Module 5: Pragmatics

- Introduction to pragmatics analysis: Key aspects and challenges

- Discourse and reference resolution

- Coreference resolution and anaphora resolution

- Constraints in meaning interpretation:

- Syntactic constraints

- Semantic constraints

- Contextual inference

- Information retrieval and question answering systems:

- Open-domain QA

- Closed-domain QA

- Text summarization and generation:

- Extractive summarization

- Abstractive summarization

- Dialogue systems and conversational AI:

- Dialogue systems

- Chatbots

- Task-oriented assistants

Question Marks

Define discourse reference resolution. 2

- Discourse reference resolution identifies what words or expressions refer to in a larger context or
conversation.
- It connects pronouns or noun phrases (like “he” or “this”) to the correct entity.
- It ensures the meaning of a sentence is coherent in the entire text.
- Example: In "Ravi went to the shop. He bought bread", “he” refers to Ravi.
Question Marks

What is a reference phenomenon? 2

- Reference phenomenon involves how words like “he,” “she,” “it,” or “this” point to actual people or
things in context.
- It helps identify what the speaker or writer is talking about.
- It includes coreference, anaphora, and deixis.
- Example: “It is raining.” – “It” refers to the weather or current situation.

Explain syntactic constraints. 2

- Syntactic constraints are grammar rules that affect sentence structure and meaning.
- They determine which sentence patterns are acceptable.
- For example, “The cat sat on the mat” follows subject-verb-object structure.
- They help ensure clarity and avoid ambiguity in sentence formation.

What are semantic constraints? 2

- Semantic constraints are limits based on the meaning of words and their combinations.
- They prevent illogical sentences, like “The table sang a song.”
- They ensure that words used together make sense.
- These constraints guide correct word usage based on meaning.

What is information retrieval? 2

- Information retrieval (IR) is the process of finding relevant documents or data based on a user query.
- Examples include search engines like Google retrieving web pages.
- It involves indexing, ranking, and matching content to user intent.
- IR helps in locating information quickly from large datasets.

Define a question answering system. 2

- A Question Answering (QA) system answers natural language questions automatically.

- It finds accurate responses using documents, databases, or the internet.
- QA systems are of two types: open-domain (any topic) and closed-domain (specific field).
- Example: Siri, Alexa, and Google Assistant use QA systems.

What is text summarization? 2

- Text summarization reduces a long text into a concise version without losing key information.
- It can be extractive (pulling sentences from the original) or abstractive (rephrasing content).
- It helps users understand large content quickly.
- Common in news, legal, and academic documents.
Question Marks

Question Marks

What is the difference between syntactic and semantic constraints? 5

- Syntactic constraints deal with sentence structure and grammar rules; they define how words must be
ordered for a sentence to be grammatically correct.
- Semantic constraints focus on meaning; they ensure that the word combinations make logical sense.
- Example of syntactic: “He go store” is wrong due to structure.
- Example of semantic: “The rock sang” is grammatically fine but semantically incorrect.
- Syntactic constraints are language-specific, while semantic constraints are meaning-specific.
- Syntax ensures the sentence is formally correct; semantics ensures it conveys real-world logic.
- Both works together to ensure effective communication.
- Violating either may cause confusion or misinterpretation in NLP tasks.

Explain the significance of discourse reference resolution. 5

- It connects pronouns and phrases to their actual referents in text (e.g., “he,” “she,” “this”).
- Maintains coherence across multiple sentences or conversation turns.
- Helps machines understand who or what is being discussed.
- Supports accurate summarization, translation, and chatbot understanding.
- Essential for anaphora resolution, coreference chains, and pronoun linking.
- Example: In “Sara called Emma. She was crying,” resolving “she” is critical.
- Makes dialogue systems and QA systems more human-like.
- Enhances user experience by avoiding ambiguity in responses.

Describe how text summarization is used in NLP. 5

- Text summarization shortens long documents into brief, meaningful summaries.

- Used in applications like news summarization, academic abstracts, and legal case briefs.
- Two main types: Extractive (selects key sentences) and Abstractive (generates new concise sentences).
- Helps users grasp key points without reading entire texts.
- Improves efficiency in search engines, QA systems, and chatbots.
- Summarization requires understanding context, importance, and redundancy.
- NLP models use techniques like TF-IDF, BERT, and Transformers for summarization.
- Useful in real-time scenarios like news apps, legal tools, and research assistants.
Question Marks

Implement a simple question answering system. 5

- QA systems take user queries and return accurate, relevant answers.

- They are of two types: Closed-domain (e.g., medical QA) and Open-domain (e.g., Google search).
- Use techniques like keyword matching, named entity recognition, and semantic similarity.
- A simple system can use a predefined document and extract answers from it.
- Python libraries like spaCy, NLTK, or HuggingFace Transformers can be used.
- Implementation steps: preprocess text → identify question type → extract answer span.
- Models like BERT can be fine-tuned for answer prediction.
- QA systems are used in virtual assistants, chatbots, and customer service bots.

Apply text summarization techniques to a document. 5

- First, decide between extractive and abstractive summarization.

- Extractive method selects important sentences using scores (e.g., TF-IDF, TextRank).
- Abstractive summarization rewrites content using neural models like T5 or GPT.
- Use libraries like Sumy, spaCy, HuggingFace, or BART.
- Start with preprocessing: remove stopwords, tokenize, and segment sentences.
- Identify the main topic or focus sentence(s).
- Apply summarization and evaluate by comparing with a human-written summary.
- Ensure the final summary is coherent, non-redundant, and contextually accurate.
 Question Marks

Use discourse reference resolution in a given text. 5

- Identify pronouns, named entities, and noun phrases that may refer to the same entity.
- Determine coreference chains (e.g., “Ravi” → “he” → “the boy”).
- Tools like spaCy, NeuralCoref, or AllenNLP help resolve references.
- Context matters: sentence position and narrative flow influence resolution.
- Use entity recognition + context windows to link references accurately.
- Helps in summarization, QA, and machine translation tasks.
- Example: “The dog chased the cat. It escaped.” → “It” = “the cat”.
- Accurate resolution improves coherence and reader understanding in NLP outputs.

Develop a basic dialogue system or chatbot. 5

- A dialogue system or chatbot interacts with users via natural language.

- Basic components: input understanding, dialogue management, and response generation.
- Use rule-based, retrieval-based, or generative models.
- Python tools: Dialogflow, ChatterBot, Rasa, or transformer-based models.
- Add intent recognition (e.g., “Book ticket”) and entity extraction (e.g., “Mumbai”).
- Maintain context to handle follow-up queries.
- Train using sample dialogues, FAQs, or conversation datasets.
- Chatbots are widely used in customer support, booking, and educational tools.

Last-minute revision cheat sheet for Module 5: Pragmatics in NLP

NLP Module 5 – Pragmatics Cheat Sheet

Pragmatics Overview

• Pragmatics: Study of meaning in context—how speakers and listeners interpret language beyond literal
words.
• Challenges: Ambiguity, context sensitivity, indirect references, and speaker intention.

Discourse & Reference Resolution

• Discourse: Continuous stretch of language (like paragraphs or dialogue).

• Reference Resolution: Linking pronouns/phrases to correct entities.

• Coreference: Identifying multiple expressions referring to the same entity.

Example: “Ravi went home. He was tired.” → He = Ravi

• Anaphora Resolution: Resolving backward references (pronouns → nouns).

• Tools: spaCy, NeuralCoref, AllenNLP.

Constraints in Meaning

• Syntactic Constraints: Based on sentence structure (grammar).

Example: “She go market” → syntactically invalid.

• Semantic Constraints: Based on logical meaning.

Example: “The rock sang a song” → semantically invalid.

Contextual Inference

• Understanding meaning requires context: previous sentences, speaker intent, or world knowledge.

• Important in dialogue systems, QA, and chatbots.

Question Answering (QA) Systems

• Open-domain QA: Any topic (e.g., Google, ChatGPT).

• Closed-domain QA: Specific field (e.g., medical QA).

• Conversational QA: Context-aware, multi-turn QA.

Text Summarization

• Extractive: Selects key sentences from original text.

• Abstractive: Generates new sentences with key ideas.

• Tools: BART, T5, HuggingFace Transformers.

Dialogue Systems & Conversational AI

• Dialogue Systems: Manage multi-turn conversations.

• Chatbots: Rule-based or generative AI (e.g., Rasa, GPT).

• Task-Oriented Assistants: Solve specific problems (e.g., book tickets).

Quick Tips:

• Coreference ≠ Anaphora (coref can be backward or forward).

• Use context to disambiguate meaning.

• Syntactic = grammar; Semantic = logic.

• Extractive = original text; Abstractive = reworded.

• QA = Retrieval + Answer Extraction.

• Chatbot = Input → Intent → Response.