---
title: "RAG as a collapsed NLG pipeline — Sivaprasad 2026"
description: "It builds a conceptual mapping between the two. 15 references. Among them: 0."
authors: ["Adarsa Sivaprasad", "Barkavi Sundararajan", "David M. Howcroft"]
year: 2026
venue: "Proceedings of the 1st Symposium on Natural Language Generation Evaluations"
source: "https://aclanthology.org/2026.retroeval-main.5/"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/sivaprasad-2026-rag-as-collapsed-nlg-pipeline/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sivaprasad-2026-rag-as-collapsed-nlg-pipeline/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/40/) / RAG as a collapsed NLG pipeline

# RAG as a collapsed NLG pipeline

**Authors:** Adarsa Sivaprasad, Barkavi Sundararajan and [David M. Howcroft](https://md.research.elamur.ai/free/en/authors/david-m-howcroft/) · **Year:** 2026 · **Venue:** [Proceedings of the 1st Symposium on Natural Language Generation Evaluations](https://md.research.elamur.ai/free/en/venues/retroeval/) ([Symposium on Natural Language Generation Evaluations 2026](https://md.research.elamur.ai/free/en/venues/retroeval/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.retroeval-main.5/) · [PDF](https://aclanthology.org/2026.retroeval-main.5.pdf) · [OpenAlex](https://openalex.org/W7216931428)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper asks how Retrieval-Augmented Generation (RAG) relates to the three-stage data-to-text architecture of Reiter and Dale (1997): document planning, microplanning and surface realisation. This is a short position paper with no experiments. It builds a conceptual mapping between the two. The classical reference is the rule-based marine forecast generator SumTime, and the contrast is a hypothetical RAG system that recommends activities for given weather. In the mapping, retrieval and reranking act as content selection, the system prompt spans document planning and microplanning, and the generator LLM serves as the surface realiser. The authors argue that RAG partly restores the modularity that end-to-end neural generation removed. However, each RAG component introduces risks the symbolic pipeline never faced:
- embeddings built without knowing future queries,
- probabilistic retrieval,
- prompt effects that depend on the specific model,
- conflicts between retrieved and parametric knowledge.

From this breakdown they derive stage-specific evaluation advice. The contribution is a diagnostic framework for locating where RAG systems fail, though the analogies have not been empirically tested.

## Key points
- **Mapping:**
  - Document planning is spread across retrieval, ranking and prompting.
  - Microplanning overlaps with both the prompt and the LLM.
  - Surface realisation happens only in the LLM generation step.
- **New failure modes:**
  - *Context faithfulness at the source* (term following Es et al., 2024, Ragas): offline, query-independent embeddings can be semantically ambiguous.
  - Non-deterministic retrieval: relevance scores do not ensure factual alignment.
  - Prompt sensitivity.
  - Retrieved-vs-parametric knowledge conflicts (Longpre et al., 2021).
- **Classical exemplars:**
  - SumTime, which uses bottom-up segmentation of wind readings and corpus-derived verbs like "backing" and "veering".
  - BT-45 for clinical summaries.
  - A comparison table sets SumTime against the weather-activity RAG scenario.
- **Proposed evaluation checklist:**
  - faithfulness checks on the knowledge source,
  - explicit retrieval evaluation,
  - prompt-strategy sensitivity tests (output stability, length, meaning preservation),
  - uncertainty estimates for context–parametric conflicts,
  - fine-grained annotation of errors, in the manner of Sundararajan et al. (2025) and the earlier Thomson and Reiter (2020).
- **Limitations:**
  - The analysis is purely conceptual and abstracts away input modality.
  - It reports no datasets or numbers.
  - Future work is to test few-shot/chain-of-thought prompting and pipeline-informed retrieval.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [University of Aberdeen](https://md.research.elamur.ai/free/en/universities/university-of-aberdeen/) — University, Aberdeen, United Kingdom · 20 works · 1,385 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [✍️ Text generation](https://md.research.elamur.ai/free/en/topics/text-generation/) · 202 works

## Cite

Cite as: Adarsa Sivaprasad, Barkavi Sundararajan and David M. Howcroft. 2026. RAG as a collapsed NLG pipeline. Proceedings of the 1st Symposium on Natural Language Generation Evaluations.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{sivaprasad2026rag,
  title = {{RAG as a collapsed NLG pipeline}},
  author = {Adarsa Sivaprasad, Barkavi Sundararajan and David M. Howcroft},
  booktitle = {Proceedings of the 1st Symposium on Natural Language Generation Evaluations},
  year = {2026},
  url = {https://aclanthology.org/2026.retroeval-main.5/}
}
```

- [← previous RadTimeline: Timeline Summarization for Longitudinal Radiological Lung Findings](https://md.research.elamur.ai/free/en/papers/zhou-2026-radtimeline-timeline-summarization-longitudinal-radiological/)
- [next → RAG or Learning? Understanding the Limits of LLM Adaptation under Continuous Knowledge Drift in the Real World](https://md.research.elamur.ai/free/en/papers/liu-2026-rag-or-learning-understanding-limits/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/40/)

## References

15 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ratish Puduppully, Li Dong, and Mirella Lapata. 2019. Data-to-text generation with content selection and planning. In Proc. of the AAAI Conference on Artificial Intelligence, volume 33, pages 6908–6915. [find](https://scholar.google.com/scholar?q=Ratish+Puduppully%2C+Li+Dong%2C+and+Mirella+Lapata%2E+2019%2E+Data%2Dto%2Dtext+generation+with+content+selection+and+planning%2E+In+Proc%2E+of+the+AAAI+Conference+on+Artificial+Intelligence%2C+volume+33%2C+pages+6908%E2%80%93691)
2. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. [Language models are few-shot learners](https://md.research.elamur.ai/free/en/papers/brown-2020-language-models-few-shot-learners/). Advances in Neural Information Processing Systems (NeurIPS), 33:1877–1901.
3. Ehud Reiter. 2007. An architecture for data-to-text systems. In Proc. of the 11th European Workshop on Natural Language Generation (ENLG), pages 97– 104. [find](https://scholar.google.com/scholar?q=Ehud+Reiter%2E+2007%2E+An+architecture+for+data%2Dto%2Dtext+systems%2E+In+Proc%2E+of+the+11th+European+Workshop+on+Natural+Language+Generation+%28ENLG%29%2C+pages+97%E2%80%93+104%2E)
4. Shahul Es, Jithin James, Luis Espinosa Anke, and Ehud Reiter. 2025. Springer. [find](https://scholar.google.com/scholar?q=Shahul+Es%2C+Jithin+James%2C+Luis+Espinosa+Anke%2C+and+Ehud+Reiter%2E+2025%2E+Springer%2E)
5. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). Advances in Neural Information Processing Systems (NeurIPS), 30.
6. Ehud Reiter and Robert Dale. 1997. Building applied natural language generation systems. Natural Language Engineering, 3(1):57–87. [find](https://scholar.google.com/scholar?q=Ehud+Reiter+and+Robert+Dale%2E+1997%2E+Building+applied+natural+language+generation+systems%2E+Natural+Language+Engineering%2C+3%281%29%3A57%E2%80%9387%2E)
7. Sander Schulhoff, Michael Ilie, Nishant Balepur, Konstantine Kahadze, Amanda Liu, Chenglei Si, Yinheng Li, Aayush Gupta, HyoJung Han, Sevien Schulhoff, and 1 others. 2024. [The prompt report: A systematic survey of prompt engineering techniques](https://md.research.elamur.ai/free/en/papers/schulhoff-2024-prompt-report-systematic-survey-prompt/). Preprint, arXiv:2406.06608.
8. Kurt Shuster, Spencer Poff, Moya Chen, Douwe Kiela, and Jason Weston. 2021. [Retrieval augmentation reduces hallucination in conversation](https://md.research.elamur.ai/free/en/papers/shuster-2021-retrieval-augmentation-reduces-hallucination-conversation/). In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 3784–3803.
9. Somayajulu Sripada. 2003. SumTime-Mousam: Configurable marine weather forecast generator. Expert Update, 6(3):4–10. [find](https://scholar.google.com/scholar?q=Somayajulu+Sripada%2E+2003%2E+SumTime%2DMousam%3A+Configurable+marine+weather+forecast+generator%2E+Expert+Update%2C+6%283%29%3A4%E2%80%9310%2E)
10. Somayajulu Sripada, Ehud Reiter, Jim Hunter, and Jin Yu. 2002. Sumtime-meteo: Parallel corpus of naturally occurring forecast texts and weather data. Technical report, Computing Science Department, University of Aberdeen, Aberdeen, Scotland, Tech. Rep. AUCS/TR0201. [find](https://scholar.google.com/scholar?q=Somayajulu+Sripada%2C+Ehud+Reiter%2C+Jim+Hunter%2C+and+Jin+Yu%2E+2002%2E+Sumtime%2Dmeteo%3A+Parallel+corpus+of+naturally+occurring+forecast+texts+and+weather+data%2E+Technical+report%2C+Computing+Science+Department%2C+Un)
11. Somayajulu G. Sripada, Ehud Reiter, Ian Davy, and Kristian Nilssen. 2004. Lessons from deploying nlg technology for marine weather forecast text generation. In Proc. of the 16th European Conference on Artificial Intelligence, ECAI’04, page 760–764, NLD. IOS Press. [find](https://scholar.google.com/scholar?q=Somayajulu+G%2E+Sripada%2C+Ehud+Reiter%2C+Ian+Davy%2C+and+Kristian+Nilssen%2E+2004%2E+Lessons+from+deploying+nlg+technology+for+marine+weather+forecast+text+generation%2E+In+Proc%2E+of+the+16th+European+Conference+on)
12. Somayajulu G Sripada, Ehud Reiter, Jim Hunter, and Jin Yu. 2003. Generating english summaries of time series data using the gricean maxims. In Proc. of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining, pages 187– 196. [find](https://scholar.google.com/scholar?q=Somayajulu+G+Sripada%2C+Ehud+Reiter%2C+Jim+Hunter%2C+and+Jin+Yu%2E+2003%2E+Generating+english+summaries+of+time+series+data+using+the+gricean+maxims%2E+In+Proc%2E+of+the+ninth+ACM+SIGKDD+international+conference+on)
13. Barkavi Sundararajan, Somayajulu Sripada, and Ehud Reiter. 2025. Input matters: Evaluating input structure’s impact on LLM summaries of sports play-by-play. In Proc. of the 18th International Natural Language Generation Conference, pages 795–809, Hanoi, Vietnam. Association for Computational Linguis… [find](https://scholar.google.com/scholar?q=Barkavi+Sundararajan%2C+Somayajulu+Sripada%2C+and+Ehud+Reiter%2E+2025%2E+Input+matters%3A+Evaluating+input+structure%E2%80%99s+impact+on+LLM+summaries+of+sports+play%2Dby%2Dplay%2E+In+Proc%2E+of+the+18th+International+Natural+)
14. Ilya Sutskever, Oriol Vinyals, and Quoc V Le. 2014. [Sequence to sequence learning with neural networks](https://md.research.elamur.ai/free/en/papers/sutskever-2014-sequence-sequence-learning-neural-networks/). Advances in Neural Information Processing Systems (NeurIPS), 27.
15. Craig Thomson and Ehud Reiter. 2020. A gold standard methodology for evaluating accuracy in data-to-text systems. In Proc. of the 13th International Conference on Natural Language Generation (INLG), pages 158–168, Dublin, Ireland. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Craig+Thomson+and+Ehud+Reiter%2E+2020%2E+A+gold+standard+methodology+for+evaluating+accuracy+in+data%2Dto%2Dtext+systems%2E+In+Proc%2E+of+the+13th+International+Conference+on+Natural+Language+Generation+%28INLG%29%2C+p)

*Updated 1 November 2026*