Pith. Sign in

RECORD

Language Models at the Syntax-Semantics Interface: A Case Study of the Long-Distance Binding of Chinese Reflexive ziji

PaperPDF from arXiv

data files

Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Xiulin Yang

Computation and Language · Submitted 2 Apr 2025

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim Evaluating 21 language models against this dataset and comparing their performance to judgments from native Mandarin speakers, we find that none of the models consistently replicates human-like judgments.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID2504.02116 (version 1)
Pith IDPLBWE7GI, Pith's permanent record of this paper and its review
CitationXiulin Yang. "Language Models at the Syntax-Semantics Interface: A Case Study of the Long-Distance Binding of Chinese Reflexive ziji." arXiv:2504.02116 (2025). https://pith.science/paper/PLBWE7GI
Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 4,288 input tokens.

What the abstract itself states, as probabilities: claims first 0.03 · states limitation 0.32 · human participants 0.47 · controlled comparison 0.69 · randomized assignment 0.02 · states quantitative result 0.87 · compares with prior results 0.72 · mentions code or data availability 0.03.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; .

This paper explores whether language models can effectively resolve the complex binding patterns of the Mandarin Chinese reflexive ziji, which are constrained by both syntactic and semantic factors. We construct a dataset of 240 synthetic sentences using templates and examples from syntactic literature, along with 320 natural sentences from the BCC corpus. Evaluating 21 language models against this dataset and comparing their performance to judgments from native Mandarin speakers, we find that none of the models consistently replicates human-like judgments. The results indicate that existing language models tend to rely heavily on sequential cues, though not always favoring the closest strings, and often overlooking subtle semantic and syntactic constraints. They tend to be more sensitive to noun-related than verb-related semantics.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 This paper explores whether language models can effectively resolve the complex binding patterns of the Mandarin Chinese reflexive ziji, which are constrained by both syntactic and semantic factors. contribution 0.88 0.55 0.04 unclear 0.42
s1 We construct a dataset of 240 synthetic sentences using templates and examples from syntactic literature, along with 320 natural sentences from the BCC corpus. method 1.00 0.96 0.02 asserted 0.99
s2 Evaluating 21 language models against this dataset and comparing their performance to judgments from native Mandarin speakers, we find that none of the models consistently replicates human-like judgments. result 0.99 0.97 0.74 asserted 1.00
s3 The results indicate that existing language models tend to rely heavily on sequential cues, though not always favoring the closest strings, and often overlooking subtle semantic and syntactic constraints. interpretation 0.64 0.92 0.54 hedged possibility 0.79
s4 They tend to be more sensitive to noun-related than verb-related semantics. result 0.92 0.85 0.09 asserted 0.94

What kind of paper the classifier reports this is: empirical study 0.99 · evidence from observation or experiment 0.93.

What was examined, and what was not

sources read
  • arXiv listing abstract v1
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper
Original abstract

This paper explores whether language models can effectively resolve the complex binding patterns of the Mandarin Chinese reflexive ziji, which are constrained by both syntactic and semantic factors. We construct a dataset of 240 synthetic sentences using templates and examples from syntactic literature, along with 320 natural sentences from the BCC corpus. Evaluating 21 language models against this dataset and comparing their performance to judgments from native Mandarin speakers, we find that none of the models consistently replicates human-like judgments. The results indicate that existing language models tend to rely heavily on sequential cues, though not always favoring the closest strings, and often overlooking subtle semantic and syntactic constraints. They tend to be more sensitive to noun-related than verb-related semantics.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

What this paper does with the arXiv papers it cites

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.