Advanced NLP: Word Embeddings to BERT

Advanced NLP: Word Embeddings to BERT

For centuries, computers treated language as meaningless symbols. To process text, engineers manually crafted features (character counts, keyword presence, etc.). Then, a revolution: what if we could learn that the words "king" and "queen" are similar? That "Paris" is to "France" as "Berlin" is to "Germany"? Word embeddings made this possible, and they fundamentally changed NLP. Today, models like BERT understand context deeply—they know that "bank" means something different in "river bank" vs. "money bank."

From One-Hot Encoding to Dense Embeddings

Traditionally, text was represented via one-hot encoding: - Vocabulary size: 10,000 words - Each word represented as vector of length 10,000 - Word i has 1 at position i, 0 everywhere else - "king" = [0, 0, 0, 1, 0, ...] - "queen" = [0, 0, 0, 0, 1, ...] Problems: enormous vectors, no semantic information. "king" and "queen" are orthogonal (dot product = 0), despite being similar words.

Dense Embeddings: Represent each word as a small vector (typically 100-300 dimensions) where semantically similar words have similar vectors. - "king" ≈ [0.2, 0.5, -0.3, 0.1, ...] - "queen" ≈ [0.3, 0.4, -0.2, 0.1, ...] - Cosine similarity between "king" and "queen" ≈ 0.95 (very similar!)

Word2Vec: Learning from Context

Word2Vec (Mikolov et al., 2013) learns embeddings by predicting words from context. Key insight: a word is known by the company it keeps (distributional semantics). Train a neural network to predict neighboring words, and the hidden layer becomes the word embedding.

Skip-gram Model: Given target word "banking", predict context words: "the", "crisis", "system" (within 5-word window). Loss = -log P(context | target) = -log(softmax over vocabulary) For each training pair (target, context), the network learns to make context words more likely. But training over full vocabulary (10,000+ words) is expensive.

Solution: Negative Sampling Instead of softmax over all words, just do binary classification: is this a real context word (positive) or random word (negative)? For pair ("banking", "crisis"): 1 positive example Sample 5 negative examples: random words like "penguin", "butter" Now loss is just: Σ[log σ(u_context · v_target) + Σ_negatives log σ(-u_neg · v_target)] Where σ is sigmoid and u, v are embedding vectors.

import numpy as np
import torch
import torch.nn as nn

class Word2VecSkipGram(nn.Module):
    def __init__(self, vocab_size, embedding_dim):
        super().__init__()
        self.target_embeddings = nn.Embedding(vocab_size, embedding_dim)
        self.context_embeddings = nn.Embedding(vocab_size, embedding_dim)

    def forward(self, target_ids, context_ids, negative_ids):
        # target shape: (batch_size,)
        # context_ids shape: (batch_size,)
        # negative_ids shape: (batch_size, num_negatives)

        target_emb = self.target_embeddings(target_ids)  # (batch, embedding_dim)
        context_emb = self.context_embeddings(context_ids)  # (batch, embedding_dim)
        negative_emb = self.context_embeddings(negative_ids)  # (batch, num_negatives, embedding_dim)

        # Positive score: dot product
        pos_score = torch.sum(target_emb * context_emb, dim=1)  # (batch,)
        pos_loss = -torch.mean(torch.log(torch.sigmoid(pos_score)))

        # Negative scores
        neg_scores = torch.bmm(negative_emb, target_emb.unsqueeze(2)).squeeze(2)  # (batch, num_negatives)
        neg_loss = -torch.mean(torch.log(torch.sigmoid(-neg_scores)))

        return pos_loss + neg_loss

Remarkable Property - Analogy: v_king - v_man + v_woman ≈ v_queen Subtracting "man" removes maleness; adding "woman" adds femaleness. This algebraic property emerges from training only on context prediction!

GloVe: Global Vectors

GloVe (Pennington et al., 2014) combines count-based methods with learning. Create a co-occurrence matrix: how often word i appears near word j. Then factorize this matrix to get embeddings. Loss = Σ_ij f(X_ij) (w_i · w_j + b_i + b_j - log X_ij)² The function f(X_ij) weights rare co-occurrences less heavily (they're noisy). This balances the global structure of language (co-occurrence statistics) with local learning.

GloVe embeddings are competitive with Word2Vec but leverage global statistics explicitly.

Contextual Embeddings: Word Meaning Depends on Context

Word2Vec and GloVe give the same embedding for a word regardless of context: - "The bank approved the loan" - "I sat by the river bank" Both uses of "bank" get the same vector. But "bank" has completely different meanings! This is the fundamental limitation of static embeddings.

Solution: Contextual embeddings compute word vectors based on surrounding context. The same word gets different embeddings in different contexts.

Transformers: The Architecture Revolution

Transformers (Vaswani et al., 2017) introduced self-attention: the ability to compare a word with all other words in the sentence simultaneously. No recurrence (unlike RNNs), so all words are processed in parallel.

Self-Attention Mechanism: For each word, compute: - Query Q: "What am I looking for?" - Key K: "What am I offering?" - Value V: "What information do I have?" Attention scores = softmax(Q × K^T / √d_k) × V For each word, attention computes a weighted combination of all word values, with weights determined by query-key similarity.

import torch
import torch.nn as nn

class MultiHeadAttention(nn.Module):
    def __init__(self, d_model=512, num_heads=8):
        super().__init__()
        self.num_heads = num_heads
        self.d_k = d_model // num_heads

        self.W_q = nn.Linear(d_model, d_model)
        self.W_k = nn.Linear(d_model, d_model)
        self.W_v = nn.Linear(d_model, d_model)
        self.W_o = nn.Linear(d_model, d_model)

    def forward(self, Q, K, V, mask=None):
        batch_size = Q.shape[0]

        # Linear transformations and split into heads
        Q = self.W_q(Q).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
        K = self.W_k(K).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
        V = self.W_v(V).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)

        # Attention: (batch, num_heads, seq_len, seq_len)
        scores = torch.matmul(Q, K.transpose(-2, -1)) / math.sqrt(self.d_k)

        if mask is not None:
            scores = scores.masked_fill(mask == 0, -1e9)

        attention = torch.softmax(scores, dim=-1)
        context = torch.matmul(attention, V)

        # Concatenate heads
        context = context.transpose(1, 2).contiguous()
        context = context.view(batch_size, -1, self.num_heads * self.d_k)

        # Final linear transformation
        output = self.W_o(context)
        return output, attention

BERT: Bidirectional Encoder Representations from Transformers

BERT (Devlin et al., 2018) is a transformer trained to predict masked words. Given: "The [MASK] sat on the [MASK]" Predict: - Position 1: "cat", "dog", "bird" - Position 2: "chair", "mat", "table" By predicting masked words, BERT learns deep contextual representations. Unlike previous models that process left-to-right, BERT uses bidirectional context—every word attends to all other words.

Training Objectives: 1. Masked Language Model (MLM): Predict masked tokens 2. Next Sentence Prediction (NSP): Given two sentences, predict if second follows first These objectives force BERT to learn both word-level and sentence-level representations.

BERT Variants: - BERT-base: 12 transformer layers, 110M parameters - BERT-large: 24 layers, 340M parameters - RoBERTa: Improved BERT training (longer training, more data) - DistilBERT: Smaller, faster BERT (40% smaller, 60% faster) - Multilingual BERT: Trained on 104 languages, works cross-lingual

Transfer Learning with BERT

BERT is typically not trained from scratch. Instead: 1. Use pre-trained BERT (trained on massive text corpus like Wikipedia) 2. Fine-tune on downstream task (sentiment, named entity, question answering) For sentiment classification: - Add a classification head on top of BERT - Fine-tune on sentiment data - Achieves SOTA with minimal task-specific data This is revolutionary: one model learned on general language, applicable to many tasks.

Recent Models: Beyond BERT

GPT Series: Generate text by predicting next token. Unlike BERT's bidirectional context, GPT is unidirectional (left-to-right). But GPT-3 (175B parameters) shows remarkable few-shot learning. T5: Text-to-text transfer transformer. Frames all NLP as text generation: "classify: toxic [sentence]", "translate english to german: [text]" Multilingual Models: mBERT, mT5 work across languages. XLM-RoBERTa trained on 100+ languages. Domain-Specific Models: BioBERT (biology), SciBERT (science), FinBERT (finance) fine-tuned on domain corpora.

Indian NLP and Languages

Challenges: Most BERT models trained primarily on English. Indian languages (Hindi, Tamil, Telugu, Bengali) are underrepresented. Progress: Indic-BERT, mBERT with fine-tuning on Indian languages. Companies like Koo, Dailyhunt use multilingual NLP. Research: IIT researchers work on low-resource language NLP, code-mixed text (English-Hindi), and Indian language MT.

Key Takeaways

• Word embeddings map words to dense vectors where similar words are close
• Word2Vec learns embeddings by predicting context from target word
• Negative sampling makes Word2Vec training efficient
• GloVe combines global co-occurrence statistics with factorization
• Contextual embeddings adjust word vectors based on context (unlike static embeddings)
• Transformers use self-attention to process all words in parallel
• BERT uses masked language modeling and bidirectional context
• Transfer learning enables BERT to solve many tasks with minimal fine-tuning
• Recent models (GPT, T5) show remarkable few-shot learning
• Indian language NLP is improving with multilingual models and domain-specific research

Deep Dive: Advanced NLP: Word Embeddings to BERT

At this level, we stop simplifying and start engaging with the real complexity of Advanced NLP: Word Embeddings to BERT. In production systems at companies like Flipkart, Razorpay, or Swiggy — all Indian companies processing millions of transactions daily — the concepts in this chapter are not academic exercises. They are engineering decisions that affect system reliability, user experience, and ultimately, business success.

The Indian tech ecosystem is at an inflection point. With initiatives like Digital India and India Stack (Aadhaar, UPI, DigiLocker), the country has built technology infrastructure that is genuinely world-leading. Understanding the technical foundations behind these systems — which is what this chapter covers — positions you to contribute to the next generation of Indian technology innovation.

Whether you are preparing for JEE, GATE, campus placements, or building your own products, the depth of understanding we develop here will serve you well. Let us go beyond surface-level knowledge.

Transformer Architecture: The Engine Behind GPT and Modern AI

The Transformer architecture, introduced in the landmark 2017 paper "Attention Is All You Need," revolutionised NLP and eventually all of deep learning. Here is the core mechanism:

# Self-Attention Mechanism (simplified)
import numpy as np

def self_attention(Q, K, V, d_k):
    """
    Q (Query): What am I looking for?
    K (Key):   What do I contain?
    V (Value): What do I actually provide?
    d_k:       Dimension of keys (for scaling)
    """
    # Step 1: Compute attention scores
    scores = np.matmul(Q, K.T) / np.sqrt(d_k)

    # Step 2: Softmax to get probabilities
    attention_weights = softmax(scores)

    # Step 3: Weighted sum of values
    output = np.matmul(attention_weights, V)
    return output

# Multi-Head Attention: Run multiple attention heads in parallel
# Each head learns different relationships:
# Head 1: syntactic relationships (subject-verb agreement)
# Head 2: semantic relationships (word meanings)
# Head 3: positional relationships (word order)
# Head 4: coreference (pronoun → noun it refers to)

The key insight of self-attention is that every token can attend to every other token simultaneously (unlike RNNs which process sequentially). This parallelism enables efficient GPU training. The computational complexity is O(n²·d) where n is sequence length and d is dimension, which is why context windows are a major engineering challenge.

State-of-the-art developments include: sparse attention (reducing O(n²) to O(n·√n)), mixture of experts (MoE — activating only a subset of parameters per input), retrieval-augmented generation (RAG — grounding responses in external documents), and constitutional AI (alignment through principles rather than RLHF alone). Indian researchers at institutions like IIT Bombay, IISc Bangalore, and Microsoft Research India are actively contributing to these frontiers.

India's Scale Challenges: Engineering for 1.4 Billion

Building technology for India presents unique engineering challenges that make it one of the most interesting markets in the world. UPI handles 10 billion transactions per month — more than all credit card transactions in the US combined. Aadhaar authenticates 100 million identities daily. Jio's network serves 400 million subscribers across 22 telecom circles. Hotstar streamed IPL to 50 million concurrent viewers — a world record. Each of these systems must handle India's diversity: 22 official languages, 28 states with different regulations, massive urban-rural connectivity gaps, and price-sensitive users expecting everything to work on ₹7,000 smartphones over patchy 4G connections. This is why Indian engineers are globally respected — if you can build systems that work in India, they will work anywhere.

Advanced Algorithms: Dynamic Programming and Graph Theory

Dynamic Programming (DP) solves complex problems by breaking them into overlapping subproblems. This is a favourite in competitive programming and interviews:

# Longest Common Subsequence — classic DP problem
# Used in: diff tools, DNA sequence alignment, version control

def lcs(s1, s2):
    m, n = len(s1), len(s2)
    dp = [[0] * (n + 1) for _ in range(m + 1)]

    for i in range(1, m + 1):
        for j in range(1, n + 1):
            if s1[i-1] == s2[j-1]:
                dp[i][j] = dp[i-1][j-1] + 1
            else:
                dp[i][j] = max(dp[i-1][j], dp[i][j-1])

    return dp[m][n]

# Dijkstra's Shortest Path — used by Google Maps!
import heapq

def dijkstra(graph, start):
    dist = {node: float('inf') for node in graph}
    dist[start] = 0
    pq = [(0, start)]  # (distance, node)

    while pq:
        d, u = heapq.heappop(pq)
        if d > dist[u]:
            continue
        for v, weight in graph[u]:
            if dist[u] + weight < dist[v]:
                dist[v] = dist[u] + weight
                heapq.heappush(pq, (dist[v], v))

    return dist

# Real use: Google Maps finding shortest route from
# Connaught Place to India Gate, considering traffic weights

Dijkstra's algorithm is how mapping applications find optimal routes. When you ask Google Maps to navigate from Mumbai to Pune, it models the road network as a weighted graph (intersections are nodes, roads are edges, travel time is weight) and runs a variant of Dijkstra's algorithm. Indian highways, city roads, and even railway networks can all be modelled this way. IRCTC's route optimisation for trains across 13,000+ stations uses graph algorithms at its core.

Research Frontiers and Open Problems in Advanced NLP: Word Embeddings to BERT

Beyond production engineering, advanced nlp: word embeddings to bert connects to active research frontiers where fundamental questions remain open. These are problems where your generation of computer scientists will make breakthroughs.

Quantum computing threatens to upend many of our assumptions. Shor's algorithm can factor large numbers efficiently on a quantum computer, which would break RSA encryption — the foundation of internet security. Post-quantum cryptography is an active research area, with NIST standardising new algorithms (CRYSTALS-Kyber, CRYSTALS-Dilithium) that resist quantum attacks. Indian researchers at IISER, IISc, and TIFR are contributing to both quantum computing hardware and post-quantum cryptographic algorithms.

AI safety and alignment is another frontier with direct connections to advanced nlp: word embeddings to bert. As AI systems become more capable, ensuring they behave as intended becomes critical. This involves formal verification (mathematically proving system properties), interpretability (understanding WHY a model makes certain decisions), and robustness (ensuring models do not fail catastrophically on edge cases). The Alignment Research Center and organisations like Anthropic are working on these problems, and Indian researchers are increasingly contributing.

Edge computing and the Internet of Things present new challenges: billions of devices with limited compute and connectivity. India's smart city initiatives and agricultural IoT deployments (soil sensors, weather stations, drone imaging) require algorithms that work with intermittent connectivity, limited battery, and constrained memory. This is fundamentally different from cloud computing and requires rethinking many assumptions.

Finally, the ethical dimensions: facial recognition in public spaces (deployed in several Indian cities), algorithmic bias in loan approvals and hiring, deepfakes in political campaigns, and data sovereignty questions about where Indian citizens' data should be stored. These are not just technical problems — they require CS expertise combined with ethics, law, and social science. The best engineers of the future will be those who understand both the technical implementation AND the societal implications. Your study of advanced nlp: word embeddings to bert is one step on that path.

Key Vocabulary

Here are important terms from this chapter that you should know:

🏗️ Architecture Challenge

Design the backend for India's election results system. Requirements: 10 lakh (1 million) polling booths reporting simultaneously, results must be accurate (no double-counting), real-time aggregation at constituency and state levels, public dashboard handling 100 million concurrent users, and complete audit trail. Consider: How do you ensure exactly-once delivery of results? (idempotency keys) How do you aggregate in real-time? (stream processing with Apache Flink) How do you serve 100M users? (CDN + read replicas + edge computing) How do you prevent tampering? (digital signatures + blockchain audit log) This is the kind of system design problem that separates senior engineers from staff engineers.

The Frontier

You now have a deep understanding of advanced nlp: word embeddings to bert — deep enough to apply it in production systems, discuss tradeoffs in system design interviews, and build upon it for research or entrepreneurship. But technology never stands still. The concepts in this chapter will evolve: quantum computing may change our assumptions about complexity, new architectures may replace current paradigms, and AI may automate parts of what engineers do today.

What will NOT change is the ability to think clearly about complex systems, to reason about tradeoffs, to learn quickly and adapt. These meta-skills are what truly matter. India's position in global technology is only growing stronger — from the India Stack to ISRO to the startup ecosystem to open-source contributions. You are part of this story. What you build next is up to you.

Crafted for Class 10–12 • Programming & Coding • Aligned with NEP 2020 & CBSE Curriculum