NLP Applications and Ethics

NLP technology is now embedded in virtually every sector of society — from healthcare and education to social media and law. With this widespread adoption comes significant ethical responsibilities. This lesson surveys the major applications of NLP and examines the ethical challenges that practitioners must address.

---

Major NLP Applications

Machine Translation

Machine translation (MT) automatically translates text between languages.

Generation	Approach	Example
1st generation	Rule-based	Hand-crafted grammar rules
2nd generation	Statistical (SMT)	Phrase-based translation
3rd generation	Neural (NMT)	Seq2Seq with attention
Current	Transformer-based	Google Translate (Transformer), DeepL

from transformers import pipeline

translator = pipeline("translation", model="Helsinki-NLP/opus-mt-en-de")
result = translator("Machine translation has improved dramatically.")
print(result[0]['translation_text'])

Question Answering

Type	Description	Example
Extractive QA	Extracts an answer span from a given passage	SQuAD dataset
Abstractive QA	Generates a new answer (not just a span)	GPT-based QA
Open-domain QA	Finds answers from a large knowledge base	Google Search, Bing Chat

from transformers import pipeline

qa_pipeline = pipeline("question-answering", model="distilbert-base-cased-distilled-squad")

context = """
Natural Language Processing is a subfield of artificial intelligence
that focuses on the interaction between computers and humans through
natural language. The ultimate objective of NLP is to read, decipher,
understand, and make sense of human languages in a manner that is valuable.
"""

result = qa_pipeline(question="What is the objective of NLP?", context=context)
print(f"Answer: {result['answer']}")
print(f"Confidence: {result['score']:.4f}")

Text Summarisation

Type	Description
Extractive	Selects the most important sentences from the original text
Abstractive	Generates new sentences that capture the key ideas

from transformers import pipeline

summarizer = pipeline("summarization", model="facebook/bart-large-cnn")

article = """
The Transformer architecture was introduced in 2017 in the paper
'Attention Is All You Need' by Vaswani et al. It replaced recurrent
neural networks with self-attention mechanisms, allowing for parallel
processing of input sequences. This innovation led to significant
improvements in machine translation and later became the foundation
for models like BERT, GPT, and T5, which have achieved state-of-the-art
results across virtually all NLP tasks.
"""

summary = summarizer(article, max_length=60, min_length=20)
print(summary[0]['summary_text'])

Chatbots and Conversational AI

Component	Description
Intent recognition	Understanding what the user wants
Entity extraction	Identifying key information (names, dates, numbers)
Dialogue management	Tracking conversation state and deciding the next action
Response generation	Producing a natural language reply

Information Extraction

Task	Description
Named Entity Recognition	Identifying people, places, organisations
Relation Extraction	Finding relationships between entities
Event Extraction	Identifying events and their participants
Knowledge Graph Construction	Building structured knowledge from unstructured text

Content Moderation

NLP powers automated content moderation on social media platforms:

• Hate speech detection
• Toxic comment classification
• Misinformation flagging
• Spam filtering
• NSFW content detection

---

Ethical Challenges in NLP

1. Bias in Language Models

Language models learn from human-generated text, which contains societal biases. These biases are reflected — and sometimes amplified — in model outputs.

Bias Type	Example
Gender bias	"The doctor... he" vs "The nurse... she" — models associate professions with genders
Racial bias	Sentiment analysis scores African American English more negatively
Cultural bias	Models trained primarily on English/Western data may not represent other cultures fairly
Socioeconomic bias	Language associated with lower socioeconomic status may be rated as less professional

# Demonstrating gender bias in word embeddings
# king - man + woman = queen (intended)
# programmer - man + woman = homemaker (biased!)

from transformers import pipeline

fill_mask = pipeline("fill-mask", model="bert-base-uncased")

# These may reveal gender biases
print(fill_mask("The nurse said [MASK] would be right back."))
print(fill_mask("The engineer said [MASK] would be right back."))

2. Privacy Concerns