Semantica

The Geometry of Meaning: A Word Embedding Exploration Tool

Semantica is a navigational tool for the hidden "map" of human language. It treats language not as a sequence of letters, but as a high-dimensional geometric space where word meaning is defined by physical coordinates.

---

The Theory: How Words Become Space

1. The Language of Coordinates

Humans understand words through context and emotion. Computers, however, only understand numbers. To bridge this gap, we use Word Embeddings.

Imagine every word in the English language is a floating point in a dark, infinite void.

• Words with similar meanings (like Ocean and Sea) are clustered tightly together.
• Opposites or unrelated words (like Apple and Justice) are miles apart.

2. The 300-Dimensional "Point Cloud"

In a 3D software like Houdini or Maya, a point is defined by 3 coordinates: $(X, Y, Z)$. Semantica operates in a 300-dimensional space. Every word is a vector—a "pointer" from the center of the universe $(0,0,0...)$ to a specific coordinate in this massive 300-axis grid. These extra dimensions allow the computer to capture subtle nuances: one axis might represent "masculinity," another "royalty," and another "temperature."

3. Static Embeddings (GloVe)

This project utilizes GloVe (Global Vectors for Word Representation). These are "static" embeddings, meaning the computer has pre-scanned billions of lines of text (from Wikipedia and the web) to calculate these coordinates based on co-occurrence. If two words frequently appear near each other in the real world, they are moved closer together in the vector space.

---

Semantic Arithmetic & Spatial Logic

The true magic of Semantica is that because words are numbers, you can perform math on them to uncover cultural relationships and navigate linguistic clusters.

1. Relational Analogies (+ and -)

By adding and subtracting vectors, you can "transport" meanings across the map.

$$Vector(\text{"King"}) - Vector(\text{"Man"}) + Vector(\text{"Woman"}) \approx Vector(\text{"Queen"})$$

By subtracting the "man" vector from "king," we mathematically strip away the concept of masculinity while retaining "royalty." Adding "woman" applies femininity to that royal essence, landing us at the coordinates for "queen."

2. Outlier Detection: The "Odd One Out" (|)

The | operator allows you to perform Spatial Filtering. If you provide a list of words, Semantica will identify which word mathematically "doesn't belong."

Example: apple | banana | meat | orange $\rightarrow$ "meat"

How it works:

1. The Centroid: Semantica calculates the "Center of Gravity" (the average coordinate) for all words in your list.
2. The Distance Scan: It then measures the distance from each word to that center.
3. The Outlier: Words like apple, banana, and orange exist in a tight "Fruit Cluster." Meat, however, is located in a distant part of the 300D space. Semantica flags the word with the highest distance from the group as the outlier.

---

Tech Stack & Architecture

Semantica is built as a lightweight, high-performance engine for spatial language analysis.

The Backend (The Engine)

• Language: Python 3.x
• Math Engine: NumPy. We use vectorized matrix operations to calculate Euclidean distance and Cosine Similarity across 400,000+ words in milliseconds.
• Data Source: GloVe 6B (300-dimensional vectors).

The Frontend (The Interface)

• CLI / Web Interface: A minimalist interface designed for rapid expression evaluation.
• Parser: A custom regex-based expression evaluator that translates human-readable strings (like paris - france + italy) into NumPy-executable operations.

---

Getting Started

1. Prerequisites

• Python 3.10+
• NumPy

2. Database Setup

The core logic uses GloVe embeddings.

1. Download glove.6B.zip.
2. Extract the contents.
3. Place glove.6B.300d.txt into the /database folder.

3. Usage

• Install Python,
• Download/clone repo,
• Download embedings file (database setup),
• Fix and run semantica.bat (change path to Python).