hole

language, models, space, linguistics; logic; computer science; artificial intelligence; cognitive science; semiotics
meaning making, translation, metaphors, slip, address, relativity, latency, hole
REFS: 
🎻==very important 
for LLMs (large language models), computer science:
🎻*(1)https://www.howdoai.org/en/transformer/embeddings-explained/?utm
*(2)https://www.geeksforgeeks.org/nlp/word-embeddings-in-nlp/?utm
*(3)von Rütte, D., Anagnostidis, S., Bachmann, G., & Hofmann, T. (2023). A Language Model’s Guide Through Latent Space. 
*(4)Washio, K., & Kato, T. (2021). Neural latent relational analysis to capture lexical semantic relations in a vector space. 
🎻*(5)book: Kornai, A. (2023). Vector Semantics.
for cognitive science and psychology :
* (6)Evangelopoulos, N. E. (2013). Latent semantic analysis: Introduction and overview. 
*(7)Grand, G., Blank, I., Pereira, F., & Fedorenko, E. (2023). Semantic projection: Recovering human knowledge of multiple, distinct object features from word embeddings
🎻*(8)Spivak, G. C. (1993). The politics of translation.
*(9)book: Lakoff, G., & Núñez, R. E. (2000). Where mathematics comes from: How the embodied mind brings mathematics into being.,
*(10)book: Lakoff, G. (1987). Women, fire, and dangerous things: What categories reveal about the mind.
*(11)deconstruction: Derrida, J. (1976). Of grammatology (G. C. Spivak, Trans.)
(on Kristeva, the semiotic and the symbolic)
*(12)Talebian Sadehi, C. (2012). Beloved and Julia Kristeva’s the semiotic and the symbolic.
*(13)Ducard, D. (2020). The semiotic chora and the inner life of language.
for poetry and other outputs:
🎻*(14)Stein, G. (1914). Tender buttons: Objects, food, rooms.
*(15)Parrish, A. (2019‑2021). 22 Compasses [computer‑generated poetry].
🎻*(16)Lispector, C. (1973/2012). Agua viva (R. Buchwald, Trans.)
*(17)some memes and twitter i gotta add here***

it started to take real shape with these whatsapp messages on a group of my beloved developers:


[13:44, 09/09/2025] Asoş: kanka latent space siksok teoride çok geçiyo takılma
[13:44, 09/09/2025] Asoş: olay şu
[13:44, 09/09/2025] Asoş: yz modeli düşün, language
[13:44, 09/09/2025] Asoş: şimdi bu model cümle kuruyo işte olayı bu
[13:45, 09/09/2025] Asoş: pardon asıl olayı
[13:45, 09/09/2025] Asoş: dil bilmesi
[13:45, 09/09/2025] Asoş: diyelim türkçe biliyor
[13:45, 09/09/2025] Asoş: arkasında kelimeler grup halinde oluyo ya, mesela "ayı" kelimesi bi izdüşümde hayvan kategorisinde başka bi iz düşümde yıldız takımı kategorisinde, başka bi iz düşümde "ay ile başlayan kelimeler"
[13:46, 09/09/2025] Asoş: falan böyle yüz farklı clusterda kelimeler,,, anlamlarını veya kullanımlarını anlaması için modelin
[13:46, 09/09/2025] Asoş: bunu biliyosundur amk Yoksa ben mi yanlışım yok mu böyle bişi
[13:46, 09/09/2025] Asoş: bu 100D uzay işte latent sapce
[15:22, 09/09/2025] Asoş: yani şimdi her kelimenin bu 100-D uzayda adresi var
[15:23, 09/09/2025] Asoş: hatta bu modeli insan beyninde düşünürsek bu uzayda anısal/duygusal boyutları (koordinatları) da var kelimelerin
[15:24, 09/09/2025] Asoş: 2D uzayda A'dan B'ye giderken C'nin yanından geçersin D'nin üzerinden ya mesela,,, D noktasına gitmek aslında bi yolculuktur hani (vektör icabı
[15:24, 09/09/2025] Asoş: bu 100D uzayda da bi kelimeye ulaşmadan önce beyninde yolculuk yapıyosun
[15:25, 09/09/2025] Asoş: o kelimenin saf sözlük anlamı dışında bir sürü farklı katmanda adresi var beyninde ve bundan dolayı bi kelimeden başka kelimeye geçerken kafandaki yolculukta bir sürü farklı kelime/anlam/tarz/duygunun yanından/üstünden vs geçiyosun
[15:26, 09/09/2025] Asoş: fikir bu,,, kelimelerin bi nokta gibi değil aslında yolculuk gibi çalıştığı
[15:27, 09/09/2025] Asoş: zaten bu yüzden bazen kafamızdaki şey yanlış/absürt kelime olarak çıkıyo ağzımızdan bazen bu yolculukta kayma oluyo falan
[15:27, 09/09/2025] Asoş: ya da mecaz yapmak da bi nevi kaymak
[15:27, 09/09/2025] Asoş: şiirde de benzer
[15:29, 09/09/2025] Asoş: teorim böyle,,,,, kelimelere sözlük dışı bi bakış, sözlük anlamını siktir etmece ve geriye kalanlarla eğlenmece
[15:30, 09/09/2025] Asoş: kelimeye anlam turizmi yapıcaz


i was saying there:
Behind that knowledge, words aren’t just isolated — they sit in clusters of meanings/usages.
A word can belong to many overlapping projections/clusters like this, which helps the model understand meaning and usage. All those clusters live in a multidimensional space (like 100D). That space of word-meanings and relations is what’s called latent space.
👉 latent space = the high-dimensional representational space where words (and concepts) are mapped in relation to each other, based on how the model has learned them.
• Like in a 2D map, going from A to B might force you to pass by C or over D → same in this multidimensional mental space:
To reach one word, you mentally pass through or near other associations, memories, feelings. •
That’s why: Sometimes the “wrong” or absurd word slips out — you deviated on the journey.
Metaphor (mecaz) is basically a deliberate slip.
Poetry also thrives on those detours.
theory: instead of treating a word as a fixed “point” with one dictionary meaning, treat it as a journey through overlapping coordinates.
“anlam turizmi” (meaning tourism) → playing with all the leftover associations outside the dictionary sense. forget dictionary meanings, let’s travel through the side-streets of a word’s emotional, cultural, poetic neighborhoods.



şimdi article’a başlıyoruz:

chatgpt-written abstract:
This *thinking pattern* proposes a relational and topological understanding of language through the lens of latent space and extends it into a poetic-psychological model termed “anlam turizmi” (meaning tourism). In artificial intelligence models, language is not generated merely through sequential syntax but emerges from a high-dimensional representational space in which words are embedded as vectors. Each word occupies multiple overlapping clusters corresponding to various contextual usages and associations—for instance, the Turkish word “ayı” (bear/constellation/prefix ay) inhabits several semantic neighborhoods simultaneously. While in machine learning these dimensions are abstract mathematical features, they parallel the mental coordinates of the human brain, which include not only semantic but also emotional and mnemonic components.

Traversing from one word to another, whether computationally or cognitively, constitutes a journey through associative terrain—passing through nearby meanings, memories, and affective residues. Errors, metaphors, and poetic expressions can thus be reinterpreted as intentional or accidental deviations within this multidimensional space. “Anlam turizmi” proposes abandoning static, dictionary-based meanings in favor of exploring the side-streets of language, where words reveal their fluid, overlapping, and affective dimensions. This framework reframes metaphor not as ornamental substitution but as movement through latent space, positioning both AI language models and human thought as travelers across interconnected topologies of meaning.

let’s go first with the oldest:::: LSA (LATENT SEMANTIC ANALYSIS): OLDER METHOD. LSA = count + matrix + SVD (linear algebra)
Each word gets one fixed vector ___> freezing the entire language into one sculpture
The idea of LSA influenced everything that came later, but no major modern NLP model literally implements LSA inside it.
LSA =co-occurrence matrix → SVD (singular value decomposition) → reduce dimensions → word meaning emerges from patterns.
It’s a pure linear algebra method for distributional semantics (1990s).
“If you want, I can explain:
🌈 Why LSA fails for polysemy
🧠 How contextual meaning killed LSA
Just say “tell me.””
in word2vec for example they use prediction instead of counting to come up with the “word vectors”


  • Static embeddings (Word2Vec / GloVe / fastText): one fixed vector per word → can’t handle context-dependent meaning very well.

  • Contextual embeddings (ELMo / BERT / GPT): generate a different vector for a word depending on the sentence → meaning updates with context.


    • “Distributional” means:
    you understand a word by looking at the company it keeps.

    This idea comes from the linguist Zellig Harris (1954):

    “Words that appear in similar contexts have similar meanings”


    >>>>>>>>>>>>>>>>>>>>>>>

    first chapter.first chapter.first chapter.first chapter.first chapter


    let’s go now with
    newer Distributional Semantics: non-contextual (stable) vs contextual (dynamic) semantics

    >>>>>>>>>>>>>>>>>>>>>>>
    word2vec → GloVe → fastText → ELMo → BERT → GPT
    with the degree of contextuality, dynamism, and fuzziness
    1- word2vec. glove. fasttext. > noncontextual
    (statistical similarity) (static embedding models:::::Meaning represented as stable vectors)
    learned via prediction, not svd.
    context-free embeddings.
    the naive learning of the corpora >> from texts we get statistical meaning to each word that corresponds to a non-dynamic vector
    different kinds of relations are not heard here.
    for the non-contenxualized and non-dynamic embeddings

    2- elmo. bert. gpt.

    contextualized word representations

    dynamic semantics: meaning = context-updating:::: LLMs instantiate a probabilistic, large-scale dynamic semantics.     dynamic embedding model

    contextual >> word vectors that are sensitive to the context in which they appear.

    **. LLMs behave like “dynamic semantic” systems

    In human language, each sentence updates a discourse context.
    LLMs do the same: each token → updates the hidden state → which shapes the next token.
    So an LLM is basically:a probabilistic context-updating machine.
    This makes them partial models of human meaning-making. Cognitive model? Partially yes, fundamentally no.

    **.LLMs do NOT model human cognition internally

    This is Andreas’ strongest caution:
    LLMs lack:perception,, grounded world models,, conceptual category formation,, causal reasoning,, human-like memory systems

    **.Human-likeness comes from statistical patterns in data

    Humans learn through embodied interaction.
    LLMs learn from corpora — human linguistic traces.
    Therefore:The behavior is similar, the mechanisms are different.


    LLMs dont  think perceive and reason causally like humans.



    >>>>>>>>>>>>>>>>>>>>>>>

    another chapter.another chapter.another chapter.another chapter


    NOW >>>>> NON DISTRIBUTIONALS:::
    they use:
    rules,
    logic,
    manually written knowledge,    human annotations
    language-external signals (vision, sensors, grounding)

    Meaning is defined by hand-written rules, not statistics or Meaning is defined by humans, not by data distribution



    1-4lang algebraic,Meaning represented as polytopes in n-dimensional space>> from (5)

    it’s not a vector, it’s a polytope! (algebraic)
    This is analytical, structural, not distributional.


     

    2- embodied cognition: meaning = grounded in sensory/motor experience (human)
    three refs:
    Lakoff, metaphors we live by;  Spicak, the politics of translation; Barsalou, Grounded cognition.

    1. Embodied cognition claim

    Embodied cognition argues that:
    • Human thinking is shaped by the body, sensory experience, and physical interaction with the world.
    • Our concepts are not abstract and disembodied, but grounded in perception, action, and lived experience.

    2. Conceptual Metaphor Theory claim

    CMT (Lakoff & Johnson, 1980) proposes that:
    • We understand abstract concepts (e.g., time, love, morality) through metaphors based on concrete, bodily experience.
    • Example:

      • “More is up” → based on seeing levels rise when quantity increases.

      • “Rich is heavy and poor is light” → grounded in physical experience of weight and value.

    3. CMT as evidence for embodied cognition

    CMT shows how:
    • The structure of the body influences the structure of thought.
    • Abstract reasoning (e.g., justice, theory, emotion) is built using patterns from sensory–motor experience.

    So metaphor isn’t just a poetic device in language — it is a mechanism of cognition.

    Zhang, Wei (2009). “Embodied Cognition and Chinese Character Semantics.”::::::;:;:;:;:;;;;

    Core idea: The physical form of Chinese characters—the strokes, their order, and spatial structure—is not arbitrary. Instead, it encodes perceptual and motor experience.

    For example, some characters imitate shapes in the physical world (like 山 “mountain” resembles peaks).

    The stroke order guides the motor actions of writing, so understanding/writing a character engages the motor system.

    Implication: Semantics in Chinese is grounded: the meaning of a character is connected to sensorimotor experience rather than just being an abstract symbol.

    💡 In short, Zhang’s work shows that even written language can carry embodied, grounded meaning, linking perception, action, and cognition.

    ----------------------------------------Barsalou, Grounded cognition-------------------------------------



    ---------------------------------------Lakoff, metaphors we live by---------------------------------------
    lol


    ---------------------------------------Spivak, the politics of translation-----------------------------------


    >>>>>>>>>>>>>>>>>>>>>>>

    fun chapter.fun chapter.fun chapter.fun chapter.fun chapter.fun chapter

    Clarice is imagining a painting that is pure event, pure presence, pure interiority — not representation:

    GERTRUDE STEIN

    >>>>>>>>>>>>>>>>>>>>>>>

    best chapter.best chapter.best chapter.best chapter.best chapter.best chapter

    a hole in the space

    ____ terminology _______-English Glossary; year ________

    glass
    aviation
    polymer
    cardiovascular
    social media related something?
    musicology?

    turkish>>>
    Translation glossary: Turkish-English Glossary for Terms on Economics and Business
    https://www.nadirkitap.com/ingilizce-turkce-ekonomi-ve-isletmecilik-terimleri-aciklamali-sozluk-halil-seyidoglu-kitap3666052.html?utm_source=chatgpt.com

    İngilizce - Türkçe Tütün Terimleri Sözlüğü — glossary of tobacco‑industry terms. Tarım ve Orman Bakanlığı
    https://www.tarimorman.gov.tr/TADAB/Belgeler/T%C3%BCt%C3%BCn%20Piyasas%C4%B1/yayinlar/sozluk.pdf?utm_source=chatgpt.com

    arapça>>>
    الميسر: معجم مصطلحات هندسة وادارة التشييد: English‑Arabic by Rudwan Sadu al‑Jurf — construction and construction‑management terms. 1995.

    https://play.google.com/store/books/details/%D8%AF_%D8%B1%D8%B6%D9%88%D8%A7%D9%86_%D8%B3%D8%B9%D8%AF%D9%88_%D8%A7%D9%84%D8%AC%D8%B1%D9%81_%D8%A7%D9%84%D9%85%D9%8A%D8%B3%D8%B1_%D9%85%D8%B9%D8%AC%D9%85_%D9%85%D8%B5%D8%B7%D9%84%D8%AD%D8%A7%D8%AA_%D9%87%D9%86%D8%AF%D8%B3%D8%A9_%D9%88%D8%A5%D8%AF%D8%A7%D8%B1?id=5JCRDwAAQBAJ&utm_source=chatgpt.com

    çince >>>

    - Chen, L. – Computer Science Terminology Dictionary (Chinese-English), 2012 Chinese-English dictionary of computing terms.

    - 医学英语词汇学习手册 (2nd Ed.): “A Concise English‑Chinese Dictionary of Medical Terminology” — focused on anatomy, systems, public health. windowsfront.com

    https://annas-archive.org/md5/cae846559ee1e482237b14034b8f09da
    https://annas-archive.org/md5/0c4ae17674fa56989816a18fc0ec4983
     
    - 中医药术语汉英双语平行语料库: A bilingual corpus / glossary specifically for Traditional Chinese Medicine (TCM) terms.


    ispanyolca >>
    Rojas, J. – Diccionario de Informática: Español-Inglés, 1999 Spanish–English dictionary of computer terms.????
    https://annas-archive.org/md5/60580d6877b93e60b422b3dc999a9a6e
    https://annas-archive.org/md5/66b1cabed4d834af4b151bfcd9c33015
    https://annas-archive.org/md5/bb58b09cd48c11ffac74ddc0045f7443

    English‑Spanish Aviation Dictionary – by David de la Hoz, bilingual aviation phrases for pilots and engineers.

    https://annas-archive.org/md5/6c5d867f5925879f45758a23e0e583af
    https://annas-archive.org/md5/0841ec5dad5579551eac79eddf56d147


    svahili>>

    ???>>
    multilingual-polymer-glossary
    https://iupac.org/multilingual-polymer-glossary/?utm_source=chatgpt.com

    Multilingual Aeronautical Dictionary — AGARD / NATO

    https://www.fzt.haw-hamburg.de/pers/Scholz/MAD.html

    1960
    1970
    1980
    1990
    2000
    2010
    2020


    POEMS:

    1.“Benedict Kolthoff, Glass Terminology; A German-English Glossary, 1967”


    2. 


    3.


    4.


    5.


    6.


    7.