nteText

Replication data and tooling for analysis of USTR National Trade Estimate (NTE) reports. Ships the section-level corpus and a fine-tuned DeBERTa-v3 hypothesis-alignment scoring layer for the intellectual property rights (IPR) issue area.

Install

remotes::install_github("jacqpark/nteText")

Quick start

library(nteText)

# Full section-level corpus
nte_load()

# IPR subset (mirrors the NTE_IPR2 working file from the source repo)
nte_subset(issue = "IPR")

# Bundled DeBERTa-scored IPR paragraphs
nte_ipr_scored()

# One-shot regeneration of every figure and table in the paper
nte_replicate_paper(dir = "out")

Scope of the DeBERTa model

The shipped DeBERTa-v3 classifier was fine-tuned on hypothesis labels for the IPR issue area. Use nte_score_ipr() to score new IPR paragraphs. For other issue areas the package provides corpus exploration tools (keyword in context, term trends, comparative wordclouds, country profiles, keyness) but ships no classifier. To classify text from another issue area, fine-tune a separate model using the NTE_DeBERTa_V3_revised_colab.ipynb notebook from the source repository as a template.

Citation for the package

Park, J. (2026). nteText: USTR National Trade Estimate Corpus and IPR Hypothesis Scores. https://doi.org/10.5281/zenodo.20028789

Citation for the paper that introduces the measure

Park J. (2026). Aid, Lending, and TRIPS. Working paper.