NLP
Background
Mentioned in the Data Science page, Natural Language Processing (NLP) is a
form of supervised machine learning. NLP essentially reads, deciphers, and
converts the human language to a manner that is understandable to machines.
The NLP code needs to be trained before it is able to do it's job–which is to
parse text and output results upon the user's demand. Examples of NLP are
Siri, spell-check, spam filters, and voice text messaging. Currently, NLP is
very difficult since the human language is very complex, but people see hope
in Deep Learning to solve more complex NLP problems.
There are two very common techniques used in NLP:
1. Syntactic Analysis: analyzes how the natural language follows grammatical
rules and tries to derive meaning from the words.
Example: Part-of-speech tagging identifies the part of speech for each word
2. Semantic Analysis: analyzes the meaning of the group of words and is the
difficult aspect of NLP
Example: Named entity Recognition (NER) determines which subset of words
can be categorized together as one group
Chemdataextractor
An example of a NLP program is Chemdataextractor.(Check
out [Link]/!) This python package is able to parse many
different file types, including html's, xml's, pdf's, etc., and output a variety of
useful chemistry-related data, such as chemical named entities, abbreviation
definitions, and chemical records. It incorporates NLP in it's code, and is split
into 5 main stages: Tokenization, Part-of-Speech Tagging, Named Entity
Recognition, Phrase Parsing, and Information Extraction.
This image was taken from the Chemdataextractor documentation, which can
be accessed here: [Link]/docs/intro
Tokenization
In this step, the program segments the text into sentences or words and
removes certain unneeded characters, like some punctuation. The smaller
pieces of text are then called tokens.
For example with sentence "US20190241449A1 - Process for the removal of
iron and phosphate ions from a chlorinated hydrocarbon waste stream-
Google Patents", the program will print:
Part-of-Speech Tagging
As mentioned above, Part-of-speech tagging labels every word in the text with
their designate part of speech.
For example, with sentence "1H NMR spectra were recorded on a 300 MHz
BRUKER DPX300 spectrometer.", the program will print:
Below are the definitions for each acronym:
Named Entity Recognition
In this step, the program is able to identify specific chemical entities and
where that entity is located. Using the code, it will return a Span, which
includes the text and start and end character number.
Phrase Parsing
The phrase parser takes a long sentence as is able to split it into constituents
with labels, specifically noun phrases, verb phrases, and prepositional
phrases. NP stands for noun phrase, VP stands for verb phrase, and PP stands
for prepositional phrase.
For example, with the phrase "The quick brown fox jumped over the lazy
dog", the program will print out:
Information Extraction
Through use of NLP, informational extraction is the process of automatically
identifying and taking out specific structured information from machine-
readable documents and other forms of texts. One example of information
extraction is the process of emails extracting only the dates from messages for
you to add to your calendar.
ROBOTICS
An industrial robot has the basic parts like arm, sensor, actuator, controller etc.
These subsystems communicate among themselves via interfaces, whose function
consists basically of decoding the transmitted information from one medium to
another. Fg. 1 shows the block diagram representation of a typical robotic
mechanical system. The input is a prescribed task, which is defined earlier. The
output of a robotic mechanical system is the actual task, which is monitored by the
sensors. These sensors sense and transmit the information in the form of feedback
signals.
Speech processing is the computational analysis, manipulation, and interpretation of spoken
language, encompassing techniques like automatic speech recognition (ASR), speech synthesis
(TTS), and speaker recognition. It bridges linguistics, computer science, and electrical
engineering to convert audio signals into text or insights, enabling devices to understand and
generate human voice.
Key Techniques and Principles
Acquisition & Digitalization: Converting analog sound waves into digital signals for
analysis
.
Feature Extraction: Identifying key components in spectral (frequency) and temporal
(time) domains, such as using Fourier transforms.
Modeling: Using probabilistic models to match patterns (e.g., Hidden Markov Models,
Neural Networks) to recognize phonemes and words.
Speech Synthesis: Converting text or data into audible, synthesized human speech.