Natural Language Processing (AD-802 (A)) Unit 2 Notes

How unit 2 is examined

This unit covers the levels of linguistic analysis (sounds, words, sentences, meaning, use) and the word-level tools built on them; no question has been asked recently, so every topic is short but complete.

Linguistic essentials: Phonetics and Phonology

Not asked since 2022

Definition. Phonetics studies the physical production and perception of speech sounds, while phonology studies how a language organises those sounds into patterns and meaningful units called phonemes.

Key points.

Phonetics has three branches: articulatory (how sounds are produced), acoustic (their waves) and auditory (how they are heard).
A phoneme is the smallest sound unit that changes meaning, for example /p/ and /b/ separate "pat" and "bat".
Allophones are variants of one phoneme that do not change meaning.
NLP uses these ideas in speech recognition and text-to-speech.

Morphology: Morphens

Definition. Morphology is the study of the internal structure of words, and a morpheme is the smallest meaningful unit of a language.

Key points.

A free morpheme can stand alone as a word ("book"), whereas a bound morpheme must attach to another ("-s", "un-").
Bound morphemes are either derivational (they create a new word or class, as in "happy" to "happiness") or inflectional (they add grammatical information such as plural or tense).
"Unhappiness" has three morphemes: un- + happy + -ness.
A root carries the core meaning and affixes (prefix, suffix, infix, circumfix) modify it.

Syntax

Definition. Syntax is the study of how words combine into phrases and grammatical sentences according to the rules of a language.

Key points.

Syntax works with word order and grammatical categories such as noun, verb and phrase.
Sentence structure is shown by a parse tree built from grammar rules such as S to NP VP.
A sentence can be grammatical yet meaningless ("Colourless green ideas sleep furiously").
Syntactic analysis, or parsing, is the NLP task that recovers this structure.

Semantics

Definition. Semantics is the study of the literal meaning of words, phrases and sentences, independent of the context of use.

Key points.

Lexical semantics deals with word meaning and sense relations such as synonymy, antonymy, hyponymy and polysemy.
Compositional semantics builds the meaning of a sentence from the meanings of its parts and their structure.
Ambiguity is the central problem: "bank" may mean a river edge or a financial institution.
NLP represents meaning using logic and performs word sense disambiguation.

Pragmatics

Definition. Pragmatics is the study of how context, speaker intention and shared knowledge shape the meaning of an utterance beyond its literal content.

Key points.

"Can you pass the salt?" is literally a question but pragmatically a request (speech act).
Deixis words such as "I", "here" and "now" depend on the speaker and situation.
Implicature is meaning implied but not stated, guided by Grice's maxims of quantity, quality, relation and manner.
Semantics gives literal meaning; pragmatics adds intended meaning.

Semiotics

Definition. Semiotics is the study of signs and symbols and of how they create and communicate meaning.

Key points.

A sign joins a signifier (the form, such as a word or image) with a signified (the concept it stands for).
Peirce classified signs as icon (resembles the object), index (points to it, like smoke for fire) and symbol (arbitrary convention, like words).
Language is a system of symbols whose meaning is agreed by convention.
It helps explain meaning in text, emoji and multimodal data.

Discourse Analysis

Definition. Discourse analysis studies language beyond the single sentence, examining how sentences connect to form coherent text or conversation.

Key points.

Cohesion is the surface linking of sentences using pronouns, conjunctions and repetition.
Coherence is the logical, meaningful flow that makes a text make sense as a whole.
Anaphora resolution links a pronoun to its earlier referent, as in "Ravi left because he was tired".
NLP uses it in summarisation, dialogue systems and coreference resolution.

Psycholinguistics

Definition. Psycholinguistics is the study of the mental processes by which humans acquire, comprehend, produce and store language.

Key points.

Its main areas are language acquisition, comprehension, production and the mental lexicon.
Garden-path sentences ("The horse raced past the barn fell") show that people parse incrementally and may need to reanalyse.
Priming shows that related words are activated together in memory.
Its findings inspire cognitively motivated NLP models.

Corpus Linguistics

Definition. Corpus linguistics studies language by analysing large, real-world collections of text or speech called corpora, using empirical and statistical methods.

Key points.

A corpus is a large, machine-readable, representative body of authentic text, for example the Brown Corpus.
Annotated corpora carry labels such as POS tags or parse trees, and treebanks are the syntactic kind.
Typical outputs are word frequencies, concordances and collocations.
Corpora are the training and test data for statistical NLP.

Word Formation Processes

Definition. Word formation processes are the ways a language creates new words from existing material.

Key points.

Derivation adds prefixes or suffixes to change meaning or class ("teach" to "teacher").
Compounding joins two words ("blackboard"); inflection only changes grammatical form ("walk" to "walked").
Other processes are conversion (noun to verb, "email"), blending ("brunch"), clipping ("phone"), acronyms ("NASA") and back-formation ("edit" from "editor").
Reduplication repeats a word or part of it, common in Hindi ("chhota-chhota").

Morphological analysis

Definition. Morphological analysis breaks a word into its root and affixes and identifies their grammatical features.

Key points.

For "cats" it returns root "cat" + plural suffix -s, so it gives both structure and features.
Stemming crudely chops suffixes ("studies" to "studi") and needs no dictionary.
Lemmatization uses a lexicon and morphology to return the valid base form ("better" to "good").
It reduces vocabulary size and helps search, translation and tagging.

Morphological Finite State Transducers

Definition. A finite state transducer (FST) is a finite automaton with two tapes that maps an input string to an output string, so it can convert between the surface form of a word and its lexical form.

Key points.

Each arc carries a pair input:output, and the machine reads a surface word while writing its analysis.
Example: surface "cats" maps to lexical "cat +N +PL".
The same FST works in both directions, for analysis (parsing) and generation.
Spelling changes such as "fox" + "s" to "foxes" (e-insertion) are handled by orthographic rules compiled as FSTs, which are cascaded with the lexicon.

Last-minute revision

Phonetics is speech sounds; phonology is sound patterns; the phoneme is the smallest meaning-changing sound.
A morpheme is the smallest meaningful unit; free morphemes stand alone, bound ones do not.
Derivational morphemes change word or class; inflectional ones add grammar only.
Syntax is sentence structure; semantics is literal meaning; pragmatics is meaning in context.
Grice's maxims: quantity, quality, relation, manner.
Peirce's signs: icon, index, symbol.
Cohesion is surface linking; coherence is logical flow.
A corpus is a large annotated text collection; a treebank holds parse trees.
Word formation: derivation, compounding, conversion, blending, clipping, acronym, back-formation.
Stemming chops suffixes; lemmatization returns the dictionary form.
An FST maps surface to lexical form ("cats" to "cat +N +PL") in both directions.

Memory hooks

Levels of language, small to big: Sounds, Morphemes, Syntax, Semantics, Pragmatics, Discourse.
Icon looks like it, index points to it, symbol is agreed.
Stem is stupid (chops), lemma is learned (lexicon).
FST has two tapes: surface in, lexical out.

Coverage checklist

Linguistic essentials: Phonetics and Phonology: definition, phoneme, branches (no past questions).
Morphology: Morphens: morpheme types (no past questions).
Syntax: parse structure (no past questions).
Semantics: lexical and compositional meaning (no past questions).
Pragmatics: speech acts, implicature (no past questions).
Semiotics: sign types (no past questions).
Discourse Analysis: cohesion, coherence (no past questions).
Psycholinguistics: language processing in the mind (no past questions).
Corpus Linguistics: corpora and treebanks (no past questions).
Word Formation Processes: derivation, compounding and others (no past questions).
Morphological analysis: root, affixes, stemming vs lemmatization (no past questions).
Morphological Finite State Transducers: two-tape FST (no past questions).

Account

Natural Language Processing (AD-802 (A)) - Unit 2 Short Notes

How unit 2 is examined

Linguistic essentials: Phonetics and Phonology

Morphology: Morphens

Syntax

Semantics

Pragmatics

Semiotics

Discourse Analysis

Psycholinguistics

Corpus Linguistics

Word Formation Processes

Morphological analysis

Morphological Finite State Transducers

Last-minute revision

Memory hooks

Coverage checklist