What causes "Bad pattern: " in stanfordnlp/CoreNLP?

Fix the regex syntax in the sourcePattern string (validate it with Pattern.compile in isolation or a regex tester)

stanfordnlp/CoreNLP · error · RuntimeException

Bad pattern

Error message

Bad pattern: <sourcePattern>

What it means

The GrammaticalRelation constructor compiles the optional sourcePattern string into a java.util.regex.Pattern. If the string is not a syntactically valid regular expression, Pattern.compile throws PatternSyntaxException, which is wrapped and rethrown as this RuntimeException. It indicates a malformed regex in a relation definition.

Solutions

Fix the regex syntax in the sourcePattern string (validate it with Pattern.compile in isolation or a regex tester)
Escape metacharacters that were meant literally (e.g. use \. for a dot)
If no source pattern is needed, pass null so the field is left null instead of compiling a bad string
Test the pattern string against the exact Java regex dialect before adding the relation

Example fix

// before
new GrammaticalRelation(lang, "nsubj", "nominal subject", "**bad[", null, ...);
// after
new GrammaticalRelation(lang, "nsubj", "nominal subject", "^(?!.*\\b(expletive)\\b).*$", null, ...);

Defensive patterns

Strategy: validation

Validate before calling

try {
    java.util.regex.Pattern.compile(sourcePattern);
} catch (java.util.regex.PatternSyntaxException e) {
    throw new IllegalArgumentException("Invalid sourcePattern: " + sourcePattern, e);
}

Try / catch

try {
    GrammaticalRelation r = new GrammaticalRelation(lang, name, longName, sourcePattern, ...);
} catch (RuntimeException e) {
    if (e.getMessage().startsWith("Bad pattern:")) {
        // fix regex or fall back to null sourcePattern
    } else throw e;
}

Prevention

Test regex strings with Pattern.compile before wiring them into relations
Escape regex metacharacters intended literally
Prefer null sourcePattern when no source constraint is needed
Use a regex linter/tester tuned to Java syntax

When it happens

Trigger: Constructing a new GrammaticalRelation(...) passing a sourcePattern argument containing invalid regex syntax (e.g. unbalanced parentheses, dangling quantifier like 'foo**' or stray '[class').

Common situations: Defining custom grammatical relations in code or config with a mistyped regex; copying patterns between regex dialects where escaping differs; recent edits introducing a typo in the pattern literal.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/c1d1d73d2a44b309. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/trees/GrammaticalRelation.java:296

                             String sourcePattern,
                             TregexPatternCompiler tregexCompiler,
                             String[] targetPatterns,
                             String specificString) {
    this.language = language;
    this.shortName = shortName;
    this.longName = longName;
    this.parent = parent;
    this.specific = specificString; // this can be null!

    if (parent != null) {
      parent.addChild(this);
    }

    if (sourcePattern != null) {
      try {
        this.sourcePattern = Pattern.compile(sourcePattern);
      } catch (java.util.regex.PatternSyntaxException e) {
        throw new RuntimeException("Bad pattern: " + sourcePattern);
      }
    } else {
      this.sourcePattern = null;
    }

    for (String pattern : targetPatterns) {
      try {
        TregexPattern p = tregexCompiler.compile(pattern);
        this.targetPatterns.add(p);
      } catch (edu.stanford.nlp.trees.tregex.TregexParseException pe) {
        throw new RuntimeException("Bad pattern: " + pattern, pe);
      }
    }

    GrammaticalRelation previous;
    synchronized (stringsToRelations) {
      Map<String, GrammaticalRelation> sToR = stringsToRelations.get(language);
      if (sToR == null) {

View on GitHub (pinned to 1b7edd19c4)