Glossary of Technical Terms
Many of the readings assume some familiarity with terms and concepts from
linguistics, biology, psychology, and other fields. Rather than spending
time covering these in class, I will
maintain a glossary of such terms, updated in time for each reading. I omit
terms for which a sufficient entry can be found in Wikipedia or an
online dictionary, foreign-language
terms that are translated sufficiently by
and terms that are irrelevant to the discussion or used in an obscure or
analysis of variance (ANOVA) Like the
ANOVA allows a researcher to determine whether results are statistically
significant or not -- i.e., how likely it is that the results are due
to chance, and not to the experimental conditions. ANOVA is usually the method of choice for more complicated
experimental designs. For example, a two-way ANOVA allows
the researcher to examine simultaneously the effects of two different
variables (like age and income) on a third (overall health). Since
variables are also called factors, such
multi-variable designs are often called factorial, and are
described in terms of the "cross" between the two factors, e.g.,
Treatments X Levels. A nice feature
of ANOVA is that it shows the significance both of main effects
(like the effect of age on income) and interactions between
variables (e.g., does income change the way that age affects health?).
Like the chi-square
value in the chi-square test,
ANOVA uses a value, called the F ratio, to express the strength
of the results. Often this value is expressed in terms of the
degrees of freedom
(df): F(2,61) means two degrees of freedom for the variable
in question, and 61 degrees of freedom for subjects. (Because degrees of
freedom is the number of distinctions made, this means that there were
three different values for the variable and 62 subjects.) As with other
kinds of statistical analyses, a low chance probability
p, e.g., p < .05,
is the standard criterion for determining whether results are
anti-mentalist / -mentalism See
articulation The activity of the speech organs (mainly lips and
tongue) in producing speech. Saussure uses it in this modern, standard sense
initially, but then confusingly refers to "language articulation"
(gegliederte Sprache) later in the same introductory chapter --
despite his crucial, fundamental distinction between language and speech.
atomic Yet another word with a strange history and confusing
set of meanings. Since the time of the
philosophers, science has
sought to discover the smallest, indivisible
(lit., a-tomic) units of matter. Shortly after the modern discovery of
the atom, physicists realized that atoms were in fact composed of smaller,
"sub-atomic" particles, but the name stuck. Logic and linguistics preserve
this sense of indivisibility, so that an atomic unit is one that,
like a terminal, cannot be "expanded" into
case Linguists use this term to describe how nouns and
pronouns get marked to reflect their role in the sentence.
The English sentences "She saw him" and
"He saw her" show that English does this to pronouns.
Other languages (German, Russian, Latin, Japanese, Finnish, Turkish) do
it to nouns as well.
case filter A constraint, perhaps one of the constraints
supposed to make up
Universal Grammar, that forces nouns to have an overtly marked case role;
i.e., to be the subject of a verb, the object of a verb or preposition, etc.
The example in
Pinker and Bloom (1990)
is less familiar than common
expressions like "too big of a ______"; e.g.,
He's too big of a schmuck to thank you for a favor. The
usage would be too big a schmuck, but the the noun phrase
a schmuck is "stranded" in this context -- is it a subject, an
object, or what? The Case Filter disfavors such stranded noun phrases.
The solution is to make a schmuck the object of the (meaningless)
preposition of. Ain't language weird?
chi-square Consider the answer to a single question -- e.g.,
"Do you plan to buy a car next year?" -- obtained
from a survey given to two different kinds of people (male / female,
under 30 / over 50, Republican / Democrat, etc.) The
"Null Hypothesis" is that none of these distinctions makes a
difference in car buying, so the average answer obtained from both
groups should be the same. The Chi-Square value tells us the extent to
which the Null Hypothesis is invalid -- i.e., the extent to which a
difference in the average results is meaningful. The df, or
degrees of freedom, is simply the number of
distinctions made in our survey: for two different groups, we have one
degree of freedom. By looking up these two values (Chi-Square and
in a statistical table, we can determine the probability p
of the Null Hypothesis being true. For example,
p < .01 means that there is less than a one percent chance that the
Null Hypothesis is true, making it extremely unlikely that our results
can be attributed to random chance. In the social sciences, a p
value of .05 or less is the typical criterion for publishing your results.
C.I.E. tristimulus values Look
here for a good explanation.
closed classes See
Colorless green ideas sleep furiously Look
competence (vs. performance) The fundamental distinction in
Chomskyan linguistics is between competence and performance.
Competence is the abstract "knowledge" that people have about the
structure of their language, without necessarily being able to express
this knowledge. For example, all English speakers know that the subject
comes before the verb in simple declarative sentences ("John loves Mary").
Probably few English speakers could express this fact as such, but they
clearly rely on it when trying to understand a sentence that they have never
heard before ("Blair loves Bush"). Performance
describes the way that people actually speak, including slips of the
tongue, misunderstandings, hesitation, stuttering, etc. According to Chomsky,
characterizing competence should be the true goal of linguistics.
compositional(ity) A language is said to be
compositional if it uses the same
word or symbol for the same concept in different contexts. For example,
the word "bird" refers to the same basic thing whether it's the subject of
the sentence ("The bird flew out the window."), the object
("I saw a bird in the treetop."), etc. Compositionality is such a
universal feature of human languages that it's difficult for us to imagine
an alternative way of speaking. It is clear, however, that in acquiring
language children start from a non-compositional holophrastic stage,
where a single word is used where an adult would use an entire sentence
(e.g., "milk" can mean "I want milk", "you are drinking milk", "where is the
milk", etc.), and that adults' language contains many expressions whose
meaning cannot be understood from the meanings of their component words
("hello", "what's up?", "holy shit!", etc.) More radically,
researcher Alison Wray
has argued for the holistic (non-compositional) nature of human proto-language,
Jim Hurford has argued that
proto-thought had no names for individual things.
constituent Another word for "phrase"; e.g., noun phrase
(the dog); verb phrase (bit the mailman), etc.
context-free (phrase structure grammar) A
generative grammar containing
phrase structure rules
like S -> NP VP, which only have one symbol to the left of the
arrow ->. I.e., the S symbol has no context, as opposed
to a context-sensitive rule like A B -> C D, where there is
more than one symbol
to the left of the arrow. Although the distinction may seem trivial,
context-free grammars generate a more constrained set of languages than
context-sensitive grammars, the point of Chomskyan linguistics being to
determine the constraints on the possible forms of human languages.
Context-free grammars can also be defined as grammars that generate languages
that can be processed by a
nondeterministic pushdown storage automaton.
control In sentences like John promised to leave,
it's clear that John is doing both the promising and (if he's honest)
the leaving. Chomskyan syntactic theory describes this situation by
saying that the explicit subject (John) controls
an implicit, unpronounced pronoun that is the subject of the second
cyclic ordering A principle of
stipulating that a rule must apply to a structure before applying to any
structure that contains it. The details of this rule are not important to
us, but the rule is a useful example of the sorts of principles that were
supposed to make up
derivation Another linguistic term with several
meanings (see derivational
uses it to refer to a phonological derivation;
e.g., the fact that the final -s in English noun plurals turns into a
/z/ sound after vowels and some consonants (trees, cans), and
into an /ez/ (schwa plus /z/) after others (buses, peaches). In other words,
the /z/ in dogs is derived from an "underlying" /s/.
derivational (and inflectional morphology)
In this sense, derivation refers to processes that change something
about a word, like positive/negative or
part-of-speech (verb, noun, adjective), typically by
morphology. In English and many other languages, this is done with
affixes (prefixes and suffixes). Some derivational
affixes, like un- (positive->negative adjective) and -ness (adjective->noun) are productive, whereas others, like -ity
(adjective->noun) are more like fossilized relics. See also
descriptively adequate/ descriptive adequacy
grammar is said to be descriptively adequate for a given language
if it generates all the sentences of that language. Because human languages
are so vastly complicated, no one has ever come up with a generative grammar
for an entire language. Instead, effort has focused on writing grammars for
an interesting set of phenomena in a given language. e.g., the
relationship between the active ("Oswald killed Kennedy") and passive
("Kennedy was killed by Oswald") forms of declarative sentences in English.
A grammar that not only
describes such phenomena correctly but also makes interesting, falsifiable
predictions about related phenomena, is said to have explanatory
adequacy. For example, one could write a grammar for English verbs by
listing the past-tense form of every single verb along with its present-tense
form, but such a grammar would fail to account for English speakers'
competence in forming the past tense of a novel verb ("google") by adding the
suffix -ed ("I googled it and got thousands of hits.")
discrete infinity Because human languages are
recursive, sentences can contain
an infinite number of discrete elements (words). This term is probably meant
to contrast with the infinity of a single continuous object,
like a line.
E(s) Used by psychologists and sociologists as an
abbreviation for "Experimenter(s)". See also
embedded See recursive
analysis of variance.
felicitous Linguistics jargon for "appropriate to the
finite-state automaton An automaton without any memory; essentially, a program that
goes from one step (state) to another without remembering where it has been.
In an important and influential paper, Chomsky
showed that such an automaton is not powerful enough to serve as a model
of human language.
focus The "new information" in a sentence, typically a reply
to a question. I took out the trash, spoken in reply to
What did you do with the trash?, has the focus on took out;
in reply to What did you take out?, the focus would be on
the trash. Common strategies for indicating focus include intonation
(pitch and loudness): I took out the TRASH, and movement of phrases
from their default position, as in Spanish: Se fue el hombre.
gap The empty space left in a sentence that has undergone a
transformation, like the transformation from a statement to a
(1) John likes Mary
(2) Who does John like _____ ?
The idea of there being an invisible "something" at the end of such questions
may seem odd, but consider the following examples, where (5) is something
you would probably never hear in English (except perhaps as slurred/drunken
(1) I want to win the election.
(2) I wanna win the election.
(3) I want Hillary to win the election.
(4) Who do you want _____ to win the election?
(5) *Who do you wanna win the election?
If the wanna contraction requires that want
come right before to, and there really is a gap in (4), we get a
nice explanation of the problem with (5).
garden paths See
Garden path sentence
generative (grammar) A grammar that can explicitly generate
the sentences of a language. The kind of grammar you learn in school
is mainly prescriptive: it prescribes (dictates) what is "correct"
and "incorrect" English, mainly by appeal to historical tradition or
logic (e.g., a double negative is really a positive, although for some reason
it doesn't work that way in Spanish, French, and lots of other languages).
Generative grammar, by contrast, is descriptive: it aims to
characterize how people actually use language, based on the
competence that they acquire
during childhood. This description often takes the form of rules,
like the classic S -> NP VP, which describes the fact that
an English declarative sentence consists of a noun phrase and a verb phrase.
Because these rules can be recursive
(e.g, S -> S and S), generative grammars embody Chomsky's
early observation that there is no limit to the number of sentences in a
goal If the
is the one performing an action, and the
is the one on whom the action is performed, then the goal is the
person, place or thing toward which the action is directed:
John (AGENT) gave a book (PATIENT) to Mary (GOAL).
head The head of a noun phrase (the blue cup)
is a noun (cup); the head of a verb phrase (really likes books)
is a verb (likes), etc.
homocentric This word can mean either "concentric" (having the
same center), as in the circles of a bullseye, or "human-centered",
meaning that people tend to project human qualities onto the non-human
world. It is this second, human-centric sense that Zubin and Köpcke
(1986) appear to be using in accounting for the fact that most names for predatory animals have masculine gender in German.
Humboldt Wilhelm von Humboldt (1767-1835)
illocutionary force For a good discussion, look
here. The illocutionary force of an utterance is not
always clear. Consider the story about the Russian man who, standing to
inherit a small fortune from his uncle, put a classified ad in the newspaper
saying "I'll bet anyone 200,000 rubles that my uncle lives another year".
immediate dominance rules Part of the set of constraints
that supposedly make up
inductive process / induction The process of making
generalizations based on specific cases.
inflection Another confusing term. Although most
people use it to mean "tone of voice" (which linguists call
intonation), linguists usually use inflection to refer to the
process of adding prefixes and suffixes (or making other changes) to words,
for the purpose of
subject-verb agreement ("He leaves at 6PM"), tense distinctions
("Curiosity killed the cat"), gender agreement
(Ô femme dangereuse), etc.
intention(al(ity)) Intentionality refers to the
beliefs, desires, and other mental states that we attribute to other
human and non-human beings. Although it seems reasonable for us to attribute such states to
non-human animals, there is debate surrounding whether such animals
can make the same kind of attributions about each other. E.g., can a
monkey "understand" that another monkey wants to do something? See
also Folk Psychology.
analysis of variance.
innate idea See Kantian
possibly the most influential Western philosopher of all time,
was concerned (as were Plato and many others)
with the classic nature/nurture issue of how much we are shaped by experience,
and how much is "innate" or built-into our human nature. For example,
discussion suggests that Kant saw
basic notions like causality (something causing something else to happen)
and objecthood (how we tell whether something is a single object or not) as
being preconditions to experience, things so fundamental that we could not
possibly learn them by an inductive process
. The experiments described here suggest that the innateness of
objecthood is still an open research question. Chomsky's concern, of course,
is with how much of language is innate, and how much can be learned. His
answer is, roughly, "all the important stuff is innate, and very little is
!Kung San As you can see
the exclamation point is not a typo; it represents a
postalveolar click consonant. As with
common ideas about language, there is some popular
about who speaks the languages containing these sounds. To listen to an
actual sample of someone speaking one of these languages (not necessarily
latency Response time in an experiment. E.g., how long a
subject takes to say the word "blue" after being shown a picture of a
lexicon The dictionary of arbitrary sound/meaning
correspondences for a given language, similar to a traditional dictionary.
As in a traditional dictionary, linguists put words into various
lexical categories: noun, verb, adjective, etc.
linear bounded automaton An
that can recognize (parse, understand) the sentences output by a
local maximum (plural maxima, also called
a local optimum) A "good enough"
solution to a particular problem, for example, the problem of vision.
The idea is that there may be better solutions (the human eye is far from
perfect), but modifying a locally optimal solution will likely just make
things worse. This is why genetic mutation, though necessary for evolution,
is almost always harmful in a particular case. See also
logical scope People who complain about the supposed misuse of
the word hopefully are complaining about its logical scope. In the
"correct" usage, hopefully has scope only over the verb, so
He hopefully left means something like He left in a hopeful state of
mind. In its more common ("incorrect") usage, hopefully has scope
over the entire proposition (sentence), making this example sentence mean
I hope that he left. As
Pinker and Bloom (1990)
note, this kind of
scope correlates with a more peripheral position in the sentence; e.g.,
Hopefully these assholes will stop telling us how to speak correctly.
main effect See
analysis of variance.
marked (word order) A separate sense from
case marking, this use of "marked" means
"unusual" or "atypical". For example, the usual word order for declarative
sentences in English is SVO
(Subject-Verb-Object), so violating this order in a declarative can be
used emphasize the object: Him, I don't like!
monotonic function To say that "the meaning of an expression
is a monotonic function of the meaning of its parts"
(Cann 1993) means that as you keep adding words to the
expression, the meaning can only grow more
complicated, not less.
motor program A stereotyped sequence of actions performed by
an animal without conscious control; e.g., walking, chewing.
nativist Someone who believes in
neural network Look
here for a good
definition. A recurrent neural network is a neural network
with feedback: its current output depends not just on its input, but also
on its previous output. Because of the sequential and
recursive nature of language,
this sort of feedback is essential if one wishes to use neural networks to
nondeterministic pushdown storage automaton The class of
automaton (computer program) that can recognize (parse, understand) the
sentences of a
language generated by a context-free phrase
structure grammar. Grammars and automata are two sides of the same
coin. See also
p < .01,   p < .02,   etc. See
for a good discussion.
performance (models) See
phase change By analogy with its
in physics, this term refers to an abrupt quantitative change in a value
that suggest a new qualitative situation. An example is
Figure 3 of
Kirby (2000), where the number of rules in an evolving
grammar makes a sudden jump.
philosophical grammar(ians) Chomsky seems to be using this term
to refer to philosophers (notably,
some ancient Indian and medieval European thinkers) who
tried to describe universal features of human language. It is extremely
important for Chomsky to distinguish his Universal Grammar from these
earlier ideas about language: according to Chomsky, it is the
constraints on possible grammars, and not the grammar itself,
that are universal.
phonation For the modern definition, look
unconventional usage of "organs of phonation" to refer to all of the speech
organs (as opposed to just the larynx) may be a matter of translation.
phonetic features describe the ways that people
produce and perceive the sounds of speech. For example, sounds produced
in the front of the mouth (like the /t/ in "top" and the vowel sound
in "see") are said to have the feature "frontal"; sounds produced with the
lips (like /p/, /b/, /v/, and /f/) are said to be "labial", etc.
Describing sounds this way allows phonological rules (see below) to
be simple and general.
phonetic gestures Another confusing term. Gesture
here refers to the movement of the speech
articulators in making various
sounds, not to hand gestures.
phonemic analysis a technique developed by structural linguists
for classifying the sounds of language. The idea is to find pairs of sounds
that make a contrast in some context; e.g., the first sounds in the words
"late" and "rate". The fact that these sounds make a meaningful contrast tells
us that /l/ and /r/ are two different phonemes in English. Japanese
(for example) does not have a phonemic contrast between /l/ and /r/, and
English lacks phonemic contrasts found in some other languages. Chomsky's
basic critique of structural linguistics was that this sort of classification
tells us very little about how these sounds (and other elements, like words)
are put together in actual language. For Chomsky, knowledge of language is much
more than just knowledge of minimal contrasts. The death blow for
phonemic analysis had already come from Chomsky's MIT colleague Morris Halle,
whose 1959 work The Sound Pattern of Russian showed that the
phonology of Russian cuts across phonemic categories, making phonemic analysis
an irrelevant distraction.
phonological rules describe the ways that speech sounds change
in various contexts. For example, the plurals of most English nouns end
in the letter "s", but this letter sounds like a /z/ after vowels and some
consonants ("trees", "cars"), and a vowel (schwa) is inserted before the /z/
after some consonants ("buses", "peaches"). Phonological rules are usually
based on phonetic features.
phrase structure (component) in early versions of Chomsky's
generative grammar, the phrase
structure component contained
rewrite rules for generating the basic, or
form of a sentence. E.g.,
S -> NP VP
VP -> V NP
are rewrite rules specifying standard
subject/verb/object order for
simple declaratives. This deep structure would then undergo various
transformations to convert it into a surface structure having
more or less the same meaning. The classic transformation was the conversion
from the active form of a sentence (subject/verb/object, "John loves Mary")
into its passive form ("Mary is loved by John"), although this transformation,
and the whole notion of specific transformations in specific languages,
fell out of favor as the field developed.
here for a
good discussion. Chomsky's
competence/performance distinction seems very much like Plato's distinction between the ideal forms and their earthly manifestation in physical objects.
preterminal category In a
rule like N -> John, the N (noun)
is a preterminal category; i.e.,
one that goes directly to a
terminal (in this case, John).
terminals are also known as
lexical items, preterminal
categories are often called
prisoner's dilemma You and your friend have both been arrested
and are being interrogated by the police in separate rooms.
If you rat each other out, you both get two years in jail. If you both
keep quiet out of loyalty to each other, you both get six months. If one
of you keeps quiet and the
other rats, the rat goes free and the loyal sucker gets 10 years. So
you do? Answering this question is the job of
game theory, a
discipline which turns out to have a lot to do with evolutionary biology,
international diplomacy, and other fields where the value of pursuing a
given strategy depends on the strategies pursued by others.
proposition(al logic) In philosophy, a proposition is a
statement that is either true or false (e.g., "All birds fly"). As Chomsky
points out, propositions constitute only one part of language, so efforts
to analyze language solely in terms of propositional logic are unlikely to shed
much light on the nature of language.
that change the intonation of entire phrases or sentences; for example,
the rule that yes/no questions in English must end in rising pitch
(Did you take out the trash?).
Compare segmental rules.
rank order correlation Correlation refers to
the degree to which two measured values agree with each other, whether
or not one necessarily causes the other. E.g., smoking is highly
correlated with respiratory ailments, and likely causes them, but
although height and weight are highly
correlated, one does not cause the other. Rank order is a way
of arranging the elements in a set (like students in a class), according
to some measurable criterion (like grade point average). So, rank
order correlation is a way of correlating two values according their
rank order, instead of their actual number. An example might be
future income bracket as a function of class rank.
real-time deterministic automaton
Obviously some sort of
automaton, presumably more constrained and biologically
plausible than a
nondeterministic pushdown storage automaton.
Chomsky's point here seems to be that the
component of the grammar
is less important than the
recurrent neural network See
recursive/recursion In linguistics, this term refers to the
fact that a sentence or phrase can contain (embed)
another sentence or phrase --
much like a box within a box, or a picture of someone holding a picture.
Common recursive structures include (1) subordinate clauses; e.g.,
He said that she left, where she left is itself a sentence;
(2) relative clauses; e.g., She's the one who took the book;
(3) coordinated clauses; e.g., She took the book and left, which
contains two verb phrases connected by the coordinating conjunction and.
especially the Indo-European material in the
Function and Meaning section.
rewrite rules Also called
phrase structure rules.
Kirby (2000) uses a special kind of rewrite rule that
specifies not just syntactic (structure) but also semantic (meaning)
Rogerian analyst See
S(s) Used by psychologists and sociologists as an
abbreviation for "Subject(s)". See also
segmental phonology The study of
that affect individual
sounds, like the rule for English plurals -- as contrasted with rules
that change the properties (pitch, intonation) of larger units
(syllables, phrases); or the set of such rules in a particular language.
Compare prosodic rules.
serial interface like Saussure's linearity property, this
term expresses the fact that we communicate in a series of
discrete units: phonemes, syllables, words, etc.
spectral (changes) The pitch of a sound (like the human voice)
can be distinguished from the additional spectral components of the
sound, sometimes called overtones or resonances. Spectral information is what allows you to distinguish between, e.g., a note played on a flute and the same note
played on a guitar, the difference between the vowels in "hit" and "hat", etc. In human speech, much of the information is contained in the changes , or transitions in the spectrum of the speech signal, as opposed to the
absolute spectral values themselves. See also
stop See stop consonant.
subjacency A constraint on grammatical
transformations that prohibits, e.g., questions like (4) below:
(1) Grover stole a car.
(2) Which car did Grover steal?
(3) Ernie saw a monster who stole a car.
(4) *Which car did Ernie see a monster who stole?
successor function A function that allows
the natural (counting) numbers to be defined recursively,
providing the basis for arithmetic; essentially, the function that adds
one to a number. If we accept that zero is a number, and that the successor
of any number is a number (note the embedding of number in that
second part!), we can define all the numbers recursively, without having
to list them.
Hauser, Chomsky and Fitch (2002)
suggest that the ability of humans to think this way may be related to,
or even the basis of, linguistic recursion, i.e., the
"Faculty of Language -- Narrow" that we do not seem to share with other species.
synthetic Synthetic languages like Russian and
features like noun suffixes to indicate the
of nouns. Such language tend to have a freer word order
than analytic (a.k.a. isolating)
languages, like English and Chinese, that don't mark nouns in this way.
terminal symbols The symbols in a
phrase structure grammar
that can only appear on the right side of the arrow: essentially, the
symbols that actually appear in a sentence (John, loves, and
Mary in John loves Mary),
as opposed to the abstract symbols (NP, VP).
See also preterminal category.
here; see also
Trubetzkoy Nikolai Trubetzkoy (1890-1938)
(two-tailed) t-test A statistical test used in cases where
the variable of interest isn't a categorical (yes/no, tu/vous)
distinction, but rather a measurement (IQ, lifespan, etc.) The t-test
deals with the same
issues (degrees of freedom, probability of results due to random chance)
as the Chi-Square test, and is
often considered a simple form of analysis
(two-way) analysis of variance See
Analysis of Variance.
[+voice], [-voice] See
VOS language Languages are often classified in terms of the
way that they order the subject (S), verb (V), and object (O) in the sentence.
So a VOS language is one in which the order is verb-object-subject. This
order, evolved in one of the simulations reported in
Kirby (2000), is actually rare or unattested in actual
human languages, which strongly favor putting the subject before the object.
Hence, the only widely-attested orders are SVO (e.g., English),
SOV (e.g., Japanese), and VSO (e.g., Irish). The emergence of an unattested
word order poses no embarrassment in Kirby's model, however, because he makes
no attempt to control for word order, and ends up with other orders
Whitney William Dwight Whitney (1827-1894)