ling exam 2

symbol

a form that represents something other than itself (ex words and phonemes)

segment

a unit of sound that is the manifestation of a phoneme with acoustics as its form

pure tone

sinusoidal function (ex tuning fork regular waves)

wave details

1Hz = 1 cycle per sec, higher freq = high tone/pitch,

speech sounds

not pure tones, complex waveforms

resonants (e.g. vowels)

sounds produced by vocal tract that differ between different vowels

ideograms

images without symbolism, represents an idea not a word/sound

pictograms

visual/iconic, non lexical symbols

logograms

written character/symbol representing semantic component of lg, ex hieroglyphics or chinese letters (not iconic)

phonographic system

glyphs represent vocal sound

polyphonia

different sound, similar semantics (kǒu ‘mouth’ ; míng ‘speak)

rebus

similar sound, different semantics (ex: wáng ‘king’, wǎng ‘to go toward)

determinatives

glyphs combines with unit characters to resolve ambiguity

code examples

human lg, mathematical expressions, programming codes

statements

expressions that can be true or false

binary numeric systems

on = 1 off = 0

each spot or column is a bit

a chunk of bits is a byte

8 digit system = 8 bits per byte

256 possible values in an 8 bit space

high level code

characters, human-readable commands

low level code

bits, bytes, cryptic commands

code function

to get a computer to do something that a human wants it to do (commands)

function

a code command that performs some operation

modeling language

an artificial construction which performs some of what the real thing does (speech synthesis, voice recognition, search, error detection, pedicitive text)

problems with whole-word text to speech

need multiple versions of each word, intense storage and search, can’t handle novel words

advantages of phonemic text to speech

smaller number of recordings, use program to generate waveform, manipulate turbulence, resonance, loudness, less storage space

disadvantages to phonemic text to speech

imperfect mapping, physical segments differ by context (sounds different depending on vowels/consonants)

concatenate

string objects tg and play them back

parser

some program/device that detects symbolic categories (phonemes or words) from audio input

speech recognition hurdles

variation, ambiguity, noisy signals, search time

variation

people produce different signals, frequencies are relative not absolute, accents (speech recog hurdle, e.g. formants or fricatives)

phonological ambiguity

interpretation of segment depends on context, handled by tracking transitions rather than simple segments

word ambiguity

parser needs more info to choose among options (ex: where wear ware)

noisy signals

mumbled/fast speech, noisy background, parser needs more info to choose

top down

reconstructing message phrase —> word —> phoneme strong —> audio signal

bottom up

reconstructing message acoustic signal —> phoneme strong —> word —> phrase

predictive recognition

every word has a probability of occurrence

conditional probability

each subsequent word has a p based on the previous word (the probability of an event a occurring, given that another event b has already occurred)

corpus

large collection of texts and/or speech

written corpora

amalgamation of texts, books, articles, websites

spoken corpora

amalgamation of recorded conversations eg phone calls (esp useful if transcribed/annotated)

bigram frequency

given x what are the chances of y (the occurrence of pairs of letters)

trigram frequency

given xy what are the chances of z (a sequence of three letters representing one phoneme)

large language models

AI programs

next word predictive model

built with enormous corpora

many predictive dimensions

require trillions of words for training data

whole-word approach to text to speech

database has recording of each word, string them tg and play them back, more natural

phonemic approach to text to speech

database has recordings of phonemes, convert text to string of phonemes

acoustic analysis

physical properties of phonemes differ by context

intonation

the rise and fall of the voice in speaking (a problem w text to speech)

spoken lg

immediate, under expressive

written lg

not bound by immediacy, even more under expressive, editable, zombie rules

immediacy

the length of your turn in an exchange is proportional to the time and effort to transmit it (time-bound and impermanent)

resonants

What type of segment is characterized by regular periodic sound waves that are observable as formants? includes vowels

fricatives

What type of segment is characterized by random and turbulent sound waves?

formant

What term refers to an amplified overtone of a sound wave? (resonating frequencies are harmonics)

abstraction

it reduces an image down to lines, it represents a complicated idea with simple form, little detail even in logograms

discreteness

language can be broken down into smaller, identifiable segments, easily distinguished from one another and can be combined to create new meanings