Elamur Research
☰
home / papers / 2023 / Evaluating Verifiability in Generative Search Engines

Evaluating Verifiability in Generative Search Engines

Written for the Research library (claude-opus-5-5) from the paper text. Original — aclanthology.org/2023.findings-emnlp.467.pdf

Summary

Generative search engines answer queries with synthesized text that contains in-line citations, but it is unclear whether those citations actually back up the claims. The authors call this property verifiability and audit four commercial systems (Bing Chat, NeevaAI, perplexity.ai and YouChat) through human evaluation. The queries come from varied sources, including historical Google searches from NaturalQuestions and open-ended questions collected from Reddit.

Annotators rated each response for fluency and perceived utility on Likert scales. They also checked sentence-level support to compute two proposed metrics, plus a combined citation F1:

  • Citation recall: the share of verification-worthy sentences that their citations fully back.
  • Citation precision: the share of citations that genuinely support their sentence.

The answers read smoothly and look helpful. However, only about half of the sentences are fully supported, and roughly one in four citations fails to support its claim. Moreover, the systems judged most useful tend to cite least accurately. Because these tools already reach millions of users, the study exposes a real trust gap and supplies metrics and data for building more reliable systems.

Key points

  • Average citation recall across the four engines is 51.5%, and average citation precision is 74.5%.
  • Perceived utility and citation precision are strongly negatively correlated across systems (r = −0.96).
  • Higher precision appears to come from copying or closely paraphrasing cited pages: BLEU and BERTScore similarity to the cited evidence correlates with precision at r = 0.80.
  • Statements are evaluated at the sentence level. Partially supporting citations count toward precision only in restricted cases, and finer-grained claim splitting is left for future work.
  • 34 annotators carried out the evaluation, and the annotations are released at github.com/nelson-liu/evaluating-verifiability-in-generative-search-engines.

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

Fields: Natural language processing

Cited by

157 papers of our library cite it (of the 140673 whose bibliographies we read); 9 of them have a page here, 148 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.findings-emnlp.467

@inproceedings{liu2023evaluating,
  title = {{Evaluating Verifiability in Generative Search Engines}},
  author = {Liu, Nelson and Zhang, Tianyi and Liang, Percy},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.findings-emnlp.467},
  url = {https://doi.org/10.18653/v1/2023.findings-emnlp.467}
}

References

26 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Michiel Bakker, Martin Chadwick, Hannah Sheahan, Michael Tessler, Lucy Campbell-Gillingham, Jan Balaguer, Nat McAleese, Amelia Glaese, John Aslanides, Matt Botvinick, and Christopher Summerfield. 2022. Fine-tuning language models to find agreement among humans with diverse preferences. In Proc. of N… find
  2. Abeba Birhane, Vinay Uday Prabhu, and John Whaley. 2022. Auditing saliency cropping algorithms. In Proc. of WACV. find
  3. Bernd Bohnet, Vinh Q. Tran, Pat Verga, Roee Aharoni, Daniel Andor, Livio Baldini Soares, Massimiliano Ciaramita, Jacob Eisenstein, Kuzman Ganchev, Jonathan Herzig, Kai Hui, Tom Kwiatkowski, Ji Ma, Jianmo Ni, Lierni Sestorain Saralegui, Tal Schuster, William W. Cohen, Michael Collins, Dipanjan Das, D… find our card
  4. Joy Buolamwini and Timnit Gebru. 2018. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Proc. of FAccT. find
  5. Sihao Chen, Daniel Khashabi, Wenpeng Yin, Chris Callison-Burch, and Dan Roth. 2019. Seeing things from a different angle: Discovering diverse perspectives about claims. In Proc. of NAACL. find
  6. Angela Fan, Yacine Jernite, Ethan Perez, David Grangier, Jason Weston, and Michael Auli. 2019. ELI5: Long form question answering. In Proc. of ACL. find our card
  7. Luyu Gao, Zhuyun Dai, Panupong Pasupat, Anthony Chen, Arun Tejasvi Chaganty, Yicheng Fan, Vincent Y. Zhao, Ni Lao, Hongrae Lee, Da-Cheng Juan, and Kelvin Guu. 2022. RARR: Researching and revising what language models say, using language models. ArXiv:2210.08726. find our card
  8. Amelia Glaese, Nat McAleese, Maja Tr˛ebacz, John Aslanides, Vlad Firoiu, Timo Ewalds, Maribeth Rauh, Laura Weidinger, Martin Chadwick, Phoebe Thacker, Lucy Campbell-Gillingham, Jonathan Uesato, Po-Sen Huang, Ramona Comanescu, Fan Yang, Abigail See, Sumanth Dathathri, Rory Greig, Charlie Chen, Doug F… find our card
  9. Ben Green and Yiling Chen. 2019. Disparate interactions: An algorithm-in-the-loop analysis of fairness in risk assessments. In Proc. of FAccT. find
  10. Svetlana Kiritchenko and Saif Mohammad. 2018. Examining gender and race bias in two hundred sentiment analysis systems. In Proc. of *SEM. find our card
  11. Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. 2019. Natural… find our card
  12. Faisal Ladhak, Esin Durmus, He He, Claire Cardie, and Kathleen McKeown. 2022. Faithful or extractive? on mitigating the faithfulness-abstractiveness trade-off in abstractive summarization. In Proc. of ACL. find
  13. Amin Ghafouri, Marcelo Menegali, Yanping Huang, Maxim Krikun, Dmitry Lepikhin, James Qin, Dehao Chen, Yuanzhong Xu, Zhifeng Chen, Adam Roberts, Maarten Bosma, Vincent Zhao, Yanqi Zhou, Chung-Ching Chang, Igor Krivokon, Will Rusch, Marc Pickett, Pranesh Srinivasan, Laichee Man, Kathleen Meier-Hellste… find our card
  14. Joshua Maynez, Shashi Narayan, Bernd Bohnet, and Ryan McDonald. 2020. On faithfulness and factuality in abstractive summarization. In Proc. of ACL. find
  15. Yusuf Mehdi. 2023. The new Bing and Edge – progress from our first month \| Bing search blog. Accessed on March 28, 2023. find
  16. Jacob Menick, Maja Trebacz, Vladimir Mikulik, John Aslanides, Francis Song, Martin Chadwick, Mia Glaese, Susannah Young, Lucy Campbell-Gillingham, Geoffrey Irving, and Nat McAleese. 2022. Teaching language models to support answers with verified quotes. ArXiv:2203.11147. find our card
  17. Danaë Metaxa, Joon Sung Park, James A. Landay, and Jeff Hancock. 2019. Search media and elections: A longitudinal investigation of political search results. In Proc. of CSCW. find
  18. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating text generation with BERT. In Proc. of ICLR. find
  19. Panagiotis Takis Metaxas and Yada Pruksachatkun. 2017. Manipulation of search engine results during the 2016 us congressional elections. find
  20. Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, Xu Jiang, Karl Cobbe, Tyna Eloundou, Gretchen Krueger, Kevin Button, Matthew Knight, Benjamin Chess, and John Schulman. 2021. WebGPT: Browser-assis… find our card
  21. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a method for automatic evaluation of machine translation. In Proc. of ACL. find our card
  22. Denis Peskoff and Brandon Stewart. 2023. Credible without credit: Domain experts assess generative language models. In Proc. of ACL. find
  23. Fabio Petroni, Aleksandra Piktus, Angela Fan, Patrick Lewis, Majid Yazdani, Nicola De Cao, James Thorne, Yacine Jernite, Vladimir Karpukhin, Jean Maillard, Vassilis Plachouras, Tim Rocktäschel, and Sebastian Riedel. 2021. KILT: a benchmark for knowledge intensive language tasks. In Proc. of NAACL. find our card
  24. Hannah Rashkin, Vitaly Nikolaev, Matthew Lamm, Lora Aroyo, Michael Collins, Dipanjan Das, Slav Petrov, Gaurav Singh Tomar, Iulia Turc, and David Reitter. 2022. Measuring attribution in natural language generation models. ArXiv:2112.12870. find our card
  25. Ronald E. Robertson, David Lazer, and Christo Wilson. 2018. Auditing the personalization and composition of politically-related search engine results pages. In Proc. of WWW. find
  26. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, find