A Brief NLP + AI

Erin M. Buchanan

06/03/2019

What are we going to talk about?

Natural Language Processing

Computational Linguistics

Dealing with language (which is messy)

Artificial Intelligence

What will you learn?

How key concepts from text mining and linguistics are used to describe and analyze language

How data structures and algorithms are used in text mining and NLP

How we can expand these techniques and apply them to artificial intelligence

What is NLP?

Natural language processing

Roots in computer science, artificial intelligence, and linguistics

Focuses on human language and how to analyze language data

What is language? How do we deal with such a messy construct?

Origins of NLP

Turing Test – Intelligence (1950)

Chinese Room Thought Experiment by Searle (1980)

Georgetown Experiment – Machine Translation (1954)

NLP Systems (1960s)

SHRDLU

ELIZA

Explosion in research given computational power increases, corpus linguistics, and machine learning

Why Study NLP?

80% of “big data” is unstructured data

Images

Videos

Human language (text, recordings)

Text Mining (text analytics, sentiment analysis, etc.)

Linguistic, statistical, and machine learning techniques used to derive high-quality information from text

Traditional Approaches to Text Analytics

Semantics

Readability

Student interest indices

Vocabulary

Frequency, frequency, frequency

Factor/cluster analysis

Word clouds

Pages, chapters, etc.

Terms to Know

Corpus: a body of linguistic data

Corpus of Contemporary American English

Terms to Know

Corpora have changed the face of NLP

The avaliability of data on the internet (and sharing!) has given us a world of possibilities when it comes to analyzing language

Also a large increase in corpora that AREN’T in English!

Terms to Know

Terms to Know

Token: total number of words in a text

Types: number of distinct words

Frequency distribution: a list of all the unique tokens (types) and count of how many times they appear

Terms to Know

Dispersion plot: a graphical representation of the location of tokens in a text

Terms to Know

Collocation: a sequence of words that occur together often

n-Gram: n words that occur together

How to Compute Language

Basic Statistics

Frequency: Counts of characters, words, sentences

Lexical Diversity: percentage of unique word tokens

Lexical Dispersion: position of word tokens in the text

How to Compute Language

Word Sense Disambiguation

Determine which word was intended in a given context

serve: help with food or drink; hold an office; put ball into play

dish: plate; course of a meal; communications device

Contextual clues:

The lost children were found by the searchers (agentive)

The lost children were found by the mountain (locative)

The lost children were found by the afternoon (temporal)

How to Compute Language

Pronoun Resolution

Pronouns refers to a noun – like I/you/this

The noun it refers to is called the antecedent

Examples

The thieves stole the paintings. They were subsequently sold.

The thieves stole the paintings. They were subsequently caught.

The thieves stole the paintings. They were subsequently found.

How to Compute Language

Generating Language Output

Question Answering

For example, who sold the paintings?

Machine Translation

Being able to translate from one language to another

Search for google translate fails

Spoken Dialog Systems

Siri, Ok Google, etc.

How to Compute Language

What can I do with NLP?

What can I do with NLP?

What is next?

Artificial intelligence is implemented all around you now, and certainly, you use it on a day to day basis:

Ok Google, Siri, any automated phone system (human speech processing)

Gaming

Watson and intelligent searching

How can this be applied to health?

What is next?

The healthcare industry has finally reached a point of understanding how to more effectively use its data.

Image processing has shown vast possibilities for reading scans quickly and efficiently to develop detection algorithms

What is next?

Paired with Apple + Research Kit, researchers are developing algorithms to look for markers of autism, help track chronic illness, epilepsy, and more.

Paired with NLP and behavioral economics, we might be able to “nudge” patients into better outcomes

Questions?