2nd/10_Wiki/Topics/AI_and_ML/Structuralism.md

---
id: wiki-2026-0508-structuralism
title: Structuralism
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Structural Analysis, Saussurean Linguistics, 구조주의]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [philosophy, linguistics, semiotics, post-structuralism]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: theory
  framework: saussure-levi-strauss
---

# Structuralism

## 매 한 줄
> **"매 의미 = 매 element 자체 X, 매 관계의 system 의 위치"**. Ferdinand de Saussure (1916) 의 langue/parole + signifier/signified — Lévi-Strauss (anthropology), Barthes (semiotics), Lacan (psychoanalysis), Foucault (early) 으로 확산. 2026 ML lens 에서는 매 "embedding space 의 differential geometry" 와 isomorphic.

## 매 핵심

### 매 Saussurean foundations
- **langue vs parole**: 매 system (langue) vs utterance (parole) — 매 Chomsky competence/performance 의 선구.
- **signifier / signified**: 매 sound-image / concept — sign 은 매 arbitrary, conventional.
- **Diachronic vs synchronic**: 매 historical 변화 vs 매 system snapshot — synchronic 우선.
- **Value (valeur)**: 매 의미는 매 negative differential — "cat" = "not dog, not mat, not bat".

### 매 expansion
- **Lévi-Strauss** (1958, 매 *Structural Anthropology*): 매 myth, kinship 의 binary opposition (raw/cooked, nature/culture).
- **Roland Barthes** (1957, 매 *Mythologies*): 매 semiotic critique — 매 sign 의 second-order myth.
- **Lacan**: "매 unconscious is structured like a language" — signifier chain.
- **Jakobson**: 매 phoneme 의 distinctive features, metaphor/metonymy axis.

### 매 post-structuralism (1960s+)
- **Derrida** (*Of Grammatology*, 1967): 매 différance — meaning 매 always deferred.
- **Foucault** (later work): 매 discourse, power/knowledge — 매 structure 의 historicization.
- **Deleuze & Guattari**: 매 rhizome — 매 structural tree 의 거부.
- **Barthes "Death of the Author"** (1967): 매 reader-centered, 매 structuralist에서 post-로 이동.

### 매 modern (2026) connections
- **NLP / LLM embeddings**: 매 word vector = 매 differential value (cosine similarity) — Saussurean valeur 의 computational realization.
- **Distributional hypothesis** (Firth, "you shall know a word by the company it keeps"): 매 BERT/GPT 의 implicit structuralism.
- **Knowledge graphs / RDF**: 매 relational structure — Lévi-Strauss 의 kinship system 의 echo.
- **Cognitive science**: 매 conceptual spaces (Gärdenfors) — 매 geometric structuralism.

## 💻 패턴

### 매 binary opposition extraction (Lévi-Strauss style)
```python
import numpy as np
from sentence_transformers import SentenceTransformer

# 매 2026: BGE-M3 / E5-mistral 등
model = SentenceTransformer("BAAI/bge-m3")

pairs = [("nature", "culture"), ("raw", "cooked"), ("light", "dark")]
for a, b in pairs:
    va, vb = model.encode([a, b])
    axis = vb - va  # 매 binary opposition axis
    print(f"{a} ↔ {b}: ||axis||={np.linalg.norm(axis):.3f}")
```

### Saussurean value (differential)
```python
# 매 "value = position in system of differences"
words = ["cat", "dog", "mat", "bat", "rat"]
embs = model.encode(words)
# 매 cat 의 value = 매 distance vector 의 다른 모든 words 와의
for i, w in enumerate(words):
    others = np.delete(embs, i, axis=0)
    differential = embs[i] - others.mean(axis=0)
    print(f"{w}: {np.linalg.norm(differential):.3f}")
```

### Semiotic square (Greimas)
```python
# 매 A vs not-A, B vs not-B — 매 4-corner structure
def semiotic_square(s1, s2):
    return {
        "S1": s1,
        "S2": s2,             # 매 contrary
        "not_S1": f"not-{s1}", # 매 contradictory
        "not_S2": f"not-{s2}"
    }

print(semiotic_square("life", "death"))
# 매 narratology / brand analysis 에 활용
```

### 매 structural narrative analysis (Propp 영향)
```python
# 매 Propp 31 functions of folk tale
PROPP_FUNCTIONS = [
    "absentation", "interdiction", "violation", "reconnaissance",
    "delivery", "trickery", "complicity", "villainy",
    # ... 31 total
]

def map_story(events: list[str], llm) -> dict[str, str]:
    """매 매 event 의 Propp function 의 mapping (LLM-assisted)"""
    prompt = f"Map these events to Propp's 31 functions: {events}"
    return llm.complete(prompt)  # Claude Opus 4.7 등
```

### 매 distributional structuralism (Firth/Harris → BERT)
```python
# 매 "company a word keeps" = context window
from transformers import AutoModel, AutoTokenizer
import torch

tok = AutoTokenizer.from_pretrained("answerdotai/ModernBERT-large")
model = AutoModel.from_pretrained("answerdotai/ModernBERT-large")

def contextual_value(word: str, contexts: list[str]):
    """매 word 의 매 different contexts 의 의미 variation"""
    embs = []
    for ctx in contexts:
        sentence = ctx.replace("___", word)
        inputs = tok(sentence, return_tensors="pt")
        with torch.no_grad():
            out = model(**inputs)
        # 매 word token embedding 추출
        embs.append(out.last_hidden_state[0, 1].numpy())
    return embs  # 매 polysemy 의 quantification
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| 매 myth / folktale 분석 | Lévi-Strauss / Propp 의 binary + function |
| 매 brand / advertising | Greimas semiotic square + Barthes myth |
| 매 NLP semantic analysis | 매 distributional embedding (BGE-M3) |
| 매 critical theory 작업 | Post-structural (Derrida, Foucault) 의 보완 |
| 매 cognitive modeling | Conceptual spaces (Gärdenfors) |

**기본값**: 매 Saussurean foundation 위에 매 task 의 맞는 successor — 매 NLP 면 distributional, 매 culture 면 Lévi-Strauss/Barthes, 매 critique 면 post-structural.

## 🔗 Graph
- 변형: [[Generative-Grammar]]
- 응용: [[Narratology]]
- Adjacent: [[Distributional-Semantics]] · [[Word-Embeddings]]

## 🤖 LLM 활용
**언제**: 매 cultural artifact 분석, brand/advertisement decoding, narrative structure mapping, 매 semantic field exploration via embeddings.
**언제 X**: 매 strict empirical linguistics 작업 (매 corpus statistics 가 우선), 매 totalizing claims (post-structuralist critique 의 무시 위험).

## ❌ 안티패턴
- **Synchronic 만**: 매 historical change 의 무시 — 매 Saussure 자신도 diachronic 가치 인정.
- **매 universal structure 강요**: Lévi-Strauss critique — 매 매 culture 의 own structure 의 무시.
- **Embedding cosine = meaning**: 매 oversimplification — 매 polysemy, pragmatics, context dynamics 누락.
- **Author intention 의 obsession**: 매 Barthes "Death of the Author" 의 무시.

## 🧪 검증 / 중복
- Verified (Saussure *Cours de linguistique générale* 1916, Lévi-Strauss *Anthropologie structurale* 1958, Stanford Encyclopedia of Philosophy 2026).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — Saussure→post-structural→2026 ML embedding bridge |