stanfordnlp/CoreNLP · error · RuntimeException

Warning: bad pattern "%s" + pattern

Error message

Warning: bad pattern "%s"
 + pattern

What it means

SemanticGraph.matchPatternToVertex walks a pattern string and matches it against a vertex's relations/children. When a pattern element's type is not one of the recognized kinds (e.g. a relation string that is neither a wildcard nor a valid pattern token), it throws a RuntimeException declaring the pattern bad. The message itself contains a bug: "Warning: bad pattern \"%s\"\n" + pattern is concatenated, not formatted, so %s is printed literally.

Solutions

1. Print/correct the pattern string and fix the offending token (the raw pattern is appended after the literal %s)
2. Verify the pattern syntax against the version of the library in use; patterns for other matchers (tregex) may not be valid here
3. If a custom pattern element type is passed, convert it to a supported representation before calling
4. Upgrade/patch the message to use String.format if the literal %s makes debugging hard

Example fix

// before
throw new RuntimeException("Warning: bad pattern \"%s\"\n" + pattern);
// after
throw new RuntimeException(String.format("Warning: bad pattern \"%s\"%n", pattern));

// validate pattern tokens against supported kinds before matching
boolean knownPatternElement(Object elem) {
  return elem instanceof String || elem instanceof GrammaticalRelation
      || elem instanceof java.util.regex.Pattern;
}

try {
  boolean ok = graph.matchPatternToVertex(pattern, vertex);
} catch (RuntimeException e) {
  if (e.getMessage().startsWith("Warning: bad pattern")) {
    // log full pattern, skip or re-parse with a corrected grammar
  } else throw e;
}

When it happens

Trigger: Calling matchPatternToVertex (or public wrappers that evaluate a pattern against a SemanticGraph) with a pattern string containing an unrecognized element type in the branch reached at line 414 — e.g. a pattern token that is not a GrammaticalRelation, regex, or one of the supported node/relation specifiers.

Common situations: Typos in hand-written tregex/semgrex-style patterns; copying patterns from docs of a different library version where the token grammar changed; RTE-era helper code encountering a pattern element class it does not handle.

Related errors

• Error on line
• Invalid txtGraph
• Invalid hypGraph
• Null index array
• Index array length

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/3ebe0512a81cf2d7. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/semgraph/SemanticGraph.java:414

          for (Pair<GrammaticalRelation, IndexedWord> pair : children) {
            if (pair.first().toString().equals("det"))
              continue;
            IndexedWord child = pair.second();
            String lemma = child.get(CoreAnnotations.LemmaAnnotation.class);
            if (lemma.isEmpty()) {
              lemma = child.word().toLowerCase();
            }
            if (lemma.equals(word)) {
              match = true;
              break;
            }
          }
          if (!match) {
            return false;
          }
        }
      } else {
        throw new RuntimeException("Warning: bad pattern \"%s\"\n" + pattern);
      }
    }
    return true;
  }

  // todo [cdm 2013]: Completely RTE-specific methods like this one should be used to a static class of helper methods under RTE
  public boolean matchPatternToVertex(String pattern, IndexedWord vertex) {
    if (!containsVertex(vertex)) {
      throw new IllegalArgumentException();
    }
    String pat = pattern.replaceAll("<", ",<");
    pat = pat.replaceAll(">", ",>");
    String[] nodePath = pat.split(",");
    for (String s : nodePath) {
      if (s.isEmpty()) {
        continue;
      }
      String word = s.substring(1);

View on GitHub (pinned to 1b7edd19c4)