The document discusses natural language processing (NLP), its applications, and issues. It describes how NLP involves analyzing input text, converting it to a machine-readable format, and processing speech grammar and meaning. Applications include machine translation, personal assistants, spell checkers, chatbots, and spam filters. NLP faces challenges from language being ambiguous, imprecise, incomplete, and inaccurate. It then outlines the steps in NLP: input, segmentation, syntactic analysis, semantic analysis, pragmatic analysis, and parsing.
2. Natural Language Processing
- Analysis of input of text or speech form of human language
- Conversion of analysis output to the machine readable format
- Involves processing of speech grammar and meaning
Applications of NLP:
- Machine translation
- Personal assistants
- Auto spell check
- Chatbots
- Spam filters
3. Natural Language Processing (Issues)
1. Ambiguous
- Not clear in meaning
- In a sentence, word wise meaning analysis differs from the whole clause
analysis
- For example: He is beating around the bush.
- Direct meaning : He beats near to the bush
- Actual meaning : He is wasting time
4. Natural Language Processing (Issues)
2. Imprecise
- Not exact
- Human language makes hypothetical assumptions of understanding even
without providing exact facts.
- Such assumption depends on the situations
- For example : I am waiting for a long time.
- Situation 1 (If waiting for a bus) - assumes to be for hours
- Situation 2 (If waiting to return village) - assumption may be years
5. Natural Language Processing (Issues)
3. Incomplete
- Natural language make use of incomplete structures
- Humans understands the structure through situation analysis
- For example : Go there
- Situation 1 (Teacher asking student) - Assumes to go to next class
- Situation 2 (Friends during trek) - Assumes to go to the next destination
6. Natural Language Processing (Issues)
4. Inaccurate
- Conflict in spelling and pronunciation
- A same word may be spelled differently by people of different accent
- For example : At the beginning, Google Assistant do not understand
English spoken by Nepalese people due to the difference in tone and
accent.
7. Steps in Natural Language Processing
1. Input Source:
- Input includes text or speech
- Applies GIGO principle
- Quality of output depends on quality of input
- The input source should be similar to that of the expected system
8. Steps in Natural Language Processing
2. Segmentation
- Process of extracting a smallest possible fragment from the given input
- Each fragment is known as frame.
- The scope of frame depends on the scope of application
- For eg: In character recognition (frame represents each letter)
- In sentiment analysis (frame represents words or sentence)
9. Steps in Natural Language Processing
3. Syntactic Analysis
- Checks the grammar or syntax of the language
- Represents the word level analysis
- The input should not violate the rules of the language for how words may
be combined
- For example : “He the go the to school” - It will be rejected.
10. Steps in Natural Language Processing
4. Semantic Analysis
- Derives the actual meaning of the context
- Regarded as sentence level analysis
- Determines possible meanings of a sentence
- The derivation is based on the task domain.
- Each structure created from the syntactic analyzer is mapped with the
objects in task domain
- The structure with no mapping possible are rejected
- For eg: “Colorless green ideas …” will be rejected as colorless as well as
green makes no sense
11. Steps in Natural Language Processing
5. Pragmatic Analysis
- Regarded as context level analysis
- Involves deriving meaning that depends upon facts about the real world
- Example : Reference of pronouns, Meaning of a set of propositions
12. Parsing
- Parsing is the process of building syntactic analyzer to analyze the
grammar of the input
- In generative transformational grammar theory, phrase structure rules
illustrate mathematically our knowledge on how the basic units of a
sentence are assembled.
14. Parsing (Providing Necessary Data)
The rules may be defined as:
- <Det> a | an | the
- <N> cow | dog | cat | lion | tiger
- <V> ate | chased | got
15. Parsing (Phase Structure Tree)
S = The dog chased the cat
S
NP VP
Det N V NP
The dog chased Det N
The cat
16. Machine Translation
- Translation of text from one natural language to another.
- Difficult because it requires in-depth understanding of the text
- First approach (analyze source language text into knowledge
representation and then generate sentences in the target language from
that representation)
- Problem : creating complete knowledge representation of everything,
parsing into that representation and generating sentence from that
representation
- Other approach - Transfer model (uses database of translation rules)
17. Machine Vision
- Approach to make machine to visualize the objects
- Main goal is to create a model of real world from images
- Mostly used in robotics and object detection and recognition