Phonology.pptx

MR. JAYANAND KAMBLE
WELCOME TO JAK’S TUTORIAL
Prof. Jayanand Kamble
1

COURSE CODE: 20UIT704EA
TITLE: SPEECH RECOGNITION & NATURAL LANGUAGE
PROCESSING
2
Weekly Teaching Hrs Evaluation Scheme
Credit
L T P CT MSE ESE
3 0 0 20 20 60 3

CONTENT: 03 PHONOLOGY
• Speech sounds,
• Phonetic transcription,
• Phoneme and phonological rules,
• Optimality theory
• Machine learning of phonological rules
3

• Phonological Aspects of prosody and speech
synthesis
• Pronunciation, spelling and N-grams:
• Spelling errors
• Detection and elimination using probabilistic
models
• Pronunciation variations
4

• Decision tree model
• Counting words in corpora
• Simple N-grams, smoothing,
• N-gram for spelling and pronunciation
5

SPEECH SOUNDS
• Knowing a language includes knowing the sounds of that
language
• Phonetics is the study of speech sounds
• We are able to segment a continuous stream of speech
into distinct parts and recognize the parts in other words
• Everyone who knows a language knows how to segment
sentences into words and words into sounds
6

IDENTITY OF SPEECH SOUNDS
• Our linguistic knowledge allows us to ignore nonlinguistic
differences in speech (such as individual pitch levels, rates
of speed, coughs)
• We are capable of making sounds that are not speech
sounds in English but are in other languages
7

• The science of phonetics aims to describe all the sounds
of all the world’s languages
• Acoustic phonetics: focuses on the physical properties of the
sounds of language
• Auditory phonetics: focuses on how listeners perceive the
sounds of language
• Articulatory phonetics: focuses on how the vocal tract produces
the sounds of language
8
उच्चार
श्रवणववषयक
ध्वविक

THE PHONETIC ALPHABET
• Spelling does not consistently represent the sounds of
language
• Some problems with ordinary spelling:
• The same sound may be represented by many letters or
combination of letters:
• he people key
• believe seize machine
• The same letter may represent a variety of sounds:
• father village
9

• A combination of letters may represent a single
sound
• shoot character Thomas
• A single letter may represent a combination of
sounds
• xerox
10

PHONOLOGY
• Phonology is the study of the sound system of a language.
• A language's sound system is made up of a set of phonemes which are
used according to phonological rules.
• Phonology describes sound contrasts which create differences in meaning
within a language.
• Phonological systems are made up of phonemes and each language has
its own phonological system.
• This means that the study of phonology is language-specific.
11

• A phoneme is the smallest unit of meaningful sound.
• Phonemes are the basic phonological units and form the
building blocks of speech sounds.
• Phonemes are single sounds represented by a single
written symbol.
12

• Phonotactics is the study of the rules governing the
possible phoneme sequences in a language
13

PHONOLOGICAL RULES
• Phonological rules are related to the spoken or written
principles which control the changes of sounds during
speech
14

• These describe the process of articulation (how a speaker
produces speech sounds stored in the brain).
• Phonological rules help us understand which sounds
change, what they change to, and where the change
happens.
• Examples of phonological rules can be divided into four
types: assimilation, dissimilation, insertion, and deletion.
15

ASSIMILATION
• Assimilation is the process of changing one feature of a
sound to make it similar to another.
• This rule can be applied to the English plural system:
• The -s can change from voiced to voiceless depending on
whether the preceding consonant is voiced or unvoiced.
16

• So, the English plural -s can be pronounced in different
ways depending on the word it is part of, for example:
• In the word snakes, the letter 's' is pronounced /s/.
• In the word baths, the letter 's' is pronounced /z/.
• In the word dresses, the letter 's' is pronounced /ɪz/.
17

DISSIMILATION
• Dissimilation is the process of changing one feature of a
sound to make it different.
• This type of rule makes two sounds more distinguishable.
It can help non-native speakers to pronounce words.
• The pronunciation of the word chimney [ˈʧɪmni] as
chimley [ˈʧɪmli], with the change of [n] to an [l].
18

INSERTION
• Insertion is the process of adding an extra sound between two
others.
• For example, we usually insert a voiceless stop between a nasal
and a voiceless fricative to make it easier for English speakers
to pronounce a word.
• In the word strength /strɛŋθ/, we add the sound 'k' and it
becomes /strɛŋkθ/.
19

DELETION
• Deletion is the process of not pronouncing a sound (consonant,
vowel, or whole syllable) present in a word or phrase, to make it
easier to say.
• For example:
• In the phrase “you and me” [ju: ənd mi:] it is possible not to say the
sound /d/.
• You and me [ju:ənmi:].
20

• This also occurs in some words:
• /h/ in him [ɪm].
• /f/ in fifth [fɪθ].
21

PHONETIC TRANSCRIPTION
• Phonetic transcription is a written guide to
pronouncing specific words.
• Typical transcriptions feature the words people say
exactly, including thinking words and sounds such as
“um,” “like,” “uh,” or “hmm.”
22

• Phonetic transcription and traditional transcription use different
languages.
• These languages have phonetic symbols, each representing a consonant
or vowel sound.
• Phonetic transcription can transcribe any language in the world.
• Phonetic transcription features symbols from the International Phonetic
Alphabet (IPA).
• The IPA is the most widely used and recognized system for phonetic
transcription.
23

• Some words, like “dress,” look very similar to their English
spellings when spelled out using the IPA: drɛs.
• Other words, like the actual word “other,” look quite
different: ˈʌð ər.
• These differences occur because the Latin alphabet and
the IPA have different systems and ways of transcribing
audio material.
24

EXAMPLE
25

WHY IS PHONETIC TRANSCRIPTION IMPORTANT?
• All languages have differences in accents and dialects,
country to country, region to region, and even town to
town.
• Professional voice actors should be able to adjust their
default accents to those needed for whatever job they’re
doing.
• This may be tricky, however, which is where phonetic
transcription comes in handy.
26

OPTIMALITY THEORY
• In a traditional phonological derivation, we are given an underlying
lexical form and a surface form.
• Surface forms of words are those found in natural language text.
• The corresponding lexical form of a surface form is the lemma
followed by grammatical information (for example the part of
speech, gender and number).
• In English "give, gives, giving, gave and given" are surface forms of
the verb give.
• The lexical form would be "give", verb.
27

• The phonological system then consists of one component:
a sequence of rules which map the underlying form to the
surface form.
• Optimality Theory (OT) (Prince and Smolensky, 1993)
offers an alternative way of viewing phonological
derivation, based on two functions (GEN and EVAL) and a
set of ranked violable constraints (CON).
28

• Given an underlying form, the GEN function produces all imaginable surface
forms, even those which couldn’t possibly be a legal surface form for the input.
• The EVAL function then applies each constraint in CON to these surface forms in
order of constraint rank.
• The surface form which best meets the constraints is chosen.
29

MACHINE LEARNING OF PHONOLOGICAL RULES
• The task of a machine learning system is to automatically
induce a model for some domain, given some data from
the domain.
• Thus, a system to learn phonological rules would be given
at least a set of (surface forms of) words to induce from.
• A supervised algorithm is one which is given the correct
answers for some of this data, using these answers to
induce a model which can generalize to new data it hasn’t
seen before.
30

• An unsupervised algorithm does this purely from the data.
While unsupervised algorithms don’t get to see the
correct labels for the classifications, they can be given
hints about the nature of the rules or models they should
be forming. For example, the knowledge that the models
will be in the form of automata is itself a kind of hint. Such
hints are called a learning bias.
31

• Ellison (1992) showed that concepts like the consonant and vowel distinction, the syllable structure of a language, and harmony relationships could
be learned by a system based on choosing the model from the set of potential models which is the simplest. Simplicity can be measured by
choosing the model with the minimum coding length, or the highest probability.
• Daelemans et al. (1994) used the Instance-Based Generalization algorithm (Aha et al., 1991) to learn stress rule for Dutch; the algorithm is a
supervised one which is given a number of words together with their stress patterns, and which induces
• generalizations about the mapping from the sequences of light and heavy syllable type in the word to the stress pattern(A syllable is a part of a
word that contains a single vowel sound and that is pronounced as a unit).
• Johnson (1984) gives one of the first computational algorithms for
• phonological rule induction. His algorithm works for rules of the form b-->a/C, where C is the feature matrix of the segments around a. Johnson’s
algorithm sets up a system of constraint equations which C must satisfy, by considering
• both the positive contexts
32

TEXT BOOK
33

Phonology.pptx

More Related Content

Similar to Phonology.pptx (20)

Recently uploaded (20)

Phonology.pptx