Enabling computers to understand, interpret, and generate human language

Human-computer interaction

Automation of language-heavy workflows

Insights from large-scale text data

Ambiguity (lexical, syntactic, semantic)

Context and pragmatics

Domain shift and low-resource languages

Noise in real-world text (typos, slang, code-switching)

Sentiment analysis

Topic labeling

Spam detection

Intent detection

Part-of-speech (POS) tagging

Named entity recognition (NER)

Chunking

Relation extraction

Event extraction

Keyphrase extraction

Dependency parsing

Constituency parsing

Information retrieval (search)

Semantic similarity

Question answering (retrieval-based)

Machine translation

Summarization (extractive, abstractive)

Dialogue systems / chatbots

Text completion

Speech-to-text (ASR)

Text-to-speech (TTS)

Normalization (lowercasing, cleaning)

Tokenization

Stopword handling

Stemming / lemmatization

Sentence segmentation

Bag-of-Words (BoW)

N-grams

TF-IDF

Word embeddings (Word2Vec, GloVe, FastText)

Contextual embeddings (BERT-style)

Naive Bayes

Logistic regression

SVM

CRF (for sequence labeling)

CNNs for text

RNN/LSTM/GRU for sequences

Transformers (self-attention)

Pretrained language models

Task-specific fine-tuning

Prompting / instruction tuning (conceptually)

Next-token prediction

Masked language modeling

Perplexity as an evaluation signal

Classification: accuracy, precision, recall, F1

Generation: BLEU, ROUGE, METEOR (use with caution)

Retrieval/Ranking: MRR, MAP, NDCG

Human evaluation: fluency, relevance, faithfulness

Data labeling and annotation guidelines

Bias, fairness, and toxicity

Privacy and security (PII handling)

Robustness and adversarial inputs

Python NLP: NLTK, spaCy

Machine learning: scikit-learn

Deep learning: PyTorch, TensorFlow

Hugging Face Transformers

Tokenizers (fast tokenization)

Model hubs and checkpoints

Datasets: Hugging Face Datasets

Experiment tracking: Weights & Biases, MLflow

Annotation tools: Prodigy, Label Studio

APIs and services (REST, batch pipelines)

Serving: TorchServe, TensorFlow Serving

Optimization: quantization, distillation, caching

Public benchmarks (GLUE/SuperGLUE, SQuAD)

Web and enterprise documents

Conversational logs (with consent and anonymization)

Define task, constraints, success metrics

Cleaning, splitting, augmentation (if applicable)

Baseline → improved model (feature-based → transformer)

Hyperparameters, regularization, early stopping

Offline metrics + error analysis

Bias and robustness checks

Latency, drift, feedback loops

Continuous improvement