Terminology Agnostic Query (TAQ)

• TAQ is a query framework that deliberately omits domain-specific terms, using paraphrased semantic descriptions to enable robust and flexible querying.
• TAQ methodologies are applied in information retrieval, mathematical knowledge management, and database systems, enhancing interoperability and conceptual retrieval.
• Empirical evidence shows TAQs improve retrieval performance and error guarantees, highlighting their practical benefits in heterogeneous system environments.

A Terminology Agnostic Query (TAQ) is a technical concept that appears across multiple domains, each time referring to a framework or method that enables querying, searching, or evaluating without dependence on specific terminology or surface forms. The key feature of TAQs is their abstraction from domain-specific or system-specific terms, enabling broader, more semantic access or understanding. This article surveys and contrasts TAQ methodologies as established in recent information retrieval benchmarks, formal mathematical query languages, approximate query processing, and higher algebraic homology.

1. Formal Definition and Fundamental Motivation

At its core, a Terminology Agnostic Query is a query formulation or query mechanism that deliberately avoids using the specific surface forms of technical or domain-specific terms. Instead, TAQs leverage structural, semantic, or descriptive devices so that they operate independently of idiosyncratic nomenclature, enhancing interoperability, robustness, and conceptual coverage.

• In information retrieval, a TAQ is a synthetic query that omits domain terms, relying exclusively on paraphrased or descriptive content to probe conceptual retrieval (Kim, 7 Jan 2026).
• In mathematical knowledge management, TAQ refers to queries that abstract over library-specific names, matching on a common semantic universe (typically as URIs in an MMT ontology) (Rabe, 2012).
• In database systems, a TAQ denotes an algorithmic approach that works independently of underlying DBMS specifics, schema, or internal statistics, issuing queries guaranteed to be compatible with any system supporting a minimal sampling interface (Zhu et al., 27 Mar 2025).

The universal motivation is to transcend superficial lexical dependencies and to provide a principled mechanism for interactions that are robust to variations in nomenclature, system, or context.

2. Domain-Specific Instantiations

2.1 Information Retrieval: TAQs in the STELLA Benchmark

Within the STELLA framework for aerospace-domain retrieval evaluation, Terminology Agnostic Queries are defined as:

• Synthetic, single-sentence queries that never include the surface form of any domain-specific technical term appearing in the target passage.
• Instead, TAQs inject concise descriptions of those terms, typically generated by prompting a LLM within local context windows.
• Formally, for candidate passage $p_{(c)}$ with technical term set $T_{(c)} = \{t_1, ..., t_k\}$ and LLM-generated term descriptions $D_{(c)} = \{d_1, ..., d_k\}$, the query $q_{(c)}$ satisfies $\forall i, t_i \notin q_{(c)}$ and is constructed by a Chain-of-Density process, incorporating the $d_i$ (Kim, 7 Jan 2026).

This process systematically disentangles lexical matching (surface overlap) from semantic matching (conceptual understanding) by pairing TAQs (term omitted, description inserted) with complementary TCQs (term present).

2.2 Mathematical Knowledge Management: TAQ in QMT

The QMT query language for formal mathematics realizes a terminology-agnostic mechanism by:

• Referring only to abstract concepts (e.g., “theory,” “constant,” “includes”) and their URIs as indexed by the underlying MMT ontology.
• Allowing queries that traverse, unify, and select objects throughout heterogeneous libraries, agnostic to specific naming conventions or concrete system syntax.
• E.g., a query to retrieve all group-theoretic constants across all imported libraries is achieved by inspecting type signatures, not by matching on the various library-specific spellings of “Group” (Rabe, 2012).

This approach enables unified search, cross-library unification, and extraction of structural patterns regardless of nomenclature diversity.

2.3 Database Systems: TAQA Algorithm in Online AQP

Within the context of approximate query processing, the TAQA (Terminology Agnostic Query Algorithm) is defined as:

• An online, two-stage algorithm for executing SQL aggregation queries with strong error guarantees, requiring no customization to schema, statistics, or DBMS extensions.
• TAQA plans and issues sampling-based queries using only standard SQL extensions (e.g., TABLESAMPLE), thus remaining compatible with any compliant DBMS (Zhu et al., 27 Mar 2025).
• The term "terminology agnostic" here refers to the system-agnosticism and independence from internal table structure, in contrast to other approaches that may rely on manual tuning or schema-specific knowledge.

3. Methodologies and Construction Pipelines

Information Retrieval (STELLA)

• Begins from large technical corpora (NASA NTRS).
• Pipeline: Document chunking → Terminology extraction → Candidate passage selection (with query intent classification) → TAQ construction by LLM-generated paraphrased term descriptions → Output synthetic, intent-compliant, terminology-omitting queries (Kim, 7 Jan 2026).
• Enforces strict “absolute term-ban policy”; all descriptions must be inferred, not copied, from context.

Mathematical QMT

• Abstract syntax over concepts and relations, supports first-order logic, set comprehensions, graph traversals, unification, and XQuery/SQL patterns.
• Queries traverse a common ontology over all libraries; operational at the level of theory graphs and term graphs, not system-specific representations (Rabe, 2012).

Database TAQA

• Two-stage online algorithm: Pilot query collects minimal statistics, enabling cost-minimizing sampling plans; the final query is rewritten with appropriate sampling rates and executed, guaranteeing user-specified error bounds with high probability.
• Remains DBMS-agnostic and does not depend on schema or terminology (Zhu et al., 27 Mar 2025).

4. Evaluation and Empirical Evidence

Information Retrieval

Empirical evaluation on the STELLA benchmark demonstrates that:

• Lexically-dependent methods (e.g., BM25) yield substantial nDCG@10 performance drops on TAQs relative to TCQs (gap: 0.228).
• State-of-the-art dense embedding models show reduced performance gaps: Llama-Embed-Nemotron (0.106), indicating stronger semantic or terminology-independent retrieval capability (Kim, 7 Jan 2026).
• Such evaluation directly measures a model’s capacity for conceptual retrieval when surface forms are deliberately inaccessible.

Model	TCQ nDCG@10	TAQ nDCG@10	Gap
BM25	0.773	0.545	0.228
Arctic-Embed-2.0-L (0.6B)	0.785	0.558	0.227
Qwen3-Embedding (8B)	0.779	0.608	0.171
Llama-Embed-Nemotron (8B)	0.841	0.735	0.106

Mathematical QMT

• Terminology-agnostic queries in QMT allow for population and traversal of objects across multiple libraries, supporting unification and retrieval without reliance on naming conventions (Rabe, 2012).
• Empirical scalability: indexing of thousands of theories in minutes, query resolution ranging from milliseconds (simple membership) to minutes (large object retrievals).

Database TAQA

• PilotDB implementation of TAQA delivers error guarantees always less than or equal to user targets, up to 126× speedup on analytic workloads (see TPC‑H, SSB, ClickBench, Instacart, DSB benchmarks).
• Remains robust across DBMSs, requiring no system-specific adaptation (Zhu et al., 27 Mar 2025).

5. Strengths, Limitations, and Open Problems

Strengths

• By construction, TAQs disentangle semantic from lexical or system-specific matching and thus act as rigorous probes of conceptual knowledge, retrieval, or system-compatibility (Kim, 7 Jan 2026, Rabe, 2012).
• Modular and extensible; applicable across domains (retrieval, semantic web, DBMS analytics).
• Empowers cross-system, cross-lingual, and cross-library workflows.

Limitations

• In IR: all TAQs are LLM-synthesized, introducing risk of style or phrasing bias; may miss real-user ambiguity or multi-hop logic (Kim, 7 Jan 2026).
• In mathematics: full terminology-agnosticism requires explicit symbol alignment between libraries (not automatic); equality theories may be limited to syntactic equivalence (Rabe, 2012).
• In DB systems: TAQA is agnostic to DBMS specifics, but query types or application domains not compatible with block-level sampling fall back to exact processing (Zhu et al., 27 Mar 2025).

Open Problems

• Automated ontology alignment for semantic equivalence discovery across formal libraries remains unresolved (Rabe, 2012).
• Enriching TAQs with support for multimodal (figure/table) and multi-hop queries is an emerging demand in technical IR (Kim, 7 Jan 2026).
• Extending practical, system-agnostic AQP to richer query types and higher-order aggregates is an active area (Zhu et al., 27 Mar 2025).

6. Connections to Related Model- and Terminology-Agnostic Frameworks

The conceptual underpinning of TAQ shares synergy with more general model-agnostic and formalism-agnostic frameworks. In ML interpretability, the SIPA (Sampling-Intervention-Prediction-Aggregation) framework unifies disparate black-box interpretation strategies by abstracting away the details of model type, system syntax, or pipeline-specific feature names, and distills interpretation into generic compositional stages (Scholbeck et al., 2019). This suggests that TAQ is an instantiation of a broader principle: robust abstraction from system or nomenclature, providing strong compositional and interoperability guarantees.

7. Summary and Significance

Terminology Agnostic Queries constitute a pivotal mechanism in modern information, knowledge, and data systems, explicitly designed to eliminate dependence on specific terms or system-internal identifiers. Across technical domains, they enable probing of true semantic content, interoperation across heterogeneous collections, and robust deployment of analytical tools. They also reveal weaknesses in models and systems overly reliant on term overlap. As scientific information, formalized knowledge, and distributed data platforms proliferate, the formulation, evaluation, and continued refinement of TAQ methodologies are likely to underpin the next generation of semantic retrieval, knowledge integration, and query processing infrastructures (Kim, 7 Jan 2026, Rabe, 2012, Zhu et al., 27 Mar 2025).