Skip to main content
To KTH's start page To KTH's start page

Efficient Information Transmission and Audience Design in Spoken Communication

Seminar by Dr. Florian Jaeger

T. Florian Jaeger is a Professor in the Brain and Cognitive Science and Computer Science at the University of Rochester. His works covers speech perception, production, and adaptation, drawing on computational models to understand how we plan and process language, and how we use it to communicate. He received his training at the Humboldt & Technical Universities of Berlin, UC Berkeley, Stanford, MIT, and UC San Diego.

Time: Tue 2023-10-03 15.05 - 16.15

Location: Fantum, Room nr: 522, Floor: 05, Lindstedtsvägen 24

Video link: Attending Online (Zoom)

Language: English


Export to calendar


Florian Jaeger
Florian Jaeger Professor, Brain and Cognitive Sciences and Computer Science Profile +1 (585) 276-3611
University of Rochester 320 Meliora Hall 270268 Rochester USA


During typical conversations, human speech transmits about 3-5 syllables every second—coordinating the encoding of meaning into articulatory motor movements in split seconds. Remarkably, the rate of information transmission seems to be rather similar across languages of the world, despite massive differences in the linguistic systems (incl. their sound, word, sentence, and discourse structure). One explanation for this is that real-time language use and, ultimately, even the latent structure underlying languages (‘grammar’) are shaped by the goal to communicate efficiently—to make the most out of the noisy motor, perceptual, and neural systems that we are endowed with.

In this seminar, I’ll present some examples of how research in my lab has tried to approach this hypothesis. I’ll start with a brief overview of research on the trade-off between redundancy and reduction during spontaneous speech production (e.g., the decision to produce “do not” vs. “don’t” depending on how inferable the negation is in its linguistic context). In all phenomena we have studied, we find that more predictable meanings are produced with less clear signals. Then I’ll present some studies that investigate how such communicatively efficient trade-offs might emerge. This includes miniature language learning studies, in which we find that learners induce more efficiency into inefficient systems—i.e., language learners deviate from the input in ways that make the language more communicatively efficient. In another study, we investigated how communicative breakdowns affect subsequent speech production. Our findings suggest that speaker can target even subtle millisecond aspects of their speech so as to increase the expected intelligibility of their speech, while aiming to avoid unnecessary effort.