Pith. sign in

REVIEW 3 major objections 3 minor 23 references

Separating Argument Structure from Logical Structure in AMR

T0 review · 3 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read AMR can be extended with a logical layer of indexed contexts that handles negation, quantifier scope, and bound variables.

desk verdict The AMR+ proposal is a clean idea, but the paper's central equivalence claim rests on a translation that is both underspecified and internally inconsistent in the headline example. read the letter →

arxiv 1908.01355 v2 pith:ARDKID4U submitted 2019-08-04 cs.CL

Johan Bos This is my paper · ORCID
classification cs.CL
keywords AbstractMeaningRepresentationAMR+scopequantificationnegationpresuppositionDiscourseTheorycontextindexing
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

The paper argues that Abstract Meaning Representation, which excels at predicate-argument structure, can be given a logical layer without losing that structure. By labeling each sub-AMR with an index and adding constraints on how those contexts relate, the paper derives a scoped formalism it calls AMR+. This is claimed to solve two known problems: the bound variable problem, where AMR graphs misread "every snake bit itself," and the scope representation problem, where AMR's predicate-style negation permits invalid inferences such as concluding "Mary left" from "Mary did not leave." The paper further claims AMR+ is equivalent in expressive power to the DRS dialect used in the Parallel Meaning Bank, which would transfer DRT's inferential properties to AMR.

What carries the argument

The central object is the indexed context layer: every slash in a PENMAN AMR is decorated with a numeric index, and a set of constraints over those indices describes the logical relationships between the corresponding sub-AMRs. The constraints use identity, negation, conditional implication, and presupposition (‘=’, ‘¬’, ‘⇒’, ‘<’), with colons expressing that a context contains a condition. This machinery recursively decomposes an AMR into sub-AMRs and then reassembles them into a scoped structure, separating logical structure from the surface-friendly argument structure that AMR is known for.

What would settle it

Apply the paper's three-step translation to "A dog scared every cat" exactly as in Section 3.5: the resulting DRS is ∀y(cat(y)→∃x(dog(x)∧scare(x,y))), which is not the surface-scope reading ∃x(dog(x)∧∀y(cat(y)→scare(x,y))). If the intended AMR+ for that sentence admits only the weaker reading, then the formalism under-generates; if both readings are intended, the constraints lack a mechanism for deriving the missing one. Either way, checking which reading the notation actually delivers would settle the claim that AMR+ correctly handles quantifier scope.

Watch

Extended reading notes

Core claim

The central claim is that adding a second, logical layer to AMRs—indices on every sub-AMR plus constraints stating how contexts relate as identical, negated, conditional, or presuppositional—makes AMR+ correct for inferences involving negation and bound variables while preserving the original predicate-argument structure. Negation becomes a scoping operator rather than a predicate attribute, universal quantification becomes a conditional context, and definite descriptions and names trigger presuppositional contexts. A three-step translation converts AMR+ into DRS: turn each sub-AMR into a DRS, merge DRSs with the same index, and combine them by following the constraints. On the paper's account, AMR+ is equivalent in expressive power to the Parallel Meaning Bank's DRS dialect, and this is what would give AMR the inferential behavior of Discourse Representation Theory.

Load-bearing premise

The claim stands on the assumption that the informally specified translation from AMR+ context constraints to DRS faithfully captures the meaning of those constraints; no model-theoretic semantics is given, and the paper's own example "A dog scared every cat" is indexed differently in Section 3.4 than in Section 3.5.

Editorial extensions

If this is right

  • Existing AMR-annotated corpora can be extended monotonically: add indices automatically by replacing every "/" with "/1/", then manually fix cases of negation, universal quantification, and definite descriptions.
  • The ":polarity" and ":mod" relations in AMR become optional or removable, because the new context constraints take over their logical work.
  • Because AMR+ can be translated into the Parallel Meaning Bank's DRS dialect through the three-step recipe, DRT's model-theoretic components and inference procedures could be applied to AMR.
  • The paper's formalism gives AMR+ the same expressive power as the Parallel Meaning Bank's DRS dialect, though not the full expressive power of general DRS.
  • A modular parsing strategy becomes possible: existing AMR parsers can remain unchanged, with a separate layer adding the logical constraints.

Reading between the lines

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

  • Taken literally, the paper's own translation of "A dog scared every cat" in Section 3.5 yields the reading ∀y(cat(y)→∃x(dog(x)∧scare(x,y))), with "every cat" wide over "a dog," rather than the surface reading ∃x(dog(x)∧∀y(cat(y)→scare(x,y))); the paper does not flag this discrepancy, so the intended scope readings are not yet fully pinned down.
  • A model-theoretic semantics for the context constraints themselves—rather than only for their DRS translations—would make the claimed inference predictions checkable and would settle whether the constraints are compositional.
  • The constraint vocabulary looks portable: other graph-based meaning representations that suffer from the same bound-variable problem could adopt indexed-context constraints, though the paper does not claim this.
  • The proposal implicitly treats quantifier scope as a property of the representation rather than of the parsing process; testing it on genuinely ambiguous sentences would clarify whether AMR+ stores one reading per structure or can encode multiple readings.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The paper proposes AMR+, an extension of Abstract Meaning Representation that adds a logical layer on top of the usual predicate-argument structure. Each sub-AMR receives an index, and additional scoping constraints over these indices (=, ¬, ⇒, <) are used to represent negation, quantification, and presuppositions. The paper gives examples for existential and universal quantification, negation, proper names, and bound variables, and sketches a translation τ from AMR+ to the DRS dialect used in the Parallel Meaning Bank. The central claim is that AMR+ 'makes correct predictions' for inferences involving negation and bound variables and is equivalent in expressive power to that DRS dialect, while preserving AMR's attractive core structure. The paper is a short proposal: it does not give a model-theoretic semantics for the labelled contexts, and the translation τ is presented only by examples.

Significance. If made rigorous, this proposal would address two well-known limitations of AMR: the bound variable problem and the scope representation problem. The idea of separating predicate-argument structure from logical structure is simple and potentially useful, and the link to DRT could enable cross-formalism comparison and inference. The paper is honest about its open ends and does not overstate its results into a full formal theory. However, the paper's headline claims of correctness and DRS equivalence are not yet supported: the semantics of the constraint language is left informal, and the translation τ is under-specified and, as the examples show, internally inconsistent in a load-bearing case. The contribution is therefore best seen as a promising research program rather than a verified formalism.

major comments (3)
  1. [§3.4 and §3.5] The same sentence "A dog scared every cat" is represented with different variable indices in the two sections: in §3.4 the dog is x/3 and the constraint is {3:2=>1}, apparently giving the dog wide scope, whereas in §3.5 the dog is x/1, already merged into τ(1), and the translation yields τ(3)=τ(2)⇒τ(1), which places the existential dog inside the consequent of the universal. The resulting DRS reads as "every cat was scared by some dog," not as "a dog scared every cat." This inconsistency shows that the mapping from AMR+ to DRS is not deterministic and undermines the claimed correctness of the formalism for scope inferences.
  2. [§3.5, translation τ] The translation τ is not a well-defined function. No rule is given for interpreting the leading label in an expression such as {3:2=>1}: the leading 3 is sometimes an inferred context (as in §3.3) and sometimes the label of a sub-AMR (the dog in §3.4). Similarly, no semantics is provided for the presupposition constraint <; in the "Mary didn't smile" example, τ(3)=¬τ(1) leaves the variable x in Agent(e,x) unbound, and the example does not show how τ(2) combines with τ(3). Without a recursive definition of l:C and a model-theoretic account of the constraints, the equivalence claim with PMB-style DRS is not testable.
  3. [§3.4, "Every student revised their paper"] The constraint set {2=3,3<1,4:2=>1} is unclear. In the AMR, x/2 is the student and y/3 is the paper, so 2=3 appears to identify the student's context with the paper's context, which is not the intended DRS reading. No explanation is provided of how these constraints encode the universal over students and the coindexation of the possessive pronoun with the student. This example suggests that the constraint language needs a precise semantics before the informal readings of the examples can be checked.
minor comments (3)
  1. [Abstract] The phrase "makes correct prediction for inferences" is grammatically incorrect; it should be "makes correct predictions for inferences."
  2. [§3.5, Step 1] The description of the two-place relation says “the second argument of the two-place relation is the main variable of the sub-AMR” twice; the second occurrence should refer to the main variable of the argument sub-AMR. The displayed formula also uses 'x' for the main variable while the surrounding text uses 'i'.
  3. [§4.1] The proposed shorthand {¬j} for {i:¬j} and {j⇒k} for {i:j⇒k} is introduced without a rule for choosing the containing context i; this is especially problematic when multiple inferred contexts are present, as in §3.4.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a formalism proposal whose DRS translation is a definitional exercise, not a fitted prediction.

full rationale

The derivation chain in this paper is not circular in the sense of the review criteria. AMR+ is introduced by assigning indices to sub-AMRs and adding constraints, and the constraints are then interpreted via a translation tau into PMB-style DRSs in Section 3.5. There is no fitted parameter, no empirical quantity predicted from a subset of its own data, and no uniqueness theorem imported from the authors to rule out alternatives. The claimed equivalence with the PMB dialect of DRS is a consequence of the way tau is defined, but that makes it a definitional or formal claim rather than a case of the formalism's inputs forcing a disguised empirical prediction; the paper openly presents the conversion as 'straightforward' and does not present the equivalence as an empirical discovery. The same-author citations, such as Bos (2016) for the recursive view of AMRs and Abzianidze et al. (2017) for PMB DRSs, support notation and background and are not load-bearing in a way that reduces the central proposal to an unverified self-citation. Concerns that tau is under-specified or that the Section 3.5 example assigns different indices than Section 3.4 are real correctness and completeness issues, not circularity: an incomplete proof of adequacy is not an input disguised as an output.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

No numeric free parameters are fit to data; the paper is a conceptual proposal. The central claim rests on four substantive assumptions about how AMRs decompose, how four context relations suffice, how the DRS translation preserves meaning, and how presuppositions are triggered. These are not derived or validated in the paper.

assumptions (4)
  • domain assumption Every slash in an AMR denotes a sub-AMR, and each sub-AMR can be assigned an index.
    Section 2 defines the recursive view of AMR that makes the indexing possible; this is a modeling choice about what counts as a sub-AMR, and it is not grounded in the AMR specification itself.
  • ad hoc to paper The scoping constraints =, ¬, ⇒, and < are sufficient to represent the semantics of negation, universal quantification, and presupposition.
    Section 2 introduces these four context relations without arguing that they cover the phenomena the paper claims to handle.
  • ad hoc to paper The translation τ from AMR+ to PMB-style DRS is meaning-preserving.
    Section 3.5 defines three translation steps but leaves out details and gives no soundness or completeness argument; the correctness of the whole proposal depends on this.
  • domain assumption Definite descriptions and proper names uniformly trigger existential presuppositions represented by <.
    Section 3.2 treats "the bear" and "Fido" with the same presuppositional context relation, without discussing cases where presuppositions project differently or are bound.
invented entities (1)
  • Inferred context labels (extra indices not present in predicate-argument structure, e.g., context 2 in {2:¬1} and context 3 in {3:2=>1})
    purpose: Host contexts for negation and conditional operators when the AMR does not provide an explicit context.
    These labels are introduced to make the logical layer well-formed. The paper gives no independent semantics for them and no empirical prediction; they are formal scaffolding.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Separating Argument Structure from Logical Structure in AMR." pith.science (2026). https://pith.science/paper/ARDKID4U

@misc{pith2026190801355,
  author       = {Pith},
  title        = {Pith review of: Separating Argument Structure from Logical Structure in AMR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ARDKID4U}},
  note         = {Machine review of arXiv:1908.01355}
}
read the original abstract

The AMR (Abstract Meaning Representation) formalism for representing meaning of natural language sentences was not designed to deal with scope and quantifiers. By extending AMR with indices for contexts and formulating constraints on these contexts, a formalism is derived that makes correct prediction for inferences involving negation and bound variables. The attractive core predicate-argument structure of AMR is preserved. The resulting framework is similar to that of Discourse Representation Theory.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 22 canonical work pages

  1. [1]

    Lasha Abzianidze, Johannes Bjerva, Kilian Evang, Hessel Haagsma, Rik van Noord, Pierre Ludmann, Duc-Duy Nguyen, and Johan Bos. 2017. The P arallel M eaning B ank: Towards a multilingual corpus of translations annotated with compositional meaning representations. In Proceedings of the 15th Conference of the European Chapter of the Association for Computati...

    work page 2017

  2. [2]

    Yoav Artzi, Kenton Lee, and Luke Zettlemoyer. 2015. Broad-coverage CCG semantic parsing with AMR . In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing , pages 1699--1710, Lisbon, Portugal. Association for Computational Linguistics

    work page 2015

  3. [3]

    Laura Banarescu, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. 2013. Abstract Meaning Representation for Sembanking . In Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse , pages 178--186, Sofia, Bulgaria, August

    work page 2013

  4. [4]

    Johan Bos and Lasha Abzianidze. 2019. Thirty musts for meaning banking. In Proceedings of the First International Workshop on Designing Meaning Representations , pages 15--27, Florence, Italy. Association for Computational Linguistics

    work page 2019

  5. [5]

    Johan Bos. 1996. Predicate Logic Unplugged . In P. Dekker and M. Stokhof, editors, Proceedings of the Tenth Amsterdam Colloquium , pages 133--143, ILLC/Dept. of Philosophy, University of Amsterdam

    work page 1996

  6. [6]

    Johan Bos. 2016. Expressive power of abstract meaning representations. Computational Linguistics , 42(3):527--535

    work page 2016

  7. [7]

    Shu Cai and Kevin Knight. 2013. S match: an evaluation metric for semantic feature structures. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages 748--752, Sofia, Bulgaria, August. Association for Computational Linguistics

    work page 2013

  8. [8]

    Ann Copestake, Dan Flickinger, Rob Malouf, Susanne Riehemann, and Ivan Sag. 1995. Translation using Minimal Recursion Semantics . In Proceedings of the Sixth International Conference on Theoretical and Methodological Issues in Machine Translation , pages 15--32, University of Leuven, Belgium

    work page 1995

Show all 23 references
  1. [9]

    Dick Crouch and Aikaterini-Lida Kalouli. 2018. Named graphs for semantic representations. In The Seventh Joint Conference on Lexical and Computational Semantics (*SEM 2018) , pages 113--118, New Orleans

    2018

  2. [10]

    Cohen, and Giorgio Satta

    Marco Damonte, Shay B. Cohen, and Giorgio Satta. 2017. An incremental parser for A bstract M eaning R epresentation. In Proceedings of the 15th Conference of the E uropean Chapter of the Association for Computational Linguistics: Volume 1, Long Papers , pages 536--546, Valencia, Spain

    2017

  3. [11]

    Bart Geurts. 1999. Presuppositions and Pronouns . Elsevier, London

    1999

  4. [12]

    Hans Kamp and Uwe Reyle. 1993. From Discourse to Logic; An Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and DRT . Kluwer, Dordrecht

    1993

  5. [13]

    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 , pages 153--158, Philadelphia

    1989

  6. [14]

    Kenneth Lai, Lucia Donatelli, and James Pustejovsky. 2020. A continuation semantics for abstract meaning representation. In The Second International Workshop on Designing Meaning Representations (DMR 2020) , Barcelona, Spain

    2020

  7. [15]

    Irene Langkilde and Kevin Knight. 1998. Generation that exploits corpus-based statistical knowledge. In COLING 1998 Volume 1: The 17th International Conference on Computational Linguistics , pages 704--710

    1998

  8. [16]

    James Pustejovsky, Nianwen Xue, and Kenneth Lai. 2019. Modeling quantification and scope in abstract meaning representations. In Proceedings of the First International Workshop on Designing Meaning Representations , pages 28--33, Florence, Italy. Association for Computational ...

    2019

  9. [17]

    Uwe Reyle. 1993. Dealing with Ambiguities by Underspecification: Construction, Representation and Deduction . Journal of Semantics , 10:123--179

    1993

  10. [18]

    Ed Stabler. 2017. Reforming AMR . In Formal Grammar 2017. Lecture Notes in Computer Science , volume 10686, pages 72--87. Springer

    2017

  11. [19]

    Van der Sandt

    Rob A. Van der Sandt. 1992. Presupposition Projection as Anaphora Resolution . Journal of Semantics , 9:333--377

    1992

  12. [20]

    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 , 7:93--108

    2017

  13. [21]

    Venhuizen, Johan Bos, and Harm Brouwer

    Noortje J. Venhuizen, Johan Bos, and Harm Brouwer. 2013. Parsimonious semantic representations with projection pointers. In Proceedings of the 10th International Conference on Computational Semantics (IWCS 2013) -- Long Papers , pages 252--263, Potsdam, Germany, March. Associa...

    2013

  14. [22]

    Noortje Venhuizen, Johan Bos, Petra Hendriks, and Harm Brouwer. 2018. Discourse semantics with information structure. Journal of Semantics , 35(1):127--169

    2018

  15. [23]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence '...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.