TL;QR: Querying Temporal Discursive Patterns in Textual Corpora

François Hublet; Vladimir Krsmanovic; Jesus Solano; Ahmed Bouhoula; Ana Oliveira da Costa

cs.CLOct 11, 2026

TL;QR: Querying Temporal Discursive Patterns in Textual Corpora

Authors: François Hublet, Vladimir Krsmanovic, Jesus Solano, Ahmed Bouhoula, Ana Oliveira da Costa

Organizations: ETH Zürich · CISPA Helmholtz Center for Information Security · IST Austria

Abstract

Many questions about large textual corpora concern patterns that unfold over time, such as a political group whose stance on a topic drifts from support to criticism over several years of debates. We present TL;QR (Too Long; Query with logical temporal Reasoning), a declarative query language and open-source engine for finding such temporal discursive patterns. In TL;QR, a single query combines natural-language predicates, whose truth is judged by a Large Language Model (LLM), with regular expressions and the operators of a temporal logic. The LLM answers only local questions about individual text passages; a symbolic engine then aggregates these judgments over time and links every match to the recorded passages and predicate decisions behind it. With this division of labor, the temporal and relational reasoning stays exact and auditable even over long time horizons. In addition, when a query includes a regular-expression pre-filter, only a small fraction of the corpus ever reaches the LLM. On a suite of controlled synthetic benchmarks with temporally and semantically complex patterns, TL;QR is substantially more accurate than LLM-only baselines. In a case study on ten years of European Parliament debates (442,593 multilingual speeches), TL;QR identifies candidate shifts in the political groups' stances on the GDPR privacy regulation.

Figures & tables

Panel of Figure 1 — Figure 1: The TL;QR pipeline, illustrated on the query of Figure 2 over a hypothetical corpus with three groups. Each panel shows the intermediate predicate it generates: the regexp pre-filter turns on-topic speeches into $\mathtt{Snip}$ events, the LLM atoms label the resulting snippets positive or negative and emit $\mathtt{Pos}/\mathtt{Neg}$ , and the symbolic monitor derives the rolling counts $\mathtt{PosN}/\mathtt{NegN}$ and $\mathtt{MostlyPos}$ , and returns the detected stance shift . Rows are aligned across the first three panels, so a gap marks a speech the filter dropped. Only the LLM-atoms panel calls a model.

		F1 $\uparrow$ / MAE $\downarrow$
Method	LLM	1.5k	3k	4.5k	Cost
Stance shift
0-shot	luna	100.0 / 0.0	93.3 / 2.2	60.0 / 0.0	1 $\times$
	sol	100.0 / 0.0	93.3 / 0.0	100.0 / 0.0	25 $\times$
5-shot	luna	100.0 / 0.0	100.0 / 2.2	93.3 / 0.0	1 $\times$
	sol	100.0 / 0.0	100.0 / 0.0	100.0 / 0.0	25 $\times$

Table 1: Full results on the three synthetic benchmarks: F1 (%) and MAE (in timestamps), separated by a slash, mean over 3 seeds, for corpus sizes

n\in\{1.5\mathrm{k},3\mathrm{k},4.5\mathrm{k}\}

. Bold: best F1 per column/benchmark. † Recall

=0

: the method localised no true instance; precision is undefined, F1 collapses and MAE is omitted.

Predicate	$n$	Agr.	$\kappa$	Acc.	P	R	F1	$\kappa$
		IAA		gpt-5.6-luna vs. human
Relevance	50	100	1.0	96	1.0	.937	.967	.920
Positive stance	50	80	.598	75	.720	.958	.822	.600
Negative stance	50	96	.913	96	.680	1.0	.810	.680

Table 2: Human validation of the LLM predicates. Inter-annotator agreement (IAA) columns report raw agreement and Cohen’s

\kappa

between the two primary annotators. Accuracy, precision, recall, F1, and Cohen’s

\kappa

are computed with respect to the adjudicated gold labels on the stratified evaluation and reweighted based on class sizes.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Figure 5: GDPR stance over time by political group, as returned by the case-study query. Rug marks are individual supportive (blue) and critical (red) speeches; the shaded lines are the $180$ -day rolling counts computed by the monitor; the purple line is the resulting negative share with its one-half threshold; dashed orange lines mark detected stance shifts. Dotted grey lines are GDPR milestones. All series are read directly off the monitor’s output, so every point is traceable to the speeches that produced it.

Temporal Logic

See all 19 papers

STAIR: Semantic-Temporal Automaton for Interpretable Reasoning in Temporal Question Answering
Aug 17, 2026Xinlong Dai, Jinchuan Zhang, Lei Gao +3Temporal QA Temporal Reasoning in Language Models
University of Electronic Science and Technology of China (UESTC)
Time Present and Time Past: Benchmarking Large Language Models on Temporally Evolving Document Understanding
Aug 9, 2026Mahbub E Sobhani, Md. Faiyaz Abdullah Sayeedi, Fahmid Hasan Chowdhury +4LLM Evaluation Document QA
1BRAC University · 2United International University · 3North South University +1
Evaluating Temporal Reasoning in Large Language Models: Event Temporal Relation Extraction with Allen's Interval Algebra, Logical Consistency, and Relative Time in Narratives
Oct 12, 2026Adib Sakhawat, Fardeen Sadab, Tamjid Hasan Fahim +9Temporal Reasoning in Language Models Relation Extraction
Islamic University of Technology, Gazipur, Bangladesh · Rajshahi University of Engineering & Technology, Rajshahi, Bangladesh

Aug 17, 2026J/K move · Enter open · S save

STAIR: Semantic-Temporal Automaton for Interpretable Reasoning in Temporal Question Answering

Xinlong Dai, Jinchuan Zhang, Lei Gao, Xinzhe Hu+2

University of Electronic Science and Technology of China (UESTC)

Temporal QA Temporal Reasoning in Language Models Neuro-Symbolic Reasoning

Open paper AskPDF arXiv

By leveraging large-scale pretraining, LLMs can interpret diverse temporal expressions and question formulations without task-specific training. However, existing prompt-based neuro-symbolic systems continue to rely on LLMs for both semantic interpretation and exact temporal inference. Consequently, discrete decisions regarding intervals, time anchors, and ordered states remain vulnerable to probabilistic errors and difficult to verify. We present STAIR, a Semantic-Temporal Automaton for Interpretable Reasoning. STAIR separates semantic interpretation from precise temporal inference: an answer-free LLM adapter maps complex question formulations to normalized temporal intents, while a deterministic temporal automaton with finite control and guarded transitions executes the corresponding policies over canonicalized evidence. Following a rule-first design, STAIR resolves standard questions without invoking an LLM and applies semantic adaptation only when the rule path fails to produce an executable intent. This approach reduces free-form reasoning, making temporal decisions verifiable and interpretable. Specifically, guarded execution supports precise point-time containment and before/after selection, while semantic adaptation handles non-exact intervals and time-anchored queries. Across the TimeQA-Easy, TimeQA-Hard, TempReason-L2, and TempReason-L3 datasets, STAIR consistently outperforms strong baselines in the TQA task using matched model settings, achieving average F1 improvements of 16.57% and 3.10% when utilizing the Qwen2.5-7B and GPT-4o-mini models, respectively. Furthermore, ablations and diagnostic analyses demonstrate that STAIR excels at handling both boundary-sensitive and order-sensitive queries, while its guarded execution and semantic adaptation ensure precise point-time reasoning and inexact intervals, respectively.

Abstract

Temporal Logic

Explore similar work

STAIR: Semantic-Temporal Automaton for Interpretable Reasoning in Temporal Question Answering

Time Present and Time Past: Benchmarking Large Language Models on Temporally Evolving Document Understanding

Evaluating Temporal Reasoning in Large Language Models: Event Temporal Relation Extraction with Allen's Interval Algebra, Logical Consistency, and Relative Time in Narratives