stanfordnlp/CoreNLP · error · RuntimeException

Between when the internal phrase edge was found and now…

Error message

Between when the internal phrase edge was found and now, the edge was somehow deleted

What it means

SetPhraseHead.evaluate() deletes the internal edges of the extracted phrase (edges among matched nodes) via sg.removeEdge(edge). If removal fails because the edge is already gone, this RuntimeException is thrown — a defensive check that the graph did not change between edge collection and deletion.

Solutions

Guard against double application: track which rules already ran or re-run semgrex matching before each evaluate
Ensure each matched phrase is edited exactly once per pipeline run
Restore the original graph between repeated evaluations (deep copy before edit)
Check prior ssurgeon operations in the pipeline for overlapping delete behavior

Example fix

// before
ssurgeon.apply(sg, rule); ssurgeon.apply(sg, rule);
// after
SemanticGraph copy = new SemanticGraph(sg);
ssurgeon.apply(copy, rule); // apply once per graph instance

Defensive patterns

Strategy: try-catch

Validate before calling

// check internal phrase edges exist before delete loop
for (SemanticGraphEdge e : deleteEdges) {
  if (sg.getEdge(e.getSource(), e.getTarget(), e.getRelation()) == null) { /* refresh matches */ }
}

Try / catch

try { op.evaluate(sg, sm); } catch (RuntimeException e) { if (e.getMessage().contains("internal phrase edge")) { /* re-match on fresh copy and retry */ } else { throw e; } }

Prevention

Guard against applying the same rule twice to one graph
Use non-overlapping ssurgeon rules or re-match between rules
Deep-copy the graph before evaluation in tests

When it happens

Trigger: evaluate() invoked when internal phrase edges were already removed (e.g. the rule ran once before on the same graph, or an overlapping rule deleted them), making removeEdge return false.

Common situations: Running a ssurgeon pipeline where two rules match overlapping node sets; junit/test fixtures reusing one SemanticGraph across assertions; manually pre-pruning edges then invoking SetPhraseHead.

Understand the failure class

Background: "This is a bug, please report it": internal invariant violations, unreachable panics, and SNH errors explained — this error's family across 47 libraries.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/f7836689f30c7000. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/semgraph/semgrex/ssurgeon/SetPhraseHead.java:151

    for (SemanticGraphEdge edge : relocateEdges) {
      SemanticGraphEdge newEdge = new SemanticGraphEdge(newHead,
                                                        edge.getTarget(),
                                                        edge.getRelation(),
                                                        edge.getWeight(),
                                                        edge.isExtra());
      boolean success = sg.removeEdge(edge);
      if (!success) {
        throw new RuntimeException("Between when the incoming edge was found and now, the edge was somehow deleted");
      }
      sg.addEdge(newEdge);
      modified = true;
    }

    for (SemanticGraphEdge edge : deleteEdges) {
      boolean success = sg.removeEdge(edge);
      if (!success) {
        throw new RuntimeException("Between when the internal phrase edge was found and now, the edge was somehow deleted");
      }
      modified = true;
    }
    for (IndexedWord other : matchedNodes) {
      if (other == newHead)
        continue;

      found: {
        for (SemanticGraphEdge existingEdge : sg.getAllEdges(newHead, other)) {
          if (existingEdge.getRelation().equals(relation)) {
            break found;
          }
        }
        SemanticGraphEdge newEdge = new SemanticGraphEdge(newHead,
                                                          other,
                                                          relation,
                                                          weight,
                                                          false);

View on GitHub (pinned to 1b7edd19c4)