Douglas W. Oard’s research while affiliated with University of Maryland Global Campus and other places

... However, since documents and queries can be generated by different types of agents, the original generative assumption, and its conceptual generalization just proposed, do not hold. This state-ofaffairs is reminiscent of that in cross-lingual retrieval where queries and documents are written in different languages [26,43]. Accordingly, re-visiting the generative theory to relevance is an interesting future direction to explore in multi-agent retrieval settings. ...

Reference:

A Multi-Agent Perspective on Modern Information Retrieval

... The recent emergence of large language models (LLMs) has enabled automated, reliable nuggetbased evaluation (Pradeep et al., 2024;Alaofi et al., 2024;Pradeep et al., 2025b;Abbasiantaeb et al., 2025). Several RAG evaluation frameworks-such as FactScore (Min et al., 2023), RUBRIC (Farzi and Dietz, 2024), and others Mayfield et al., 2024)-incorporate the nugget concept, although most of these proposed approaches are either not validated or primarily validated on traditional ad hoc retrieval, and hence their applicability to longform answers is unclear. We refer readers to Pradeep et al. (2025b) for a more detailed discussion of related work. ...

Reference:

Chatbot Arena Meets Nuggets: Towards Explanations and Diagnostics in the Evaluation of LLM Responses

... In contrast, there is little work on neural dual encoders in such a streaming scenario. Recent work by Lawrie et al. [21] investigated incrementally indexing a large corpus using PLAID-X [40] with a fixed model, but only evaluated on a fixed set of queries at the end of the stream. How the model performs during and across the stream remains unknown. ...

Reference:

MURR: Model Updating with Regularized Replay for Searching a Document Stream

... A parallel direction to our approach is recent literature leveraging distillation techniques. Yang et al. [40,39] recently proposed Translate-Distill, where a strong cross-encoder is leveraged as a teacher to distil its knowledge more efficient biencoder models in both a CLIR and MLIR setting as an alternative to translatetrain. Zhuang et al. [45] propose to augment the documents' representations with queries generated using mT5 in languages other than the original document language to mitigate the data scarcity of CLIR. ...

Reference:

Improving Low-Resource Retrieval Effectiveness using Zero-Shot Linguistic Similarity Transfer

... Recent advancements focus on end-to-end retrieval systems that eliminate translation dependencies [19,22], improving scalability in multilingual settings [17]. Techniques such as language model pretraining [39], data curation and translation [22,23], and knowledge distillation [19,37,38] have enhanced the effectiveness of neural retrieval models [16,37]. ...

Reference:

MMMORRF: Multimodal Multilingual Modularized Reciprocal Rank Fusion

... In recent years, there has also been a growing interest in evaluating cross-lingual information retrieval. HC4 [15] and HC3 [16] were recently proposed to benchmark IR methods when dealing with Chinese, Persian, and Russian documents. Similarly, the TREC NeuCLIR track [13,14] released both NeuCLIR 1, a collection of Persian, Chinese and Russian Common Crawl News documents as well as neuMARCO, a cross-language machine-translated version of MSMARCO to study the effect of neural approaches to CLIR. ...

Reference:

Improving Low-Resource Retrieval Effectiveness using Zero-Shot Linguistic Similarity Transfer

... HC4 [15] and HC3 [16] were recently proposed to benchmark IR methods when dealing with Chinese, Persian, and Russian documents. Similarly, the TREC NeuCLIR track [13,14] released both NeuCLIR 1, a collection of Persian, Chinese and Russian Common Crawl News documents as well as neuMARCO, a cross-language machine-translated version of MSMARCO to study the effect of neural approaches to CLIR. Both of these collections are now standard benchmarks for evaluating the cross-language capabilities of IR methods. ...

Reference:

Improving Low-Resource Retrieval Effectiveness using Zero-Shot Linguistic Similarity Transfer

... Within academic institutions, language technologies can facilitate seamless knowledge retrieval from diverse sources and languages, mitigating cultural and technical barriers (Litschko et al., 2022). Consequently, AI-based programming for information retrieval is essential in advancing comprehension and development within multilingual environments (Lawrie et al., 2023). Furthermore, AI-driven information retrieval systems significantly contribute to the efficiency and accuracy of knowledge dissemination. ...

Reference:

Digital Transformation in ASEAN Humanities: Integration of AI and Language Technologies for Multilingual Information Retrieval

... However, those who focus on the adaptation of pre-existing schemes, proposed, in essence, modalities of application of the principle of cultural hospitality. As Choi et al. (2022) point out, cultural hospitality constitutes "an approach to improve information systems by providing ethical resource descriptions and access" (554). In this way, "cultural hospitality refers to the ability of a system to connect existing knowledge with perspectives, expectations, and assumptions from different cultures and users" (554). ...

Reference:

Native Peoples and Knowledge Organization: Perspective from the Indigenous Subject Representation to Promote Latin American Approaches

... Although financial incentives can help with recruitment, they can also increase the likelihood of fraud (Bowen et al., 2008;Rogers et al., 2022). If the study design includes a financial reimbursement for participation, consider methods for compensating people that would make it more difficult or impossible for fraudsters or bots. ...

Reference:

Management of fraudulent participants in online research: Practical recommendations from a randomized controlled feasibility trial