Abstract
Generating grounded image descriptions requires associating linguistic units with their corresponding visual clues. A common method is to train a decoder language model with attention mechanism over convolutional visual features. Attention weights align the stratified visual features arranged by their location with tokens, most commonly words, in the target description. However, words such as spatial relations (e.g. next to and under) are not directly referring to geometric arrangements of pixels but to complex geometric and conceptual representations. The aim of this paper is to evaluate what representations facilitate generating image descriptions with spatial relations and lead to better grounded language generation. In particular, we investigate the contribution of three different representational modalities in generating relational referring expressions: (i) pre-trained convolutional visual features, (ii) different top-down geometric relational knowledge between objects, and (iii) world knowledge captured by contextual embeddings in language models.- Anthology ID:
- W19-8668
- Volume:
- Proceedings of the 12th International Conference on Natural Language Generation
- Month:
- October–November
- Year:
- 2019
- Address:
- Tokyo, Japan
- Editors:
- Kees van Deemter, Chenghua Lin, Hiroya Takamura
- Venue:
- INLG
- SIG:
- SIGGEN
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 540–551
- Language:
- URL:
- https://aclanthology.org/W19-8668
- DOI:
- 10.18653/v1/W19-8668
- Cite (ACL):
- Mehdi Ghanimifard and Simon Dobnik. 2019. What goes into a word: generating image descriptions with top-down spatial knowledge. In Proceedings of the 12th International Conference on Natural Language Generation, pages 540–551, Tokyo, Japan. Association for Computational Linguistics.
- Cite (Informal):
- What goes into a word: generating image descriptions with top-down spatial knowledge (Ghanimifard & Dobnik, INLG 2019)
- PDF:
- https://preview.aclanthology.org/nschneid-patch-1/W19-8668.pdf