Semantic Parsing

Tushar B. Kute,
http://tusharkute.com
Semantic Parsing

• Semantic parsing, a crucial task in Natural

Language Processing (NLP), delves deeper than
syntax parsing by aiming to extract the meaning
behind a sentence.
• It goes beyond understanding the grammatical
structure (syntax) and focuses on converting
natural language sentences into a formal
representation that captures their semantics.
Semantic Parsing

• Core Objective:
– Bridging the Gap: Semantic parsing bridges
the gap between human language and
machine understanding by translating natural
language into a machine-readable format
that expresses the intended meaning.
Types of Meaning Representations

• Logical Forms: These represent the meaning of a

sentence in a logical format, often using predicate
logic or lambda calculus. They capture the
relationships between entities and actions described in
the sentence.
• Abstract Meaning Representations (AMRs): This
graphical format represents the meaning of a sentence
as a network of interconnected concepts and relations.
• Execution Trees: These represent the meaning of a
sentence as a sequence of instructions or actions to be
performed by a machine.
Word Sense

• In Natural Language Processing (NLP), word

sense refers to the specific meaning a word
conveys within a particular context.
• Many words have multiple meanings, and
identifying the intended sense in a given
situation is crucial for accurate language
understanding.
Word Sense

• Understanding Word Ambiguity:

– Polysemy: This refers to words with multiple,
related meanings.
• For example, the word "bat" can refer to a
flying mammal or a wooden club used in sports.
– Homonymy: This refers to words with the same
spelling and pronunciation but unrelated
meanings.
• For example, "bat" can also refer to a piece of
fabric used to blindfold someone.
Wordnet

• WordNet is a large lexical database that groups

words in the English language into sets of
synonyms, called synsets.
• These synsets capture the different senses or
meanings that a word can have.
• It's essentially a vast electronic thesaurus combined
with elements of a dictionary, designed to provide
a deeper understanding of word relationships.
Wordnet : Structure

• Synsets:
– The core of WordNet is the synset, a collection of
synonymous words representing a single
concept.
– For example, the synset {king, monarch,
sovereign} represents the concept of a ruler.
• Semantic Relations:
– WordNet connects synsets through various
semantic relations,
Wordnet : Structure

• Semantic Relations: WordNet connects synsets

through various semantic relations, including:
• Hypernymy/Hyponymy: A hierarchical relationship
where a general term (hypernym) is linked to more
specific terms (hyponyms) that fall under it. For
example, "animal" is a hypernym of "dog" and "cat"
(hyponyms).
• Meronymy/Holonymy: A part-whole relationship
where a part (meronym) is linked to the whole it
belongs to (holonym). For example, "wheel" is a
meronym of "car" (holonym).
Wordnet : Structure

• Antonymy: A relationship between words with

opposite meanings. For example, "happy" is an
antonym of "sad".
• Parts of Speech: WordNet categorizes words into
different parts of speech (nouns, verbs, adjectives,
adverbs).
Wordnet : Why?

• Word Sense Disambiguation (WSD):

– By providing information about different senses
of a word and their relationships, WordNet helps
identify the intended meaning of a word in
context.
• Information Retrieval:
– WordNet can be used to improve the accuracy of
search engines by understanding the semantic
relationships between words in a query and
documents.
Wordnet : Limitations

• Coverage: While extensive, WordNet primarily

focuses on everyday words and might not include
domain-specific terminology.
• Granularity: Some argue that WordNet's synsets
might be too fine-grained, leading to challenges in
accurately WSD tasks.
• Language Focus: The current primary focus is on
English, with limited availability for other
languages.
Word sense disambiguation (WSD)

• Word sense disambiguation (WSD) is a crucial task

in Natural Language Processing (NLP) that deals
with identifying the specific meaning of a word
within a particular context.
• Many words in a language have multiple meanings,
and WSD aims to pinpoint the intended sense
based on the surrounding words and the overall
sentence structure.
Word sense disambiguation (WSD)

• Imagine encountering the sentence "The bank is

on the river." Here, "bank" could refer to the
financial institution or the edge of a body of
water.
• Without WSD, a machine might struggle to
understand the true meaning of the sentence.
• However, with WSD, the machine can identify
the intended sense based on the context (river
suggesting the edge meaning).
WSD: Applications

• Machine Translation: Choosing the correct translation

equivalent based on the intended meaning in the source
language.
• Text Summarization: Identifying the key points of a text
and ensuring the summary accurately reflects the meaning
of the original content.
• Sentiment Analysis: Understanding the sentiment
expressed in a sentence requires considering the specific
sense of the words used.
• Question Answering: Answering a question accurately
might rely on identifying the intended meaning of words
in the question itself and the relevant passages.
WSD: challenges

• Context Dependency: The meaning of a word can be highly

dependent on the specific context in which it appears. For
instance, "bat" can mean a flying mammal or a sports
equipment depending on the context.
• Limited Resources: Training accurate WSD models often
requires vast amounts of annotated data, where each word
instance is labeled with its specific sense in that context.
Creating such data can be expensive and time-consuming.
• Word Ambiguity: Some words have very subtle differences
in meaning that can be challenging for WSD models to
distinguish. For example, the difference between "light" as
in weight and "light" as in illumination.
WSD: Approaches

• Dictionary-based Methods: These methods rely on pre-

defined dictionaries that provide different senses for a
word. The system analyzes the context of the word and
chooses the sense that best fits the surrounding words and
the overall sentence meaning.
• Machine Learning Methods: Supervised learning algorithms
can be trained on large datasets of text annotated with
word senses. These models learn to identify patterns in the
context that indicate the intended meaning of a word.
• Unsupervised Learning Methods: These methods group
words with similar contexts together, assuming that words
appearing in similar contexts likely share the same sense.
WSD: Algorithms

• Lesk Algorithm: This classic approach leverages pre-

defined dictionaries with information about word senses
and their definitions. It compares the overlap between
the definitions of the target word's senses and the
definitions of surrounding words in the sentence. The
sense with the most overlap is chosen as the most likely
meaning.
• Gloss Overlap: Similar to Lesk, this method compares
the words used in the definitions of the target word's
senses with the words in the surrounding context. The
sense with the highest number of matching words is
considered the most likely meaning.
WSD: Algorithms

• Machine Learning Methods:

– Naive Bayes:
– Support Vector Machines
– Neural Networks (RNN)
WSD: Algorithms

• Unsupervised Learning Methods:

– Clustering
– EM Algorithms
Lesk Algorithm

• 1. Input:
– The algorithm takes two main inputs:
• A sentence containing a word with multiple
senses (ambiguous word).
• A dictionary resource that provides
definitions for the different senses of words.
• 2. Identify Ambiguous Word:
– The algorithm first identifies the word in the
sentence that has multiple possible meanings
(the ambiguous word).
Lesk Algorithm

• 3. Extract Dictionary Senses:

– The algorithm retrieves all the different senses
(definitions) listed for the ambiguous word in the
dictionary.
• 4. Analyze Context:
– The algorithm defines a "context window"
around the ambiguous word in the sentence.
This context window typically includes a few
words before and after the ambiguous word.
Lesk Algorithm

• 5. Signature Creation:
– For each sense of the ambiguous word retrieved
from the dictionary:
• The algorithm creates a "signature" which is
essentially a set of words.
– This signature can be formed by including all the
words present in the definition of that particular
sense.
Lesk Algorithm

• 6. Overlap Calculation:
– The algorithm calculates the overlap between the
signature of each sense and the words in the context
window of the sentence.
– This overlap is typically measured as the number of
words that appear in both the signature and the
context window (excluding stop words like "the", "a",
etc.).
Lesk Algorithm

• 7. Disambiguation:
– The algorithm identifies the sense of the ambiguous
word with the highest overlap between its signature
and the context window. This sense is considered the
most likely meaning based on the surrounding
words.
• 8. Output:
– The algorithm outputs the disambiguated word,
which is the ambiguous word along with its most
likely sense according to the analysis.
Lesk Algorithm: Example

• Sentence:
– "I went to the bank to deposit my money."
• Ambiguous Word: "bank"
• Dictionary Senses:
– bank (financial institution)
– bank (edge of a river)
• Context Window: "the bank" (assuming a small
window)
Lesk Algorithm: Example

• Signature Creation:
– bank (financial institution) signature: {financial
institution, money, deposit}
– bank (edge of a river) signature: {river, edge}
• Overlap Calculation:
– Overlap (financial institution): 2 (money, deposit)
– Overlap (edge of a river): 0
• Disambiguation:
– Based on the highest overlap (2), the algorithm
suggests "bank (financial institution)" as the most
likely meaning in this context.
Word Sense Induction

• Word sense induction (WSI), also known as word

discrimination, is a subfield of Natural Language
Processing (NLP) concerned with automatically
identifying the different senses of a word.
• Unlike word sense disambiguation (WSD) which
assumes a pre-defined set of senses (like a
dictionary), WSI aims to discover the possible
senses a word can have based on how it's used in a
corpus of text.
Word Sense Induction

• Objective:
– Unveiling Hidden Meanings: WSI delves into a
corpus of text to uncover the various ways a
word is used and group these usages into
clusters representing distinct senses.
– This allows NLP systems to understand the
broader semantic landscape of a word.
Semantic role labeling

• Semantic role labeling (SRL), also known as shallow

semantic parsing or slot filling, is a crucial task in
Natural Language Processing (NLP) that focuses on
identifying the semantic roles of words within a
sentence.
• It goes beyond understanding the grammatical
structure (syntax) and delves into the meaning of
the sentence by pinpointing the roles various
elements play in relation to the main verb or
predicate.
Semantic role labeling

• Core Objective:
– Unveiling the Meaningful Relationships: SRL aims
to uncover the semantic arguments associated
with a predicate (usually the verb) in a sentence.
– These arguments represent the who, what,
when, where, why, and how of the action or
event described by the verb.
Semantic role labeling: Roles

• Agent: The entity performing the action (e.g., "The baker" in

"The baker baked bread").
• Patient: The entity undergoing the action (e.g., "bread" in
"The baker baked bread").
• Instrument: The tool or means used for the action (e.g.,
"knife" in "He cut the apple with a knife").
• Location: The place where the action occurs (e.g., "park" in
"They played in the park").
• Time: The temporal aspect of the action (e.g., "morning" in
"We woke up in the morning").
• Manner: The way or how the action is performed (e.g.,
"quickly" in "She ran quickly").
Semantic role labeling: Process

• Input: The system takes a sentence as input.

• Part-of-Speech (POS) Tagging: The system identifies the part-
of-speech (POS) tag for each word in the sentence (e.g., noun,
verb, adjective).
• Dependency Parsing: The system creates a dependency parse
tree that shows the grammatical relationships between words.
• Predicate Identification: The system identifies the main verb or
predicate in the sentence.
• Role Assignment: Based on the POS tags, dependency
relationships, and potentially a predefined set of semantic
roles, the system assigns semantic roles (e.g., Agent, Patient)
to words or phrases in the sentence.
Diathesis Alternation

• In linguistics, diathesis alternation, also known as

verb alternation, refers to the phenomenon where
a single verb can be used in different syntactic
frames or with different valency (the number of
arguments a verb takes).
• These variations often lead to slight differences in
meaning.
Diathesis Alternation

• One Verb, Multiple Forms: Diathesis alternation

highlights how a verb can appear in different syntactic
constructions while conveying a similar core meaning.
This can involve changes in:
– Transitivity:
• Whether the verb takes a direct object (transitive)
or not (intransitive).
– Argument Structure: The way the verb's arguments
(e.g., subject, object) are arranged in the sentence.
Diathesis Alternation

• Examples:
– Transitive vs. Intransitive:
• "Fred ate the pizza" (transitive, "ate" takes the
object "pizza")
• "Fred ate" (intransitive, "ate" doesn't have a direct
object)
– Argument Structure Variation:
• "Ann threw the ball to Beth" (double object frame)
• "Ann threw Beth the ball" (prepositional object
frame)
Diathesis Alternation

• While the core meaning of the verb might be similar

across alternations, there can be subtle variations
in emphasis or perspective:
– Transitivity: A transitive construction might
emphasize the object being acted upon, while an
intransitive construction might focus more on
the action itself.
– Argument Structure: The order or phrasing of
arguments can influence the focus or nuance of
the sentence.
Proposition Bank

• PropBank, short for Proposition Bank, is a

corpus of text that's annotated with verbal
propositions and their arguments.
• It's a foundational resource in NLP (Natural
Language Processing) used for tasks like
semantic role labeling and understanding the
meaning of sentences.
Proposition Bank

• What it is:
– Annotated Corpus:
• PropBank is a collection of text data where each
sentence is analyzed and labeled with information
about the verb and its arguments.
– Focus on Propositions: It focuses on identifying the core
meaning or proposition expressed by the verb in the
sentence.
– Argument Roles: It also identifies the arguments (e.g.,
agent, patient) associated with the verb, providing a
deeper understanding of the sentence's meaning
structure.
Proposition Bank : Key Points

• Key Points:
– Origin: Developed by Martha Palmer et al., PropBank is
one of the first resources to provide this level of
semantic annotation for verbs and their arguments.
– Verb-Centric: PropBank primarily focuses on verbs,
unlike other resources like FrameNet that might
consider nouns and other words for semantic analysis.
– Complementary to Syntax: While syntax focuses on the
grammatical structure of a sentence, PropBank delves
into the semantic roles and underlying meaning.
Proposition Bank : Applications

• Semantic Role Labeling (SRL): PropBank serves as a

valuable training resource for SRL models, which aim to
identify the semantic roles of words within a sentence.
• Verb Sense Disambiguation (WSD): By analyzing the
arguments associated with a verb in PropBank, WSD
models can gain insights into the intended meaning of
the verb in a particular context.
• Parsing and Machine Translation: PropBank annotations
can be used to improve the accuracy of parsers and
machine translation systems by providing a deeper
understanding of sentence meaning.
FrameNet

• FrameNet is a large lexical database designed

to capture the semantic relationships between
words in a frame-based approach.
• It's a valuable resource in Natural Language
Processing (NLP) for tasks like word sense
disambiguation and semantic analysis.
FrameNet

• Frame Semantics
– FrameNet is built upon the theory of Frame
Semantics, developed by linguist Charles J.
Fillmore.
– This theory proposes that the meaning of many
words can be best understood in relation to a
semantic frame, which represents a stereotypical
situation, event, or relationship.
FrameNet : Structure

• Frames: The heart of FrameNet is the frame, which is a conceptual

category representing a specific type of situation. Examples of
frames include Buying, Communication, and Possession.
• Frame Elements (FEs): Each frame consists of a set of Frame
Elements (FEs), which are the roles or participants involved in the
situation. For instance, the Buying frame might have FEs like
Buyer, Seller, Goods, and Money.
• Lexical Units (LUs): FrameNet links words (verbs, nouns,
adjectives) to frames through Lexical Units (LUs). An LU
represents a specific sense of a word in relation to a particular
frame. For example, the verb "buy" has an LU associated with the
Buying frame, where it signifies the act of acquiring something.
VerbNet

• VerbNet is a lexical resource designed for verbs in

Natural Language Processing (NLP).
• It focuses on capturing the semantic and syntactic
properties of verbs, making it a valuable tool for
tasks like verb sense disambiguation and
understanding sentence structure.
VerbNet

• Core Function:
– Verb-Centric Analysis: Unlike FrameNet, which
encompasses various word classes, VerbNet
specifically concentrates on verbs.
– Classes and Frames: It organizes verbs into a
hierarchical classification system, with general
verb classes (e.g., communication, creation)
further divided into more specific subclasses
(e.g., saying, building).
VerbNet

• Semantic Roles and Thematic Roles: VerbNet

incorporates both semantic roles (related to meaning)
and thematic roles (related to argument structure) to
describe a verb's behavior.
• Semantic roles might include Agent (who performs the
action) and Patient (what undergoes the action), while
thematic roles could be SUBJ (subject) or OBJ (object).
• Selectional Restrictions: VerbNet also specifies
selectional restrictions, which are limitations on the
types of words that can function as arguments for a verb.
For example, the verb "eat" typically requires a Patient
argument that refers to something edible.
Lexicon

• Lexicon (plural: lexicons, rarely lexica) has a few

different meanings in linguistics and natural
language processing (NLP):
– Vocabulary of a language:
• This is the most general sense of lexicon.
• It refers to the entire collection of words and
phrases that constitute a language.
• This includes all the words you might find in a
dictionary, from common everyday words to
more technical terms.
Lexicon

• Vocabulary of a specific domain or register:

– A lexicon can also refer to the vocabulary used
within a particular field or context.
– For example, the legal lexicon includes terms like
"liability" and "tort," while the medical lexicon
encompasses words like "diagnosis" and
"prognosis."
Lexicon

• Lexicon as a resource:
– In NLP, lexicon can sometimes refer to a specific
resource or database that contains information
about words.
– This might include information about a word's
definition, part of speech, synonyms, antonyms, and
other relevant details.
– Examples of lexicon resources include dictionaries,
thesauri, and wordnets (like WordNet or FrameNet).
Lexicon for sentiment, affect and connotation

• Sentiment: A general opinion or feeling

expressed towards something, often
categorized as positive, negative, or neutral.
• Lexicon: Includes words that directly express
emotions or evaluations. These can be broad
(good, bad, happy, sad) or more specific (joyful,
frustrated, disappointed).
Lexicon for sentiment, affect and connotation

• Affect: A physiological or behavioral response to a

situation or stimulus.
– Lexicon:
• Encompasses words that describe bodily
sensations, physiological changes, or outward
expressions associated with emotions.
• Examples include sweating, trembling,
shouting, crying, etc.
Lexicon for sentiment, affect and connotation

• Connotation:
– Connotation: The implied meaning or association
beyond the literal definition of a word. It can evoke
positive, negative, or neutral feelings.
– Lexicon: Comprises words that carry subtle
emotional baggage or cultural associations in
addition to their literal meaning.
• For instance, "homely" can literally mean
homelike, but also carries a connotation of being
plain or unattractive.
Creating Lexicons

• Creating Affect Lexicons by Human Labeling

– Affect lexicons are valuable resources in Natural
Language Processing (NLP) for tasks like
sentiment analysis and emotion recognition.
– They list words and phrases associated with
specific emotions, allowing NLP systems to
understand the emotional undercurrents in text
data.
Creating Lexicons

• Process:
– Word Selection: The initial step involves selecting a
set of words to be labeled. This can be done by:
• Using existing sentiment lexicons as a starting
point.
• Randomly sampling words from a large corpus.
• Focusing on specific emotion categories of
interest.
Creating Lexicons

• Labeling Interface Design: Develop a user-friendly

interface where human annotators can label each
word with the emotions it evokes. This interface might
include:
– The word itself.
– A list of pre-defined emotion categories (e.g., joy,
sadness, anger, fear, surprise, disgust).
– An option for annotators to indicate if the word
doesn't evoke any specific emotion (neutral).
– Optionally, a scale for intensity (e.g., slightly happy
vs. very happy).
Creating Lexicons

• Human Labeling:
– Recruit annotators with good language skills and
familiarity with emotional expression.
– Train them on the labeling guidelines and ensure
consistency in their ratings.
– Each word might be labeled by multiple
annotators to improve reliability.
Creating Lexicons

• Quality Control:
– Implement measures to ensure the quality of the
labeled data:
• Use a gold standard set of pre-labeled words for
annotators to practice on.
• Monitor inter-annotator agreement (how
consistent the labels are between annotators).
• Address discrepancies in labeling through
adjudication or re-labeling.
Creating Lexicons

• Lexicon Creation:
– Once labeling is complete, compile the data and
create the lexicon. Each entry in the lexicon might
include:
• The word itself.
• A list of emotions associated with the word,
along with their frequency of being assigned by
annotators (if multiple annotators were
involved).
• Optionally, an average intensity score for each
emotion.
Semi-supervised Induction of Affect Lexicons

• While human labeling offers high accuracy for affect

lexicon creation, it can be time-consuming and expensive.
• Semi-supervised induction offers a compelling
alternative, leveraging a limited amount of labeled data
along with unlabeled data to build robust affect lexicons.
• Semi-supervised induction utilizes both labeled and
unlabeled data to learn the emotional associations of
words.
• This approach aims to exploit the rich information
present in unlabeled data while relying on the guidance
provided by labeled data.
Semi-supervised Induction of Affect Lexicons

• Techniques:
• Several techniques can be employed for semi-
supervised affect lexicon induction:
– Label Propagation: This method starts with a small
set of labeled seed words with known emotional
associations.
– The labels are then propagated to unlabeled words
based on their similarity or co-occurrence patterns.
– Words that frequently appear alongside positive
seed words are more likely to be assigned positive
sentiment themselves.
Semi-supervised Induction of Affect Lexicons

• Graph-based Methods:
– Words can be represented as nodes in a graph,
with edges connecting similar words.
• Label propagation algorithms can be applied on this
graph to propagate emotional labels from labeled
seed words to unlabeled words based on their
connections and similarities.
Supervised Learning of Word Sentiment

• Supervised learning is a powerful technique for

automatically learning the sentiment (positive,
negative, or neutral) associated with words.
• This approach utilizes labeled data to train a
model that can then predict the sentiment of
unseen words.
• Here's a deeper look at supervised learning for
word sentiment:
Supervised Learning of Word Sentiment

• Process
– Data Preparation
• Data Collection
• Pre-processing
– Feature Engineering
• Word n-grams
• POS Tagging
• Dictionary features
• Word Embeddings
Supervised Learning of Word Sentiment

– Model Selection and Training

• Machine Learning models
• Neural Network Models
– Evaluation
– Deployment
Lexicons for Sentiment Recognition

• Lexicon-based sentiment recognition is a

fundamental approach in Natural Language
Processing (NLP) for understanding the overall
emotional tone of text data.
• It relies on pre-built dictionaries or lists of words
(lexicons) that are associated with specific
sentiments (positive, negative, or neutral).
Lexicons for Sentiment Recognition

• Steps:
– Lexicon Development:
• Existing sentiment lexicons like SentiWordNet,
NRC Word-Emotion Association Lexicon, or
LIWC can be used directly.
• Alternatively, you can create a custom lexicon
tailored to your specific domain or needs.
• This might involve collecting words relevant to
your domain and manually labeling them with
sentiment.
Lexicons for Sentiment Recognition

• Steps:
– Text Processing:
• The text data you want to analyze (e.g.,
reviews, social media posts) is preprocessed.
• This typically involves cleaning the text by
removing noise, punctuation, and applying
stemming/lemmatization (reducing words to
their base form).
Lexicons for Sentiment Recognition

• Steps:
– Lexicon Matching:
• Each word in the preprocessed text is
compared against the lexicon.
• If a match is found, the sentiment score
associated with that word is retrieved.
Lexicons for Sentiment Recognition

• Steps:
– Sentiment Score Calculation:
• A sentiment score is calculated for the entire text document or
sentence. This can be done in various ways:
– Simple Sum: The sentiment scores of all matched words are
simply added together. Words with stronger sentiment
(positive or negative) will contribute more to the overall
score.
– Weighted Sum: Assign weights to words based on their
intensity (e.g., "very happy" might have a higher weight than
"happy").
– Taking the Most Frequent Sentiment: Assign the sentiment
of the most frequently occurring words in the text as the
overall sentiment.
Lexicons for Sentiment Recognition

• Steps:
– Sentiment Classification:
• Based on the calculated sentiment score, the
text is classified as positive, negative, or
neutral.
• Thresholds can be defined to determine the
sentiment category (e.g., a score above a
certain value might be positive, below a
certain value might be negative, and values in
between might be neutral).
Lexicons for Personality Recognition

• Personality recognition aims to understand a person's

character from their text.
• Lexicons are like emotional dictionaries, listing words
associated with specific traits.
• By analyzing the words someone uses and how often, a
lexicon-based approach can predict aspects of their
personality, like extroversion or agreeableness.
• However, this method might miss sarcasm or context-
dependent meanings, and machine learning models are
often used for more nuanced personality recognition.
Lexicons for Personality Recognition

• Personality recognition aims to understand a person's

character from their text.
• Lexicons are like emotional dictionaries, listing words
associated with specific traits.
• By analyzing the words someone uses and how often, a
lexicon-based approach can predict aspects of their
personality, like extroversion or agreeableness.
• However, this method might miss sarcasm or context-
dependent meanings, and machine learning models are
often used for more nuanced personality recognition.
Lexicons for Personality Recognition

• A lexicon for affect recognition is a like a giant

dictionary of emotions! It lists words and phrases linked
to specific feelings (joy, anger, sadness, etc.).
• By analyzing these lexicons, NLP systems can
understand the emotional undercurrents in text data,
like social media posts or reviews.
• These lexicons can be built manually by humans labeling
words, or through automated techniques that leverage
unlabeled data.
• They are a valuable tool for tasks like sentiment analysis
and emotion recognition in NLP.
Entity-Centric Affect

• Lexicon-based methods in entity-centric affect analysis

focus on understanding emotions associated with specific
entities (people, organizations, etc.) in text data.
– Traditional Lexicons: These methods leverage existing
sentiment lexicons like SentiWordNet, but with a twist.
• Instead of analyzing overall sentiment, they focus on
emotions linked to particular entities mentioned in
the text.
– Entity-Centric Analysis: The core idea is to analyze how
different phrases and words used around an entity
contribute to its overall emotional portrayal.
Connotation Frames

• Connotation Frames is a relatively recent concept

in NLP (Natural Language Processing) that focuses
on understanding the subtle, implied meanings
associated with a predicate (usually a verb) beyond
its literal definition.
• It delves deeper than sentiment analysis, aiming to
capture the nuances of how an action is framed and
the impact it has on the entities involved.
Connotation Frames

• It represents the rich layer of meaning conveyed by a verb

beyond its basic action. This includes:
• Writer's perspective: How the writer portrays the "actor"
(who performs the action) and the "patient" (who
undergoes the action).
• Entities' perspective: Potential emotional states or
reactions of the entities involved.
• Effect: The outcome or consequence of the action.
• Value: The implied importance or significance of the action
or the entities involved.
• Mental state: The emotional state or intention of the actor.
Summary

• Semantic parsing takes a sentence and breaks it down not just

grammatically, but also into its meaning.
• It aims to understand the roles that words play (who, what,
when, where, why, how) and the relationships between them.
• This allows NLP systems to move beyond surface meaning and
grasp the deeper intent of a sentence.
• It's crucial for tasks like question answering, where the system
needs to understand the meaning of the question and the
relationships between entities to provide an accurate answer.
• Semantic parsing can be rule-based, machine learning based,
or rely on dependency parsing to achieve this deeper
understanding.
 Thank you
This presentation is created using LibreOffice Impress 7.4.1.2, can be used freely as per GNU General Public License

@mITuSkillologies @mitu_group @mitu-skillologies @MITUSkillologies

Web Resources
https://mitu.co.in
@mituskillologies http://tusharkute.com @mituskillologies

[email protected]
[email protected]