Source-linked AI summary
Words, Spaces and Generative AI: Layers of language in contemporary architecture
Anca-Simona Horvath
TL;DR
As text-to-X generative AI becomes commonplace in architectural practice, the chapter asks how language functions as a design material and how linguistic theory can clarify this relationship. It connects historical theories of language and metaphor to generative AI, identifies three intertwined language layers, and proposes research directions for studying their architectural use.
Problem
Text-to-X generative AI is widespread in architectural practice, making the relationship between language and architectural design more important to study.
Method
The chapter connects linguistic theories and architectural research, then analyzes natural language, programming languages, and annotations as intertwined layers in generative-AI design.
Results
The chapter concludes with a research agenda combining corpus linguistics, communication and information studies, and attention to differences among natural languages.
Takeaways & Limitations
Generative-AI architectural research should examine discourse, computational languages, annotations, embodiment, and the metaphors through which design concepts are structured.
Takeaways & Limitations
Graphical and volumetric representations remain important for communicating precise architectural ideas that are difficult to express through natural language.
Abstract
from arXiv · showhide
Language can be considered a design material in architecture, and in the context of text-to-X generative AI models becoming a common tool for architectural practice, looking more closely at language is more important now than in the past. After describing some of the important developments in linguistics starting from Wittgenstein, and including the work of Chomsky, Lakoff, conceptual and generative metaphors as proposed by Schön, this chapter connects them to contemporary architectural design and generative text-to-X tools. The chapter builds on the idea that three main forms of language intertwine in architectural design done using generative AI, namely (I) discourse (or natural language which can contain professional terminology specific to our field), (II) programming languages (which are artificial languages sitting at the basis of all computational systems), and (III) annotations (as language elements attached to pieces of data). It concludes by outlining a research agenda for connecting generative metaphors to generative AI: (a) conducting corpus linguistics studies on architectural texts (using quantitative tools such as topic modelling, and qualitative tools such as discourse analysis); (b) bringing communication theory and information studies closer to architectural research and (c) taking into account that different (natural) languages come with different affordances meaning generative and conceptual metaphors differ in relation to this.
1. Introduction
The chapter examines language as a design material as text-to-X generative AI becomes widespread in architecture. It frames this inquiry through the relationship between natural language, programming languages, and annotations.
- The chapter studies language’s relationship to architectural design as text-to-X generative AI becomes prevalent in practice.It treats language as a design material and responds to the widespread use of accessible text-to-X models.
- Programming languages require attention in architectural education, practice, and research, while text-to-X AI also demands closer study of natural language.
- Architectural design with generative AI intertwines natural language, programming languages, and annotations.Natural language may include professional terminology; programming languages are artificial languages underlying computing; annotations describe images or other data.
- The chapter proceeds from recent text-to-X research to linguistics, then analyzes the three language layers in AI-supported architectural design.
2. Text-to-X Generative Models: prompting architectural concepts, structures, elements
Architectural text-to-X research spans education, conceptual and element generation, structural modeling, vernacular analysis, and prompt design. These studies show a rapidly expanding field while also motivating more structured approaches to textual input.
- Recent research applies generative AI to architectural education and to generating concepts, structures, and elements.
- Text-to-BIM systems can transform textual input into code that invokes BIM APIs and generates editable models with semantic information.
- Text-to-structure research converts natural-language prompts into conditioning vectors that guide generation of structural graphs near static equilibrium.
- Studies of vernacular architecture evaluate generated forms across contexts using standardized prompts, morphological analysis, expert assessment, and the Analytic Hierarchy Process.
- Semiotic prompt analysis is proposed to address trial-and-error workflows and produce outputs that are more semantically relevant.
3. Linguistics and its connection to architecture
The chapter connects linguistic theories of meaning, grammar, metaphor, categorization, and formal systems to architectural design. It argues that these theories help explain how language and computational processes participate in architectural production.
- Architecture as language: Architecture has been interpreted as a visual and spatial language with its own grammatical structure, notably in Christopher Alexander’s A Pattern Language.
- Language and concepts: Wittgenstein linked words, things, and communication by describing language as triggering pictures of communicated facts.He also treated drawings and physical models as important supplements for communicating ideas that spoken or written language cannot reliably convey.
- Generative metaphors: Lakoff and Johnson argued that conceptual metaphors use concrete experience to structure understanding of abstract ideas and shape perception, reasoning, and action.
- Generative metaphors: Schön proposed that generative metaphors frame complex problems and thereby generate the solution space available for thinking.He associated conflicting frames with restructuring tacit metaphors underlying perceptions of social reality.
- Generative grammar: Chomsky proposed innate universal grammar, distinguishing unconscious grammatical competence from messy, error-prone performance.His transformational generative grammar describes transformations from deep structure to surface structure.
- Formal systems: L-systems adapted formal grammars to model parallel biological growth and later supported architectural generative design, parametric form-finding, and urban planning.
- Prototype theory: Prototype theory explains categories through similarity to a generic or best example rather than explicit definitions.
- Connections to generative AI: These theories map onto generative AI architecture: Chomsky informs programming and geometric patterning, Lakoff informs design communication, and Rosch informs data annotations.
4. Layers of language in architectural design
Architectural design with generative AI intertwines natural-language discourse, programming languages, and annotations. This chapter connects these layers to communication, computational systems, and the categorization of visual data.
- Three layers of language: Generative AI tools combine natural language, programming languages, and annotations as intertwined layers of architectural design.Natural language may include professional terminology; programming languages underpin computing; annotations describe images or other data.
- Architectural discourse: Architectural discourse includes briefs, regulations, criticism, histories, classifications, and promotional language that shape buildings’ meaning and social effects.Architectural texts participate in the production, interpretation, and use of buildings rather than merely describing them.
- Research directions: Research on architectural language can combine corpus linguistics, discourse analysis, and studies of communication during BIM-based collaboration.Existing work examines architectural corpora, metaphorical discourse, and unstructured specialist communication around drawings and digital models.
- Programming languages: Programming languages translate human-readable commands into machine-readable binary strings while opening some possibilities and closing others.Their precise structures distinguish them from the messier communication of natural language, but they remain communicative and non-neutral tools.
- Annotations: ImageNet used WordNet’s noun hierarchy to organize more than 14 million images into categories, embedding assumptions about entities and their relations in AI training.WordNet organized approximately 170,000 words into around 117,000 synsets linked by semantic relations.
- Annotations: ImageNet’s categories demonstrate that apparently neutral annotation systems can encode social bias, including offensive slurs and psychiatric diagnoses.Crawford distinguishes more concrete nouns such as “apple” from more abstract or judgmental concepts such as “health.”
5. The nature of embodiment in language
Language theories frame communication as embodied, rule-governed, metaphorical, and connected to categorization. The chapter proposes further research on communication with computational systems and on how language-specific affordances shape architectural concepts.
- The nature of embodiment in language: Natural language is presented as embodied in human sensory experience, while linguistic theories describe both innate rules and differing individual experiences.Wittgenstein connects experience with the images formed from linguistic concepts; Chomsky emphasizes inherent rules and biological language capacity.
- Computational embodiment: Computational systems have forms of embodiment different from humans, so communication with them cannot be identical to communication between people.The chapter links this issue to AI systems’ inherited phenomenology from human-generated training data and calls for communication- and information-science research.
- Metaphor and categorization: Metaphor and prototype theory explain how humans understand complex concepts and organize categories around idealized mental representations.Prototype theory describes hierarchical categories such as furniture, chair, and armchair, with middle-level categories easier to image mentally.
- Embodied communication: Graphical and volumetric representations remain important for communicating precise architectural ideas that are difficult to express through natural language.This follows from the chapter’s discussion of spoken language as harder to produce for certain design concepts.
- Research directions: Future studies should examine architectural language with corpus and discourse methods, designer interaction with graphic-less interfaces, and differences among natural languages.The proposed methods include topic modelling, sentiment analysis, discourse analysis, and human-computer or human-robot interaction research.