# From phonemes to images: levels of representation in a recurrent neural model of visually-grounded language learning

arXiv: https://arxiv.org/abs/1610.03342 · HTML: https://mlsanity.com/paper/1610.03342/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 11 Oct 2016.

## Built on by (most cited first, 8 of 8)

- [Visually grounded models of spoken language: A survey of datasets, architectures and evaluation techniques](https://mlsanity.com/paper/2104.13225/) (2021)
- [Word Segmentation on Discovered Phone Units with Dynamic Programming and Self-Supervised Scoring](https://mlsanity.com/paper/2202.11929/) (2022)
- [Talk, Don't Write: A Study of Direct Speech-Based Image Retrieval](https://mlsanity.com/paper/2104.01894/) (2021)
- [Probing artificial neural networks: insights from neuroscience](https://mlsanity.com/paper/2104.08197/) (2021)
- [Attention-Based Keyword Localisation in Speech using Visual Grounding](https://mlsanity.com/paper/2106.08859/) (2021)
- [Keyword localisation in untranscribed speech using visually grounded speech models](https://mlsanity.com/paper/2202.01107/) (2022)
- [Towards visually prompted keyword localisation for zero-resource spoken languages](https://mlsanity.com/paper/2210.06229/) (2022)
- [Exploring How Generative Adversarial Networks Learn Phonological Representations](https://mlsanity.com/paper/2305.12501/) (2023)

## Abstract (arXiv)

We present a model of visually-grounded language learning based on stacked gated recurrent neural networks which learns to predict visual features given an image description in the form of a sequence of phonemes. The learning task resembles that faced by human language learners who need to discover both structure and meaning from noisy and ambiguous data across modalities. We show that our model indeed learns to predict features of the visual context given phonetically transcribed image descriptions, and show that it represents linguistic information in a hierarchy of levels: lower layers in the stack are comparatively more sensitive to form, whereas higher layers are more sensitive to meaning.