Computational Linguistics

define computational linguistics

study of written and spoken language from a computational perspective, building artifacts that usefully process and produce language

what is the goal of computational linguistics

recognize language — how system identify sounds, words, and structure of human language

comprehend language — how they analyze meaning + context to interpret what is being said

generate language — how models produce text that sounds meaningful

what is natural language processing? (NLP)

what are the types?

def — build applications that let computers use human language

• speech recognition — turning sound waves into words

• translation — bridging languages instantly

• bias detection — identifying harmful or unfair language patterns in datasets or media

• accessibility tech — creating captions, screen readers, and text to speech tools

what are the key steps and goal to speech to text

goal — turn audio signals into words

key steps

• acoustic analysis — breaking audio into small frames and extracting features

• phoneme recognition — identify the basic units of sounds

• word recognition — mapping sequences of phonemes to words using language models

what are the key steps and goal to text to speech

goal — convert written text into natural sounding speech

key steps

• text analysis — break text into units

• prosody modeling — decode how to stress words, pause, and intonate

• waveform synthesis — generate the actual audio waveform using algorithms or neural networks

what is the distributional hypothesis?

words that appear in similar contexts tend to have similar meanings

1. vector representation — capturing meaning mathematically

2. embeddings — system performs the task of predicting the next word based on the current word

what are n-grams

sequence of n words that can be used to approximate which word should go next in the sequence

• bad because they require a lot of data to identify word relationships, diverse and specific data sets

what are n-gram models ?

choose the most likely word to go next in a sequence given data on which words appear next to each other

what are transformers?

neural network architecture proposed by Google

• contain an attention mechanisms that makes them good at generating human text

• keep track of word meanings in context of the sentence

methods when talking about computational linguistics

• topic modeling — tools to create clusters of similar ideas

• keyword frequency analysis — track how words used changed across stages

• large language models — generate themes based on clusters of similar ideas

what is the lack of linguistics diversity cycle?

what does it lead to?

no or even less data → no models → no tools → speaker switch to dominant languages

• linguistic injustice — tech that claims to be universal serves only part of the world

• cultural loss — language carry unique knowledge systems, stories, worldviews

• power dynamics — reinforces english dominance