عجفت الغور
trail
Language and the Brain (class)
fall 2021 classes
Class 1
What do linguists do?
Linguists as part of cognitive science
There exists some sentences of setnences such that it allows for recursive composition
So once we setup the idea that English cannot be a FSM, what is the data?
Langauges that use case markings
case markings allow it to mtaintain a more consistent structure modulo case marking
chomsky is unconcerned with the statistical metadata that arises from a corpus
“internalist” - grammars and structures within a single sentence
“externalist” - set of sentences within a corpus
for sentences to make sense, we assign a phoronlogical structure to pronounce sentences
which then means a sentence makes sense iff:
syntax (chomsky is here) is correct
phoronological sufficent (this can be derived from statistics)
“there exists a faculty of language” - but is there a single faculty of langauge
Class 2 - 9/13
what is the neo-durkheimian relationship between langauge and thought?
is language just a set of sentences?
in the 70’s, they thought that sentences are a sequence of words or sounds
what does this leave out? -> hidden syntax and semantics
syntax cannot directly be seen
semantics sometimes contain a truth value
oftentimes this forms the \(\{form/sound, meaning\}\)
chomsky identified this as a triplet of \(\{form, syntax, meaning\}\)
sentences in english are infinite due to recursion (aka not finite structured)_
chomsky also distinguished between “lists” and “conjunctions”
lists as \(\mathbb{N}\)
however, langauge (or at least english) is “uninterestingly” infinite
as you push further out and generate more sentences, the sentences themselves become regularized
for chomsky, grammar is the compression of language, and if you memorize words you can generate an infinite set of sentences
also must make a distinction between “actual” and “potential”
“actual” sentences are bound by the maximum speaker time or computation time
“potential” sentences are the infinite set of all sentences generatable
what are the grammatical rules for telling us if a word is an actual word?
“moak” -> is this a word?
we have grammars to say this is a potential word, but to reject a word, we actually do a semi-exhaustive search through our internal dictionaries to find out
some words enter the vernacular after it is used
“jabberwocky”, for example
how do we distinguish words that are archaeic/low frequency from words that are incorrect?
what was the theory of cognitive science before the 60’s?
learning is the intentional modification of the environment in order to produce different future behaviors
what is actually being “learned” when you acquire a langauge?
a child does not do a blind word/object association (i.e. this is a ball)
the faculty of langauge extends beyond this into “teaching” or “learning” a grammar and a generative capacity
grammars have variance built in
Class 3 - 9/15
Mendel’s laws - laws of inheritances that require abstract concepts like genes
organized in such a way that gives rise to laws
linguistic laws
grimm’s law - stated in terms of independent sounds - phonemes
werner’s law
before these system laws, the idea that sound changes happened consistently/completely across langauges
when chomsky says linguistics as a cognitive science
this is not to reject the generalization finding tools
dan everett - linguist in piraha
no recursion in this language
he proposes it comes from culture
many things are understood about linguistics via numbers and color systems
how to understand rules?
through the enforcement of cultural roles
how are the roles communicated and enforced?
what happens if you violate it?
studies about generalizablity of specific groups have never been totally innocent - largely driven by social policy
does grimm’s law focus on the right things?
are phonemes the correct unit of generalization?
the persuasion of a generalization lies in the correlation of its properties
future is tied up with the concept of “irrealis”
modals, conditions
future
counterfactuals
future tense does not seem to be a a fundalmental building blocks of language
present tense is a mess because stateive and others
grimm’s laws are simple, but the explanation could be complicated
“laws” of languages often end up generalizing idiosyncratic effects to that languge
anthropoligists suggest that systematic study of langauge by pieces that exhibit discontinuities reflect culture
is culture downstream of langauge?
compositionality
meaning is composition
labels can be generated from the hash of meaning
label is the form/shape/demarker of the word?
Class 5 - 9/22
Is phornology finite state?
is it less powerful than finite state?
why do we care where in the brain something happens
it matters how the brain recongizes faces, but to do so, we must first learn where
jakobson
invariance
/t/ - phoneme, slashes denote an obstruct t/t/ - goes into \([t^h]\) (as in “top”) and \([t]\) (as in “stop”, this doesn’t actually exist in mandarin)
the other inviariance is when you have two different phonemmes, but how do we know what the nominative case is?
Jakobson says it is meaning
the features are semantic ones
therefore, it was incorrect to think of this as the level of the phoneme, but rather we should look at it from the meaning
internal/eternal
prinicple of contrast
example: “top” vs “(t^)hop” - your pronouncation of “t” does not change the menaining, therefore the asperation is not used to define meaning
feature grounding - in terms of articulation, Jakobson thought we should use acoustics (wrong)
paper fights with sausser
meaning is distinct because it arises from the rest of the words
cat vs dog, elephant, etc
sausser says language is ungrounded and that it arises from scarcity
it is because meaning can be generated through differences, like a hash function
what about religion ?
what does the invariance come from?
jakobson is still a version of structuralism
are word embeddings structuralist?
quine - early sentence embeddings -> the context is what matters, the company it keeps
possiblies of derivational frequencies
derevational family entropy says there is neural reponse
why do generalizations happen in suffixes in english?
Class 6 - 9/27
Standard linguistics metholdogy is the method of contrast principles
contrast structures that are parallels
examples:
Toler-
Teach-
Clash
tolerant (adj)
teach (/)
Clash (noun)
tolerance (noun)
teach
Clash (verb)
tolerate (verb)
tolerable (adj)
the (/) suffix is somethimes null, which contrasts with the existence sometimes
sometimes the metholodgy tells you the suffix is null
when you read toler[ate|ance|able] your brain generates the frequency distribution
we have entropy/uncertainty over the continuities, called derivational family entropy
derivational family entropy on mandarin?
example:
unflush(/) - no
unflushable - yes
why does (un) require a suffix?
what does linguistics look like with implementation within the brain?
context free - a sentence consists of a noun phrase and a verb phrase
case markings - NP-> NPACC/v-
when a noun phrase is accusative when its next to a verb, which are context sensitive
A context free grammar cannot be duplicated. Lanugages like Yoruba which exhibits duplication, “buildhouse-buildhouse” -> builder
Swiss german is the only clear examples of midly context-sensitive
Context-free is recursive (no memory)
for some languages, context sensitivy is used only for certain features, like reduplication in Yoruba
why is this a problem?
this points to an issue where human langauges have certain mathematical properties that map to the brain
langauges are not evenly distributed between context free & context sensitive distribution
maybe we can’t process certain context sensitivity
finite -> computablably enumerable are rough categories
pholonogy is theorized to be subregular
morpohology and syntax are much more complicated
Class 7 - 9/29
repetition implies context sensitivty
does english have similar properties?
np by np construction
“dog after dog” is different than “little by little”
little by little is not a noun phrase
noun phrases are potentially infinite
heinz and isardi claim that phronology is sub-regular
constraints on grammar
generally deal with locality
grammar trees can show you how things block locality
regular, natural, and CFG’s circle layering
stabler
for each grammar of a specific type (MC) there eixists an efficient parser
stabler - subregularity of syntax gives you constraints on languages
some sentences cannot be extracted from subject and adjunct
called CED by J Huang
for langauges can do subextraction from subject
example: “what did pictures of fall on Fred” -> “what” must be extracted from subject
merge ops
even if you limit ones to subregular, the merge operations does whatever you want
noun-phrase is equivilant to a noun-prep phrase, since a noun-phrase is a merge of noun-prep
let us assume subregular grammar math is the math used for human language -> formal universal
langauges have nouns and verbs as invariants -> Jakobson
Stabler sees that we parse strings of words: word by word
do you
generate sentences on your own until you find a match?
inefficient and non-deterministic
since we do not wait until the end of a sentence before parsing it, model a sentence interpretation as a probability distribution that’s incremental
lineraly porportional to the number of words
using the finite set of words and finite set of rules: the cross product of both gives you the search space
