Pith. sign in

REVIEW 3 major objections 5 minor 62 references

Multi-Modal Semantic Parsing for the Interpretation of Tombstone Inscriptions

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a fine-tuned vision-language model with mid-pipeline retrieval parses photographs of tombstones into structured semantic graphs, lifting Smatch F1 from 36.1 for the previous OCR pipeline to 89.5 on a held-out test…

desk verdict A credible 89.5 Smatch result for VLM-based tombstone parsing; needs variance and annotation-quality checks, but deserves peer review. read the letter →

arxiv 2507.04377 v3 pith:4MRCNJCK submitted 2025-07-06 cs.CV cs.CLcs.MM

Xiao Zhang , Johan Bos This is my paper
classification cs.CVcs.CLcs.MM
keywords tombstoneparsingsemanticvision-languagemodelsretrieval-augmentedgenerationdigitalheritageMeaningRepresentationSmatchcemeteryresearch
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a single fine-tuned vision-language model, helped by external lookups, can take a photograph of a tombstone and output a structured, searchable record of the person buried there, their dates, places, family roles, and occupation. On the authors' 1,200-image corpus, the best configuration reaches 89.5 Smatch F1, up from 36.1 for the previous OCR-based pipeline, which is the number the paper leads with. The reason the jump matters is that tombstones are eroding and being vandalized, and a direct image-to-structure route makes large-scale digitization feasible without brittle OCR stages. The paper also shows which design choice carries the result: retrieval inserted in the middle of generation, after the model drafts an initial structure, beats retrieval before or after.

What carries the argument

The central object is the Tombstone Meaning Representation (TMR), a rooted directed acyclic graph in PENMAN notation that encodes the deceased, their relations, dates, places, occupations, and roles, using WordNet synsets for concepts and literal identifiers for external codes such as GeoNames and HISCO. The carrying mechanism is the Tomb2Meaning (T2M) framework: a vision-language model converts the image into a TMR, and a retrieval component supplies arbitrary external identifiers. In the best-performing variant, RimAG, the model first generates a draft TMR, extracts entity names from it, queries GeoNames, HISCO, and WordNet, and then regenerates with the retrieved codes added to the prompt, so the external knowledge is grounded in what the model has already read.

What would settle it

Take a new set of tombstone photographs from a different region and language, have two independent annotators write TMRs for each stone, and compare the model's Smatch F1 against each annotator and against inter-annotator agreement; if the model matches human annotators as closely as humans match each other, the result generalizes, and if it does not, the high score is an artifact of the single-annotator corpus.

Watch

Extended reading notes

Core claim

The paper's central claim is that a fine-tuned vision-language model, augmented by retrieval in the middle of its generation, parses photographs of tombstones into structured semantic graphs better than any tested alternative: 89.50 Smatch F1 on the held-out test half, against 36.13 for the YOLO-OCR pipeline and 66.40 for the VLM-OCR baseline. The gain concentrates in external fields that cannot be read from pixels, with geographic identifier F1 rising from 10.83 to 66.58 and HISCO occupation codes from 0.00 to 60.33, while internal fields such as names, roles, and dates also improve. The authors interpret this as evidence that VLMs can replace brittle OCR pipelines for tombstone digitization, provided external knowledge is retrieved and injected rather than memorized.

Load-bearing premise

The gold-standard TMR annotations used for training and testing are treated as correct without any reported measure of agreement between independent annotators, so the 89.5 figure may measure fidelity to this specific corpus rather than general skill at reading tombstones.

Editorial extensions

If this is right

  • If the reported accuracy holds, OCR-based pipelines for tombstone interpretation can be replaced by a single fine-tuned VLM with retrieval, removing the need for segmentation, OCR, and rule-based post-correction.
  • RimAG's consistent superiority across model sizes implies that external knowledge should be injected mid-generation, after the model has read the image and drafted a structure, rather than before or after.
  • The retrieval components give each parsed tombstone stable, verifiable external identifiers, making the output directly storable in relational databases and queryable by historians.
  • The robustness results under simulated noise and the strong scores on rare fonts, rare languages, abbreviations, and coreference suggest the framework tolerates real-world damage better than OCR baselines.
  • Fine-tuning contributes the main jump from prompting-level performance to the low-to-mid 80s, so practical deployment requires per-corpus fine-tuning rather than pure prompting.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the gold annotations contain systematic choices by a single annotator, the 89.5 figure on this corpus may not transfer to other cemeteries; an inter-annotator agreement study would reveal the ceiling.
  • Because all retrieval variants pick GeoNames matches using GPS coordinates stored in image metadata, non-geotagged photographs would lose most of the geographic gain; a visual geolocation step would be a natural extension.
  • The same retrieval-in-the-middle recipe — draft a structure, look up arbitrary identifiers, regenerate — could transfer to other heritage artifacts such as memorial plaques and archival inscriptions, where canonical codes are unlearnable from pixels.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes Tomb2Meaning (T2M), a framework that combines vision-language models (VLMs) with retrieval-augmented generation (RAG) to parse photographs of tombstones into structured Tombstone Meaning Representations (TMRs). Three RAG integration strategies are introduced: RibAG (retrieval before generation), RimAG (retrieval between initial and final generation), and RieAG (retrieval-based replacement at the end). The authors evaluate the framework on a corpus of 1,200 tombstone images with five VLM backbones under few-shot prompting and LoRA fine-tuning, comparing against a traditional OCR pipeline (YOLO-OCR), a VLM-based OCR baseline, and a deterministic baseline. The headline result is an improvement from 36.13 Smatch F1 (YOLO-OCR) to 89.50 Smatch F1 (Qwen2.5-VL-7B with RimAG). Additional experiments address linguistic challenges (rare languages, fonts, coreference, abbreviations, multiple persons) and robustness to simulated image degradation. The paper also briefly states limitations, including the mainly Dutch dataset and dependence on structured external resources.

Significance. If the reported results are valid, this paper offers a practical and effective approach for automated heritage digitization, and it introduces a new multimodal semantic parsing task with a formal meaning representation (TMR) and an accompanying corpus. The empirical study is comparatively broad: five VLM families, three retrieval strategies, few-shot and fine-tuned settings, per-field F1 metrics, and a deterministic baseline that anchors the score scale. The authors also promise to release code and supplementary materials, which supports reproducibility. However, the evaluation rests on the quality of the authors' own TMR annotations, for which no inter-annotator agreement is reported, and part of the RAG improvements come from GPS metadata rather than from interpreting the inscription. These issues are material to the credibility of the 89.5 headline number, but they are arguably fixable with additional experiments and reporting.

major comments (3)
  1. [Section 4.1, Section 4.3, Figure 3] The paper does not report a separate validation set or a clear model-selection procedure. Figure 3 varies the number of few-shot examples from 0 to 6, and fine-tuning hyperparameters are described only as 'provided in supplemental materials.' If the best number of shots or LoRA settings were chosen by looking at test-set performance, the reported numbers, including the headline 89.50, may be optimistically biased. Please specify how hyperparameters and prompt exemplars were selected, report a validation split, and ideally include error bars or significance tests over multiple seeds.
  2. [Section 3.2, Table 2] The retrieval module selects GeoNames identifiers using GPS coordinates from image metadata, and the test images contain such metadata (Section 4.1). This means that the large F1-Geo gains (e.g., from 10.83 for base T2M to 66.58 for RimAG on Qwen2.5-VL-7B) are partly a lookup from EXIF data rather than a semantic interpretation of the inscription. The OCR baselines do not use this metadata, making the comparison less apples-to-apples than the abstract suggests. The authors should report an ablation that disables GPS and relies only on textual matching, and they should discuss how the framework would perform when GPS is unavailable.
  3. [Section 4.1, Table 2] The gold-standard TMR annotations are the authors' own, and no inter-annotator agreement or detailed corpus statistics are reported. Because the models are fine-tuned and evaluated on this same corpus, the 89.5 Smatch measures fidelity to one annotator's corpus, and the extent to which this transfers to general tombstone understanding is unknown. The deterministic baseline already achieves 30.17 Smatch, indicating that a fixed template captures a substantial portion of the score; this makes the effective difficulty of the benchmark unclear. Please provide inter-annotator agreement (for example, Smatch between annotators), annotation guidelines, and per-field annotation statistics.
minor comments (5)
  1. [Figure 7 and Figure 6] The x-axis label in Figure 7 contains a typo: '/glyph1197oise Level' should be 'Noise Level', and the caption of Figure 6 says 'Orginal' instead of 'Original'. Please correct these.
  2. [Section 6.1, Figure 5] The results on the five challenge subsets (rare languages, rare fonts, coreference, abbreviations, multiple persons) are only presented in a line chart without exact values, subset sizes, or inclusion criteria. Please provide these numbers in a table or in the supplementary material so that readers can judge the robustness claims quantitatively.
  3. [Table 2] All reported F1 scores are single numbers without variance or significance tests. Given that some differences between RAG variants are small (for example, RibAG vs. RimAG on Llama-3.2-vision-11B), the paper would benefit from standard deviations across multiple runs or a significance test to support the claim that RimAG consistently outperforms the alternatives.
  4. [Abstract and Section 7] The abstract claims robustness 'across diverse linguistic and cultural inscriptions,' but the dataset is mostly Dutch with a small number of non-Dutch and multilingual instances. Since the authors themselves list this as a limitation in Section 7, the abstract and Section 6 should be worded more cautiously, or the paper should report the sizes of the non-Dutch subsets and the results on them separately.
  5. [Section 2.2 and Introduction] The introduction states that this is 'the first attempt to formalize tombstone understanding using large vision-language models.' Given the recent surge in multimodal document understanding and heritage-related VLM work, it would be helpful to discuss a few such works explicitly to support the novelty claim and position the contribution more precisely.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline result is a genuine held-out evaluation against an external metric, with retrieval drawn from independent resources; self-citations are provenance, not load-bearing reductions.

full rationale

The paper's central claim, that RimAG reaches 89.50 Smatch versus 36.13 for YOLO-OCR, is a measured result on a held-out test split of the authors' corpus using Smatch, an external graph-matching metric (Cai and Knight, 2013), with no test-set parameter fitting. The retrieval-augmented components query GeoNames, HISCO, and WordNet, which are independent external knowledge bases rather than quantities derived from the model's own outputs. The TMR formalism and the manually annotated corpus originate in the authors' prior work (Bos et al. 2022; Bos 2022), but those are published annotation standards and datasets, not predicted values; citing them is normal scientific provenance rather than circular reasoning. The GPS-based selection of GeoNames codes is a direct lookup from image metadata, which explains the large F1-Geo gains, but it is transparent use of external metadata and is not a fitted parameter renamed as a prediction. The absence of inter-annotator agreement and the in-distribution split are annotation-validity and generalization concerns, not circularity under the definitions used here. No equation in the paper reduces to its own input, and no load-bearing claim is justified solely by a self-citation.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The system's measurements rest on the authors' own corpus and annotations, GPS metadata, and three external knowledge bases; none of these are validated within the paper, but all are public or independently maintained. The LoRA and noise settings are experimenter choices reported in supplemental materials rather than quantities fitted to the test set. Nothing in the central claim depends on a newly postulated physical or semantic entity.

free parameters (4)
  • LoRA rank/alpha and learning rate = not stated in main text (supplemental)
    Parameter-efficient fine-tuning configuration; chosen by the authors, reported only in supplemental materials. Affects all fine-tuned results in Table 2.
  • Noise levels for image fusion (low/medium/high) = defined only via supplemental materials
    Hand-chosen alpha-blending settings used for the robustness experiment in Section 6.2; results depend qualitatively on these cutoffs.
  • Number of few-shot examples (0 through 6) = 0 through 6
    Experimental sweep in Figure 3; the saturation behavior is a finding, but the specific shots are experimenter choices.
  • 50:50 train/test split seed = not reported
    Random split of 1,200 images; without a reported seed the exact split is unrepeatable, though the conclusions are unlikely to depend on it.
assumptions (5)
  • domain assumption The gold TMR annotations are correct and consistent enough to serve as ground truth for Smatch and F1 evaluation.
    Section 4.1 states images were 'manually annotated with a formal meaning representation following the TMR conventions'. No inter-annotator agreement is reported; the corpus is the authors' own (ref [11]).
  • domain assumption GPS coordinates in the image metadata are available and accurate for geographic disambiguation.
    Section 3.2: 'we leverage the GPS coordinates of the tombstone images to select the geographically closest match' for GeoNames codes. Deployment without GPS would lose most of the measured F1-Geo gains.
  • standard math Smatch with hill-climbing is a valid measure of semantic graph similarity.
    Section 4.2 adopts Smatch from Cai and Knight [16]; standard practice in AMR parsing, though hill-climbing is stochastic and typically needs multiple restarts.
  • domain assumption GeoNames, HISCO, and WordNet are reliable and complete enough for the entities appearing on these tombstones.
    The retrieval module depends on these external databases (Section 3.2); if a toponym or occupation is absent from the database, the RAG integration fails for that field.
  • domain assumption The generated TMRs are structurally close enough that RieAG's regular-expression replacement aligns correctly.
    RieAG replaces sections of TMR0 via 'a simple regular expression' (Section 3.2); this assumes the initial graph is well formed, consistent with the IFR numbers shown for the base model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Modal Semantic Parsing for the Interpretation of Tombstone Inscriptions." pith.science (2026). https://pith.science/paper/4MRCNJCK

@misc{pith2026250704377,
  author       = {Pith},
  title        = {Pith review of: Multi-Modal Semantic Parsing for the Interpretation of Tombstone Inscriptions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4MRCNJCK}},
  note         = {Machine review of arXiv:2507.04377}
}
read the original abstract

Tombstones are historically and culturally rich artifacts, encapsulating individual lives, community memory, historical narratives and artistic expression. Yet, many tombstones today face significant preservation challenges, including physical erosion, vandalism, environmental degradation, and political shifts. In this paper, we introduce a novel multi-modal framework for tombstones digitization, aiming to improve the interpretation, organization and retrieval of tombstone content. Our approach leverages vision-language models (VLMs) to translate tombstone images into structured Tombstone Meaning Representations (TMRs), capturing both image and text information. To further enrich semantic parsing, we incorporate retrieval-augmented generation (RAG) for integrate externally dependent elements such as toponyms, occupation codes, and ontological concepts. Compared to traditional OCR-based pipelines, our method improves parsing accuracy from an F1 score of 36.1 to 89.5. We additionally evaluate the model's robustness across diverse linguistic and cultural inscriptions, and simulate physical degradation through image fusion to assess performance under noisy or damaged conditions. Our work represents the first attempt to formalize tombstone understanding using large vision-language models, presenting implications for heritage preservation.

Figures

Figures reproduced from arXiv: 2507.04377 by the authors.

Figure 1
Figure 1. The overall idea of tombstone parsing. The input is an image of a tombstone, and the output is semantically interpreted, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the Tomb2Meaning (T2M) framework. The three retrieval-augmented generation (RAG) strategies are: [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Smatch scores for five models (solid lines) and three [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Examples of tombstones with variability and linguistic complexity (left to right): Rare Languages, Rare Font Styles, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Smatch scores of baselines and T2M on the subsets from five different dimensions. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Image fusion for generating noised tombstones. Person names have been blurred to preserve privacy. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Smatch scores of fine-tuned models on four differ [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

62 extracted references · 52 canonical work pages

  1. [1]

    Mohamed Ashraf Abdelsalam, Zhan Shi, Federico Fancellu, Kalliopi Basioti, Dhaivat Bhatt, Vladimir Pavlovic, and Afsaneh Fazly. 2022. Visual Semantic Parsing: From Images to Abstract Meaning Representation. In Conference on Computational Natural Language Learning . https://api.semanticscholar.org/ CorpusID:253116561

    work page 2022

  2. [2]

    Anthropic. 2024. Claude 3.5 Sonnet. https://www.anthropic.com/news/claude-3- 5-sonnet

    2024

  3. [3]

    Yannis Assael, Thea Sommerschield, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, Jonathan Prag, and Nando De Freitas. 2022. Restoring and attributing ancient texts using deep neural networks. Nature 603, 7900 (2022), 280–283

    work page 2022

  4. [4]

    Laura Banarescu, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider

  5. [5]

    Guntis Barzdins and Didzis Gosko. 2016. RIGA at SemEval-2016 Task 8: Impact of Smatch Extensions and Character-Level Neural Translation on AMR Parsing Accuracy. In Proceedings of the 10th International Workshop on Semantic Evalua- tion (SemEval-2016), Steven Bethard, Marine Carpuat, Daniel Cer, David Jurgens, Preslav Nakov, and Torsten Zesch (Eds.). Assoc...

    work page doi:10.18653/v1/s16-1176 2016

  6. [6]

    John A. Bateman. 1990. Finding translation equivalents: an application of gram- matical metaphor. In Proceedings of the 13th conference on Computational Linguis- tics. 13–18

    work page 1990

  7. [7]

    Bateman, Robert T

    John A. Bateman, Robert T. Kasper, Joerg F. L. Schuetz, and Erich H. Steiner. 1989. A new view on the process of translation. In Proceedings of the fourth conference on European chapter of the Association for Computational Linguistics . 282–290

    work page 1989

  8. [8]

    Bateman and Cecile L

    John A. Bateman and Cecile L. Paris. 1989. Phrasing a text in terms the user understand. In Proceedings of IJCAI. 1511–1517

    work page 1989

Show all 62 references
  1. [9]

    Michele Bevilacqua, Rexhina Blloshmi, and Roberto Navigli. 2021. One SPRING to Rule Them Both: Symmetric AMR Semantic Parsing and Generation without a Complex Pipeline. In AAAI Conference on Artificial Intelligence . https://api. semanticscholar.org/CorpusID:235349016

    2021

  2. [10]

    Fung, Cory J

    Claire Bonial, Julie Foresta, Nicholas C. Fung, Cory J. Hayes, Philip Osteen, Jacob Arkin, Benned Hedegaard, and Thomas Howard. 2023. Abstract Meaning Representation for Grounded Human-Robot Communication. In Proceedings of the Fourth International Workshop on Designing Meanin...

    2023

  3. [11]

    Johan Bos. 2022. A semantically annotated corpus of tombstone inscriptions. International Journal of Digital Humanities 3, 1 (2022), 1–33

    2022

  4. [12]

    Johan Bos. 2023. The Sequence Notation: Catching Complex Meanings in Simple Graphs. In Proceedings of the 15th International Conference on Computational Semantics (IWCS 2023). Nancy, France, 1–14

    2023

  5. [13]

    Johan Bos, Valerio Basile, Kilian Evang, Noortje Venhuizen, and Johannes Bjerva

  6. [14]

    Johan Bos, Cristian Marocico, A Emin Tatar, and Yasmin Mzayek. 2022. Auto- matically Interpreting Dutch Tombstone Inscriptions. Computational Linguistics in the Netherlands Journal 12 (2022), 253–267

    2022

  7. [15]

    Richard Brutti, Lucia Donatelli, Kenneth Lai, and James Pustejovsky. 2022. Ab- stract Meaning Representation for Gesture. In Proceedings of the Thirteenth Lan- guage Resources and Evaluation Conference , Nicoletta Calzolari, Frédéric Béchet, Philippe Blache, Khalid Choukri, Ch...

    2022

  8. [16]

    Shu Cai and Kevin Knight. 2013. Smatch: an Evaluation Metric for Semantic Feature Structures. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , Hinrich Schuetze, Pascale Fung, and Massimo Poesio (Eds.). Associ...

    2013

  9. [17]

    Sharon DeBartolo Carmack. 2002. Your Guide to Cemetery Research . Betterway Books, Cincinatti, Ohio

    2002

  10. [18]

    Zijian Chen, Tingzhu Chen, Wenjun Zhang, and Guangtao Zhai. 2024. OBI- Bench: Can LMMs Aid in Study of Ancient Script on Oracle Bones? arXiv preprint arXiv:2412.01175 (2024)

    2024 arXiv

  11. [19]

    Eva Eckert. 1993. Language change: The testimony of Czech tombstone inscrip- tions in Praha, Texas. Varieties of Czech: studies in Czech sociolinguistics (1993), 189–215

    1993

  12. [20]

    You Are An Expert Linguistic Annotator

    Allyson Ettinger, Jena Hwang, Valentina Pyatkin, Chandra Bhagavatula, and Yejin Choi. 2023. “You Are An Expert Linguistic Annotator”: Limits of LLMs as Analyzers of Abstract Meaning Representation. In Findings of the Association for Computational Linguistics: EMNLP 2023 , Houd...

    2023 doi

  13. [21]

    Moshiur Rahman Farazi, Salman Hameed Khan, and Nick Barnes. 2020. Attention Guided Semantic Relationship Parsing for Visual Question Answering. ArXiv abs/2010.01725 (2020). https://api.semanticscholar.org/CorpusID:222133100

    2020 arXiv

  14. [22]

    Christiane Fellbaum (Ed.). 1998. WordNet: An Electronic Lexical Database. MIT Press, Cambridge, MA

    1998

  15. [23]

    van Gemert

    Morris Franken and Jan C. van Gemert. 2013. Automatic Egyptian Hieroglyph Recognition by Retrieving Images as Texts. In Proceedings of the 21st ACM Inter- national Conference on Multimedia . 765–768

    2013

  16. [24]

    Gemini. 2024. Gemini: A Family of Highly Capable Multimodal Models. arXiv:2312.11805 [cs.CL] https://arxiv.org/abs/2312.11805

    2024 arXiv

  17. [25]

    Michael Wayne Goodman. 2020. Penman: An Open-Source Library and Tool for AMR Graphs. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations , Asli Celikyilmaz and Tsung- Hsien Wen (Eds.). Association for Computational ...

    2020 doi

  18. [26]

    Jan Hajič, Eva Hajičová, Jarmila Panevová, Petr Sgall, Ondřej Bojar, Silvie Cinková, Eva Fučíková, Marie Mikulová, Petr Pajas, Jan Popelka, Jiří Semecký, Jana Šindlerová, Jan Štěpánek, Josef Toman, Zdeňka Urešová, and Zdeněk Žabokrtský

  19. [27]

    Hendrix, Earl D

    Gary G. Hendrix, Earl D. Sacerdoti, Daniel Sagalowicz, and Jonathan Slocum

  20. [28]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-Rank Adaptation of Large Language Models. In International Conference on Learning Representations

    2022

  21. [29]

    Hanqi Jiang, Yi Pan, Junhao Chen, Zhengliang Liu, Yifan Zhou, Peng Shu, Yiwei Li, Huaqin Zhao, Stephen Mihm, Lewis C Howe, et al. 2024. Oraclesage: Towards unified visual-linguistic understanding of oracle bone scripts through cross-modal knowledge fusion. arXiv preprint arXiv...

    2024 arXiv

  22. [30]

    Hans Kamp and Uwe Reyle. 1993. From Discourse to Logic - Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. In Studies in Linguistics and Philosophy . https://api. semanticscholar.org/CorpusID:45807661

    1993

  23. [31]

    Robert T. Kasper. 1989. A Flexible Interface for Linking Applications to Penman’s Sentence Generator. In Proceedings of the DARPA Speech and Natural Language Workshop. Philadelphia, 153–158

    1989

  24. [32]

    Kenneth Lai, Richard Brutti, Lucia Donatelli, and James Pustejovsky. 2024. En- coding Gesture in Multimodal Dialogue: Creating a Corpus of Multimodal AMR. In Proceedings of the 2024 Joint International Conference on Computa- tional Linguistics, Language Resources and Evaluatio...

    2024

  25. [33]

    Jiangming Liu. 2024. Model-Agnostic Cross-Lingual Training for Discourse Rep- resentation Structure Parsing. In Proceedings of the 2024 Joint International Con- ference on Computational Linguistics, Language Resources and Evaluation (LREC- COLING 2024), Nicoletta Calzolari, Mi...

    2024

  26. [34]

    Jiangming Liu. 2024. Soft Well-Formed Semantic Parsing with Score-Based Selection. In Proceedings of the 2024 Joint International Conference on Compu- tational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , Nicoletta Calzolari, Min-Yen Kan, Veronique Hoste...

    2024

  27. [35]

    Chris Evin Long. 2016. Disambiguating the departed: using the genealogist’s tools to uniquely identify the long dead and little known. Library Resources & Technical Services 60, 4 (2016), 236–247

    2016

  28. [36]

    Abelardo Carlos Martínez Lorenzo, Marco Maru, and Roberto Navigli. 2022. Fully-Semantic Parsing and Generation: the BabelNet Meaning Representation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Smaranda Mur...

    2022 doi

  29. [37]

    George F Matthias. 1967. Weathering rates of Portland arkose tombstones.Journal of Geological Education 15, 4 (1967), 140–144

    1967

  30. [38]

    Jonathan May. 2016. SemEval-2016 Task 8: Meaning Representation Parsing. In Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval- 2016), Steven Bethard, Marine Carpuat, Daniel Cer, David Jurgens, Preslav Nakov, and Torsten Zesch (Eds.). Association fo...

    2016 doi

  31. [39]

    Stephan Oepen and Jan Tore Lønning. 2006. Discriminant-Based MRS Bank- ing. In Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC’06), Nicoletta Calzolari, Khalid Choukri, Aldo Gangemi, Bente Maegaard, Joseph Mariani, Jan Odijk, and Da...

    2006

  32. [40]

    OpenAI. 2024. Hello GPT-4. https://openai.com/index/hello-gpt-4o

    2024

  33. [41]

    Qwen. 2025. Qwen2.5-VL Technical Report. arXiv:2502.13923 [cs.CV] https: //arxiv.org/abs/2502.13923

    2025 arXiv

  34. [42]

    Richard P Saller and Brent D Shaw. 1984. Tombstones and Roman family relations in the Principate: civilians, soldiers and slaves. The Journal of Roman Studies 74 (1984), 124–156

    1984

  35. [43]

    Oliver Streiter, Leonhard Voltmer, and Yoann Goudin. 2007. From tombstones to corpora: TSML for research on language, culture, identity and gender differences. In Proceedings of the 21st Pacific Asia Conference on Language, Information and Computation. 450–458

    2007

  36. [44]

    Oliver Streiter, Leonhard Voltmer, and Yoann Goudin. 2007. From Tombstones to Corpora: TSML for Research on Language, Culture, Identity and Gender Differ- ences. In Proceedings of the 21st Pacific Asia Conference on Language, Information and Computation (PACLIC). Seoul Nationa...

    2007

  37. [45]

    Marjorie Templeton and John Burger. 1983. Problems in Natural-Language Interface to DBMS With Examples From EUFID. In First Conference on Applied Natural Language Processing. Association for Computational Linguistics, Santa Monica, California, USA, 3–16. https://doi.org/10.311...

    1983

  38. [46]

    Hart, Peter Stone, and Raymond J

    Jesse Thomason, Aishwarya Padmakumar, Jivko Sinapov, Nick Walker, Yuqian Jiang, Harel Yedidsion, Justin W. Hart, Peter Stone, and Raymond J. Mooney. 2020. Jointly Improving Parsing and Perception for Natural Language Commands through Human-Robot Dialog. J. Artif. Intell. Res. ...

    2020

  39. [47]

    Marco HD Van Leeuwen, Ineke Maas, and Andrew Miles. 2004. Creating a historical international standard classification of occupations an exercise in multinational interdisciplinary cooperation. Historical Methods: A Journal of Quantitative and Interdisciplinary History 37, 4 (2...

    2004

  40. [48]

    Rik van Noord, Lasha Abzianidze, Antonio Toral, and Johan Bos. 2018. Exploring Neural Methods for Parsing Discourse Representation Structures. Transactions of the Association for Computational Linguistics 6 (2018), 619–633. https://doi. org/10.1162/tacl_a_00241

    2018 doi

  41. [49]

    Rik van Noord and Johan Bos. 2017. Neural Semantic Parsing by Character-based Translation: Experiments with Abstract Meaning Representations.Computational Linguistics in the Netherlands Journal 7 (13 Oct. 2017), 93–108

    2017

  42. [50]

    Rik van Noord, Antonio Toral, and Johan Bos. 2020. Character-level Repre- sentations Improve DRS-based Semantic Parsing Even in the Age of BERT. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) , Bonnie Webber, Trevor Cohn, Yula...

    2020 doi

  43. [51]

    Richard F Veit and Mark Nonestied. 2008. New Jersey cemeteries and tombstones: history in the landscape . Rutgers University Press

    2008

  44. [52]

    Chunliu Wang, Xiao Zhang, and Johan Bos. 2023. Discourse Representation Structure Parsing for Chinese. In Proceedings of the 4th Natural Logic Meets Machine Learning Workshop , Stergios Chatzikyriakidis and Valeria de Paiva (Eds.). Association for Computational Linguistics, Na...

    2023

  45. [53]

    William A Woods. 1973. Progress in natural language understanding: an appli- cation to lunar geology. In Proceedings of the June 4-8, 1973, national computer conference and exposition. 441–450

    1973

  46. [54]

    Xiulin Yang, Jonas Groschwitz, Alexander Koller, and Johan Bos. 2024. Scope- enhanced Compositional Semantic Parsing for DRT. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , Yaser Al- Onaizan, Mohit Bansal, and Yun-Nung Chen (Eds.). ...

    2024

  47. [55]

    Na Zhang. 2024. Research on Ancient Text Recognition and Translation Assisted by Artificial Intelligence. Applied Mathematics and Nonlinear Sciences 9, 1 (2024). https://doi.org/10.2478/amns-2024-3681

    2024 doi

  48. [56]

    Xiao Zhang, Gosse Bouma, and Johan Bos. 2025. Neural Semantic Pars- ing with Extremely Rich Symbolic Meaning Representations. Computational Linguistics 51, 1 (03 2025), 235–274. https://doi.org/10.1162/coli_a_00542 arXiv:https://direct.mit.edu/coli/article-pdf/51/1/235/2483888...

    2025 doi

  49. [57]

    Xiao Zhang, Qianru Meng, and Johan Bos. 2024. Retrieval-Augmented Se- mantic Parsing: Using Large Language Models to Improve Generalization. arXiv:2412.10207 [cs.CL]

    2024 arXiv

  50. [58]

    Xiao Zhang, Chunliu Wang, Rik van Noord, and Johan Bos. 2024. Gaining More Insight into Neural Semantic Parsing with Challenging Benchmarks. In Proceed- ings of the Fifth International Workshop on Designing Meaning Representations @ LREC-COLING 2024, Claire Bonial, Julia Bonn,...

    2024

  51. [1977]

    Developing a natural language interface to complex data. In TODS. https: //api.semanticscholar.org/CorpusID:15391397

  52. [2012]

    Announcing Prague Czech-English Dependency Treebank 2.0. In Proceed- ings of the Eighth International Conference on Language Resources and Evaluation (LREC’12), Nicoletta Calzolari, Khalid Choukri, Thierry Declerck, Mehmet Uğur Doğan, Bente Maegaard, Joseph Mariani, Asuncion M...

  53. [2013]

    In Proceedings of the 7th linguistic annotation workshop and interoperability with discourse

    Abstract meaning representation for sembanking. In Proceedings of the 7th linguistic annotation workshop and interoperability with discourse . 178–186

  54. [2017]

    In Handbook of Linguistic Annotation , Nancy Ide and James Pustejovsky (Eds.)

    The Groningen Meaning Bank. In Handbook of Linguistic Annotation , Nancy Ide and James Pustejovsky (Eds.). Vol. 2. Springer, 463–496

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.