Feature Systems and Augmented Grammars

Introduction

Natural languages have agreement restrictions between words and phrases, such as number
agreement, gender agreement, and subject-verb agreement. Feature systems and
augmented grammars help enforce these agreements in linguistic structures.

Feature Systems

 Definition: A method to represent and manipulate linguistic features (e.g., tense,

number, gender, case, syntactic role) in a sentence.
 Purpose: Ensures grammatical correctness and maintains agreement in sentence
construction.
 Examples of Feature Agreement:
1. Number Agreement:
 ✅ A man (Correct: Singular)
 ❌ A men (Incorrect: Mismatch)
2. Subject-Verb Agreement:
 ✅ He writes vs. ❌ He write
3. Gender Agreement:
 ✅ She is intelligent vs. ❌ She is intelligent (incorrect if gender does not
match pronoun).
4. Case Agreement:
 ✅ She gave him a book (Correct: Object in objective case).

Augmented Grammars

 Definition: An extension of Context-Free Grammars (CFGs) that allows feature

structures (mappings of attributes to values) for improved linguistic analysis.
 Example Rule:
o NP → ART N (Only valid if NUMBER(ART) = NUMBER(N))
o Meaning: A noun phrase must contain an article and noun that agree in
number (singular/plural).
 Feature-Based Constraints:
o Instead of duplicating rules for singular and plural forms, augmented
grammars use feature structures to enforce constraints efficiently.

Advantages of Augmented Grammars

1. Prevents Rule Duplication: Instead of writing separate rules for singular and plural
forms, a single rule with feature constraints can handle both cases.
 2. Better Precision in Sentence Formation: Enforces proper grammatical structure
dynamically.
3. Efficient Parsing and Language Processing: Used in Natural Language
Processing (NLP), Machine Translation, Speech Recognition, and Grammar
Checking.
4. Handles Complex Linguistic Phenomena: Can incorporate agreement conditions
that traditional CFGs cannot manage.

Comparison: Traditional CFGs vs. Augmented Grammars

Aspect Traditional CFG Augmented Grammar

Feature Handling Cannot enforce agreement Uses feature structures
Grammar Size Large (due to rule duplication) Compact (single rule with features)
Parsing Efficiency Less efficient More efficient
Linguistic Accuracy Lower Higher

Conclusion

Feature systems and augmented grammars improve linguistic accuracy by enforcing

agreement constraints within sentences. By incorporating feature structures into
grammars, they enhance parsing efficiency, NLP applications, and automated language
processing

Morphological Analysis and the Lexicon – Detailed Notes (Exam-Oriented)

1. Introduction to the Lexicon

 The lexicon is a structured collection of words used in a language.

 It contains phonetic, grammatical, and semantic information about words.
 Each word entry in a lexicon includes:
o Spelling (orthographic representation)
o Pronunciation (phonetic representation)
o Part of speech (POS) (noun, verb, adjective, etc.)
o Morphological features (tense, number, gender, case, derivation, inflections)
o Meaning (semantics) (definitions, synonyms, antonyms)

Example of Lexical Entries

Consider the word “run” in a lexicon:

1. Noun → The marathon run was exhausting.

2. Verb (Base form) → I run every morning.
3. Verb (Past form – ran) → She ran fast yesterday.
4. Verb (Participle – running) → She is running now.
This demonstrates how a single word can have multiple meanings and forms, all stored in
the lexicon.

2. Morphological Analysis

 Morphology is the study of word formation and the internal structure of words.
 Morphological analysis breaks down words into morphemes, the smallest units of meaning.

Types of Morphemes

1. Free morphemes: Can stand alone as words.

o Examples: friend, cat, walk
2. Bound morphemes: Cannot stand alone and must attach to other morphemes.
o Examples: -s (plural in cats), -ed (past tense in walked)

Morpheme Example Analysis

 Unhappiness → un (prefix) + happy (root) + -ness (suffix)

 Friendly → friend (root) + -ly (suffix that changes a noun to an adjective)

3. Role of Morphological Analysis in NLP

Key Applications in NLP

 Lemmatization: Reducing a word to its base form.

o Example: running → run, better → good
 Stemming: Cutting off affixes to find the word root.
o Example: playing → play, cats → cat
 Part-of-Speech (POS) Tagging: Identifying a word's grammatical category using morphology.
o Example: "running" can be a verb (She is running) or a noun (Running is fun).
 Inflectional Morphology: Understanding word variations based on tense, number, or
gender.
o Example: walk → walked (past), cat → cats (plural)
 Derivational Morphology: Creating new words by adding prefixes or suffixes.
o Example: happy → happiness (noun), act → actor (noun from verb)

4. The Lexicon’s Role in Morphological Analysis

 The lexicon stores word forms and rules that help NLP systems analyze and generate
language.
 It helps determine:
o Base form of a word (lemma).
o Inflected forms based on grammar rules.
o Multiple meanings of ambiguous words.
Example: The Word “Saw” in a Lexicon

1. Noun → The saw was broken. (Tool)

2. Verb (Base form) → Jack wanted me to saw the board. (Action)
3. Verb (Past tense of "see") → I saw Jack eat the pizza.

The lexicon helps resolve ambiguity and assigns correct meanings based on context.

5. Importance of Morphological Analysis in NLP

1. Morphological analysis plays a crucial role in Natural Language Processing (NLP)

as it helps machines understand and process human language more effectively.
2. It enables various linguistic tasks by breaking words into their smallest meaningful
units (morphemes) and analyzing their structure, forms, and meanings.

6. Conclusion

Morphological analysis and the lexicon are essential for understanding language structure
and enabling effective NLP applications. A well-defined lexicon ensures accurate word
formation, interpretation, and language generation in computational linguistics.

Introduction

Augmented Transition Networks (ATNs) are an advanced type of transition network

used in Natural Language Processing (NLP) to analyze and generate sentences. They
extend Recursive Transition Networks (RTNs) by adding registers (storage variables) to
keep track of information, making them more powerful for parsing, semantic analysis, and
language generation.

4o

Structure of ATNs

1. Nodes and Arcs: ATNs consist of nodes (states) and arcs (transitions), which define
how words are processed.
2. Registers: These are special storage variables used to store values while traversing
the network.
3. Transition Rules: Conditions that define when and how the network moves from
one state to another based on input words.
Features of ATNs

1. Registers for Data Storage:

o Each network has its own set of registers to store temporary values.
o When a new network is called, a new set of registers is created to maintain
context.
2. Special Variable ("*") Usage:
o The value obtained from an arc transition can be stored in a special variable
("*") for later use.
o This helps in semantic and syntactic processing.
3. Operations on Registers:
o Assignments: Values can be assigned dynamically during parsing.
o Tests: Conditions can be checked to control flow and correctness.
o Agreement Checks: Ensures grammatical correctness (e.g., subject-verb
agreement).

Applications of ATNs

✅ Parsing: Helps in sentence structure analysis.

✅ Semantic Analysis: Extracts meaning from sentences.
✅ Language Generation: Helps in constructing grammatically correct sentences.

Advantages of ATNs

✔ Handles complex sentence structures efficiently.

✔ Supports context-aware processing using registers.
✔ Allows recursive and hierarchical parsing.

Disadvantages of ATNs

✖ Computationally expensive due to recursion.

✖ Requires detailed rule-based implementation.
✖ Difficult to debug and interpret.

Here's a more detailed tabular comparison of Top-Down and Bottom-Up Parsers in NLP:

Feature Top-Down Parsing Bottom-Up Parsing

Begins with the start symbol and Starts from the input string and
Definition expands derivations using applies reductions to reach the start
production rules. symbol.
Direction of
Left-to-right derivation (expanding) Right-to-left derivation (reducing)
Parsing
Reduces terminals and non-
Approach Expands non-terminals recursively
terminals
Derivation Type Leftmost derivation Rightmost derivation in reverse
 Feature Top-Down Parsing Bottom-Up Parsing
Recursive Descent Parser, LL(1) Shift-Reduce Parser, LR(0),
Examples
Parser SLR(1), LALR(1), LR(1) Parsers
Working Tries to match input with a Tries to reduce the input into a valid
Mechanism predefined grammar structure grammatical form
More efficient as it avoids
Efficiency Less efficient due to backtracking
backtracking
Handling Left Cannot handle left recursion without
Can naturally handle left recursion
Recursion transformation
Handling Difficult to handle ambiguous More effective in handling
Ambiguity grammars ambiguous grammars
Simple implementation but More complex implementation but
Complexity
inefficient for large grammars efficient for large grammars
Parsing Table Yes, in predictive parsing (LL Yes, in LR Parsing
Used? Parsing)
Uses an explicit or implicit
Stack Usage Uses an explicit stack for parsing
recursion stack
Lookahead Requires lookahead to determine the
Uses lookahead to decide reduction
Mechanism correct production
May require backtracking (except
Backtracking Does not require backtracking
for LL parsers)
Grammar Works with LL(k) grammars (subset Works with a larger set of
Restrictions of context-free grammars) grammars (LR grammars)
Used in rule-based NLP systems,
Used in statistical NLP, machine
Use Cases in NLP syntax checking, and simple
translation, and data-driven parsing
grammar parsing
Constructs a valid parse tree by
Parses a sentence by expanding
Example merging words into phrases and
possible structures based on
Grammar Parsing reducing them
predefined rules

Bottom up parsing

Initialize

o Start with an empty stack and the input sentence.

2. Shift Operation
o Read a word from the input and push it onto the stack.
3. Reduce Operation
o If the top of the stack matches the right-hand side of a grammar rule, replace it
with the left-hand side.
4. Repeat Steps 2 & 3
o Continue shifting and reducing until the input is consumed and the stack
contains only the start symbol.
5. Finalize
o If the stack contains only the start symbol (S), parsing is successful; otherwise,
it fails.
o