CAMRA: Copilot for AMR Annotation

TL;DR

CAMRA introduces a web-based AMR annotation tool that treats annotation as coding, integrating local Propbank autocomplete and model-in-the-loop parser copilots to improve efficiency and accuracy. The system features a three-panel, code-like editor with syntax-aware editing, two layers of auto-completion (local and parser-driven global), and persistent search windows, all supported by modular language-server backends including SPRING (BART-large) for parsing. Key contributions include the design of an annotator-centric, modular platform that enables seamless integration of different assistant models and formalisms, plus an analysis of parser response times to inform UI decisions. The work demonstrates a practical path toward safer, more interpretable AI-assisted semantic annotation and outlines future directions for expanding LLM integration, cross-domain generalization, and multilingual PENMAN support, with potential broad impact on semantic annotation workflows.

Abstract

In this paper, we introduce CAMRA (Copilot for AMR Annotatations), a cutting-edge web-based tool designed for constructing Abstract Meaning Representation (AMR) from natural language text. CAMRA offers a novel approach to deep lexical semantics annotation such as AMR, treating AMR annotation akin to coding in programming languages. Leveraging the familiarity of programming paradigms, CAMRA encompasses all essential features of existing AMR editors, including example lookup, while going a step further by integrating Propbank roleset lookup as an autocomplete feature within the tool. Notably, CAMRA incorporates AMR parser models as coding co-pilots, greatly enhancing the efficiency and accuracy of AMR annotators. To demonstrate the tool's capabilities, we provide a live demo accessible at: https://camra.colorado.edu

Figures (6)

• Figure 1: AMR for sentence "The boy must not go." in conventional graph representation format (left) and PENMAN encoding language format (right)
• Figure 2: Adding a new argument to predicate node (w / want-01) with ISI editor's interface. Each dashed line box represents an updated view after submitting the mini form. Blue colored fields of each form represent fields that are required to be filled before submitting
• Figure 3: an overview of the CAMRA editor with an annotated example sentence. Left panel is the surface text area with dynamic variable carryover from the constructed AMR code. The middle panel is the main AMR editing area where the string complies with the PENMAN encoding syntax for AMR. The right text panel renders the parser suggestions. Note: this screenshot contains only nonempty part of the UI, the UI is window size responsible.
• Figure 4: When looking up in the Propbank rolesets for the keyword "make," a persistent new window will appear at the annotator's disposal on the side of the CAMRA's main interface.
• Figure 5: When looking up in the existing AMRs corpora for the keyword "must", another persistent new window will appear at the annotator's disposal on the side of the CAMRA's main interface.