Are Multimodal Large Language Models Pragmatically Competent Listeners in Simple Reference Resolution Tasks?
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917054085332992 |
|---|---|
| author | Junker, Simeon Ali, Manar Koch, Larissa Zarrieß, Sina Buschmeier, Hendrik |
| author_facet | Junker, Simeon Ali, Manar Koch, Larissa Zarrieß, Sina Buschmeier, Hendrik |
| contents | We investigate the linguistic abilities of multimodal large language models in reference resolution tasks featuring simple yet abstract visual stimuli, such as color patches and color grids. Although the task may not seem challenging for today's language models, being straightforward for human dyads, we consider it to be a highly relevant probe of the pragmatic capabilities of MLLMs. Our results and analyses indeed suggest that basic pragmatic capabilities, such as context-dependent interpretation of color descriptions, still constitute major challenges for state-of-the-art MLLMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_11807 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Are Multimodal Large Language Models Pragmatically Competent Listeners in Simple Reference Resolution Tasks? Junker, Simeon Ali, Manar Koch, Larissa Zarrieß, Sina Buschmeier, Hendrik Computation and Language We investigate the linguistic abilities of multimodal large language models in reference resolution tasks featuring simple yet abstract visual stimuli, such as color patches and color grids. Although the task may not seem challenging for today's language models, being straightforward for human dyads, we consider it to be a highly relevant probe of the pragmatic capabilities of MLLMs. Our results and analyses indeed suggest that basic pragmatic capabilities, such as context-dependent interpretation of color descriptions, still constitute major challenges for state-of-the-art MLLMs. |
| title | Are Multimodal Large Language Models Pragmatically Competent Listeners in Simple Reference Resolution Tasks? |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2506.11807 |