### This particular property is called recursion - Sentences can be made to contain elements from other sentences infinitely - "My cat saw the guy who went with Kati, Vincent, and Victoria to the park where the New York Knicks talked to Mary..." - This can go on literally *forever* ---
--- ### Creativity and Productivity is true of all languages - We have yet to discover a language which is communicatively 'primitive' or unable to describe reality - If a language lacks elements, they will be developed or borrowed over time - Languages can differ substantially still - A given language may lack words for some concepts - ... or may have words that other languages don't have - Many words and concepts are culturally based, and may not be shared regardless - **Every natural human language that we've found is able to express a rich set of meanings** --- ### Characteristic #2 ## Languages are fully productive and creative --- ### Characteristic #3 ## Speakers have knowledge of grammar and the 'rules' of language --- ### Grammaticality Judgements - Some things sound 'like English' and are comprehensible - Some things are not English --- ### You can make grammaticality judgements - 'John went to the park with Tamara' - 'Walk Rick gate Johnny between' - 'The moose fiendishly decalcified the hieroglyphics between the Dagestani Giraffes' - 'Somebody set up us the bomb. All your base are belong to us.' --- ### Grammaticality judgements are about language, not individual words - 'Gleeble' - A small fish - 'To Garflabble' - To perform an elaborate mating display - 'Flundubble' - A large octopus-like creature --- 'Gleeble' - A small fish 'To Garflabble' - To perform an elaborate mating display 'Flundubble' - A large octopus-like creature
### Is the sentence 'Twelve Flundubble garflabbling gleeble.' grammatical? A) Yes B) No --- 'Gleeble' - A small fish 'To Garflabble' - To perform an elaborate mating display 'Flundubble' - A large octopus-like creature
### Is the sentence 'Gleebles garflabble for flundubbles all the time.' grammatical? A) Yes B) No --- ### (and by the way, how do we know the plural of 'gleeble'?) --- ![](humorimg/conspiracykeanu.jpg) * (Nah, it's just phonology) --- ### Is the following sequence grammatical? --- ### How about this one? --- ### Judgements can be more subtle - ?There looks to be a problem outside - ?There sounds to be somebody in there - ?There smells to be something rotten in there - ?There feels to be something in the bag - ?There tastes to be an issue with the cookies --- ### We as speakers have strong feelings about the 'correctness' and 'rules' of language - ... but not so much for other kinds of basic communicative gestures - So, this is yet another element of Language --- ### Characteristic #3 ## Speakers have knowledge of grammar and the 'rules' of language --- ### There are other characteristics that are important in human Language - Language is symmetrical (e.g. we can send and recieve) - Language can talk about situations not in the 'here and now' - Language is made of discrete, isolateable units (like words or sounds) - Language has slang, taboo words - Language accomplishes certain functions (e.g. asking questions, issuing commands) - All languages can be translated into any other - But the three we'll focus on are... --- ### Three important characteristics of Language - 1 - Arbitrary relationships between signs and the signified - 2 - Fully Productive and Creative - 3 - Speakers have knowledge of grammar and the 'rules' of language - **All human languages share these properties!** --- ### Given that human languages seem to share some properties - ... and we're yet to find a group of humans without language - Some theorists have argued that this is evidence for... --- # Universal Grammar --- ### Universal Grammar - An idea initially posited by Noam Chomsky --- ### A Note on the Noam ![](people/noam_chomsky.jpg) --- ### Universal Grammar An idea initially posited by Noam Chomsky - UG holds that there's some genetic basis to human language - Some properties of human language are innate - Not just the ones to do with physical structures - There are a set of constraints governing language that are 'built in' to our physiology --- ### UG is not facts about your individual language - A child of two Mandarin Chinese speakers, raised by English speakers, will speak English - Every language is *not* the same, and linguistic grammar varies substantially - *UG is about biological predisposition to Language, not to A language* --- ### Evidence for UG - No cultures, people, or societies have been found with no language - No other species has been found with human-like language - Children acquire language impressively well - Even in cultures where people don't speak to children - Signed languages show similar properties to spoken languages - Even when they arise in isolation - Some things appear to be universal - The big one is Recursion --- ### Evidence against UG - There doesn't seem to be good biological evidence for it - Many properties of human language can be derived from functional facts about communication - Recursion might not be universal - Dan Everett (and others) claim that speakers of Pirahã lack recursion - Pirahã is a pretty neat language - [Even Chomksy doesn't support this view anymore](https://dlc.hypotheses.org/1269) --- ### Are elements of Linguistic structure innate? - There was a major change between the 9th (2010) and 11th (2018) editions of the textbook - Innatist arguments have gone from 'unquestioned' to 'questionable' in the field - Lots of smart folks have lined up on both sides of this argument - ... and there's no sign of a consensus yet! - I just want you to know the idea behind UG --- ### But we've now got a pretty good handle on what Language is --- # What *isn't* Language --- ### Alternative means of coding languages - [Semaphore](https://www.omniglot.com/writing/semaphore.htm) and Morse Code - These are just re-coding an existing language - Writing systems - They're closely associated with an existing spoken or signed human language --- ### Computer Programming Languages - Languages like COBOL, C++, Java, Python, Rust ---
--- ### Are programming languages languages? - 1 - Arbitrary relationships between signs and the signified - 2 - Fully Productive and Creative - 3 - Speakers have knowledge of grammar and the 'rules' of language --- ### Are programming languages languages?
1 - Arbitrary relationships between signs and the signified
2 - Fully Productive and Creative 3 - Speakers have knowledge of grammar and the 'rules' of language --- ### Are programming languages languages?
1 - Arbitrary relationships between signs and the signified
2 - Fully Productive and Creative
3 - Speakers have knowledge of grammar and the 'rules' of language --- ### Are programming languages languages?
1 - Arbitrary relationships between signs and the signified
2 - Fully Productive and Creative
3 - Speakers have knowledge of grammar and the 'rules' of language
--- So, no. * ### (Current) Computer Programming languages are not Language --- ### (Sorry, Hal)
--- ## Do any non-humans have language? --- ## Yes, but only one so far --- ### Large Language Models can produce human language - Large Transformer-based Neural language models (like ChatGPT or OpenAssistant) are now capable of producing robust, grammatical, and used-as-humans-would language - They have 'passed the bar' for many linguists as competent language users - Most of the historical problems language models faced are now 'solved' - *This is different than saying 'These are true AI'* - [They are much less efficient language learners than children](https://arxiv.org/abs/2208.07998) - ... and work is ongoing to see whether they can do more with less --- ### Aside: ChatGPT has no plausible 'universal grammar' - The code doesn't include grammatical facts, a 'language organ', or innate grammatical patterns - English models have accidentally learned some other languages - (e.g. it appears to be able to use one languages' patterns to generalize) - [Some UG Folks are still fighting](https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html) (also [this](https://mronline.org/2023/04/24/chatgpt-and-human-intelligence-noam-chomsky-responds-to-critics/)) --- ## There is finally another kind of thing which can do human language - ... and it's a very large matrix of numbers. - Weird. --- ### What about living creatures? --- ### Lots of animals communicate - Mating calls - Alert calls - Territorial displays --- ### Mimicry
--- ### Nim Chimpsky
--- ### Nim Chimpsky - A chimpanzee raised in a human family to learn ASL by Herbert Terrace - Learned 125 signs (maybe fewer) - Combined signs to produce 2-4 word utterances --- ### Nim Chimpsky didn't seem to have full-on language - Never got past 2-4 sign utterances - 'Nim Banana Me Eat' - Only 7% of utterances served to advance the conversation - 39% of utterances were repetitions of what had just been signed - Entirely 'pragmatic' use of language - Terrace argued that Nim never learned anything - There were some methodological... oddities --- ### Nim led a pretty sad life - Became rather strong and violent, as male chimps do. - Later given back to a research facility and used for medical research - Still used signs (e.g. 'play', hug') from time to time - Died at an animal rescue ranch of a heart attack - Featured in the (sad) documentary [Project Nim](https://www.imdb.com/title/tt1814836/) --- ### The Honeybee 'Waggle Dance'
--- ### Bat Vocalizations - According to [Prat et al. 2016](https://www.nature.com/articles/srep39419), bat vocalizations have detectable information about... - The speaker - Behavior context (e.g. feeding, mating, perching, sleep) - The desired listener --- ### Prairie Dogs
--- ### Prairie Dog Language? - [Dr. Con Slobodchikoff](http://jan.ucc.nau.edu/~cns3/) argues that prairie dogs have... - Different calls for different species of predators - Calls containing descriptive information about location, size, shape, color of predators - Prairie dogs will alarm differently on the basis of your shirt - Yes, prairie dogs are judging your fit - Some signs of new calls for shapes - [A great video on this](https://www.youtube.com/watch?v=y1kXCh496U0) --- ### But is it language? - We've only seen one domain of 'language' - A rich vocabulary of various forms of 'Oh crap! Hide!' - There's no sign of non-pragmatic language use - e.g. "Hey Herbie, remember all those grubs we ate last month? Yeah, that was great" - So, we'd need more data to call this 'Language' - ... and I need to see this work replicated! --- ### Whale whale whale - We've just discovered that whales produce vowels as a part of their clicks - This points us towards new approaches of understanding whale language! --- ### So, do non-human animals have Language? - Nobody's truly doing human language yet - There's evidence of complex information being encoded - ... but we're still gonna need more evidence before it's 'Language' --- ### Quick note - I expect this portion of the lecture to continue to change during my career - There are too many interesting studies coming out - I didn't expect language to emerge in LLMs, and then it did - ... and the move away from UG in the field will broaden some perspectives - We don't necessarily understand the means by which animals communicate - An alien who doesn't see visible light or hear frequencies between 20Hz and 20kHz would think us telepaths - At some level, western society is *very* interested in *not* discovering animal language - Doing so would trigger many uncomfortable conversations! --- ### So, we know what language is. - We know it's a universal thing, although with variation - And we know that animals don't seem to do it (yet?) - Now, for the (potentially) scarier side of Language --- ### We think in language, regularly - We use linguistic concepts in categorizing the world - We discuss the world using our most available set of terms - We lack terms for some more abstract feelings - So, language is a part of our thinking process - This leads many to ask... --- ## Does the language we speak determine how we think? --- ### The Sapir-Whorf Hypothesis - Also known as 'Linguistic Relativity' - 'The language you speak influences how you think about the world' - This is fascinating and terrifying --- ### There are many possible forms - **No effect**: 'The language we speak has *zero* effect on our thought or cognition' - **Weak**: 'The language we speak has some effect on our thought and cognition' - **Strong**: 'The language we speak guides our thought and cognition' - **Determinism**: 'The language we speak *determines* our thought and cognition' --- ### The debate is now mostly about effect size and domain - It's clear that there's *some* effect - [See Dr. Lera Boroditsky at UCSD make a great case](https://www.ted.com/talks/lera_boroditsky_how_language_shapes_the_way_we_think) - But it's also clear that it's not deterministic - We can recognize and discuss objects, events, or colors which we don't have words for - The debate is now focused on the size and nature of the effect - ... and whether/how much it *actually* matters in our lives --- ### This is an amazing area of research - ... and it's one I wish we could spend more time on --- ## For next time... - Watch [Dr. Boroditsky's TED Talk](https://www.ted.com/talks/lera_boroditsky_how_language_shapes_the_way_we_think) - Not required, but worth your time ---
Thank you!