What causes "only support settings for CTB and PKU now." in stanfordnlp/CoreNLP?

Set dict2name to a custom non-dictionary file path so the constructor builds NonDict2 from your resource

stanfordnlp/CoreNLP · error · RuntimeException

only support settings for CTB and PKU now.

Error message

only support settings for CTB and PKU now.

What it means

NonDict2's constructor builds the non-dictionary (unseen word) feature dictionary based on corpus flags. It only supports CTB and PKU: if flags.useAs, useHk, or useMsr is set, it refuses and throws RuntimeException('only support settings for CTB and PKU now.') because no .non dictionary resource exists for those corpora.

Solutions

Set dict2name to a custom non-dictionary file path so the constructor builds NonDict2 from your resource
Disable useAs/useHk/useMsr and use CTB or PK settings instead
Supply your own corpus-specific .non dictionary file and adjust the constructor path

Example fix

// before (Properties)
useAs=true
// after (Properties)
dict2name=asbc.non
useAs=false

Defensive patterns

Strategy: validation

Validate before calling

Properties p = /* segmenter props */;
boolean unsupported = Boolean.parseBoolean(p.getProperty("useAs","false"))
  || Boolean.parseBoolean(p.getProperty("useHk","false"))
  || Boolean.parseBoolean(p.getProperty("useMsr","false"));
if (unsupported && (p.getProperty("dict2name") == null || p.getProperty("dict2name").isEmpty())) {
  throw new IllegalArgumentException("Set dict2name or use CTB/PK settings (NonDict2 limitation)");
}

Try / catch

try {
  segmenter.initialize();
} catch (RuntimeException e) {
  if (e.getMessage().contains("CTB and PKU")) {
    props.setProperty("dict2name", "custom.non");
    segmenter.initialize();
  } else throw e;
}

Prevention

For ASBC/HK/MSR data, always provide dict2name pointing to a valid .non dictionary
Don't combine corpus flags; choose CTB or PK for NonDict2 features
Verify dictionary files exist under corporaDict/dict/ at startup

When it happens

Trigger: Instantiating NonDict2 (during segmenter setup) with useAs=true, useHk=true, or useMsr=true and no custom dict2name specified.

Common situations: Segmenting ASBC (Taiwan), HK, or MSR Sighan data with the NonDict2 feature enabled; mixing corpus flags; users assuming all Sighan corpora are supported like CTB/PK.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/80308ee6d3c21cb1. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/wordseg/NonDict2.java:42

  public final String corporaDict;
  private final CorpusDictionary cd;

  private static Redwood.RedwoodChannels logger = Redwood.channels(NonDict2.class);

  public NonDict2(SeqClassifierFlags flags) {
    if (flags.sighanCorporaDict != null) {
      corporaDict = flags.sighanCorporaDict; // use the same flag for Sighan 2005,
      // but our list is extracted from ctb
    } else {
      corporaDict = DEFAULT_HOME;
    }

    String path;
    if (flags.dict2name != null && !flags.dict2name.equals("")) {
      path = corporaDict + "/dict/" + flags.dict2name;
      logger.info("INFO: dict2name specified | building NonDict2 from "+path);
    } else if (flags.useAs || flags.useHk || flags.useMsr) {
      throw new RuntimeException("only support settings for CTB and PKU now.");
    } else if ( flags.usePk ) {
      path = corporaDict+"/dict/pku.non";
      logger.info("INFO: flags.usePk=true | building NonDict2 from "+path);
    } else { // CTB
      path = corporaDict+"/dict/ctb.non";
      logger.info("INFO: flags.usePk=false | building NonDict2 from "+path);
    }

    cd = new CorpusDictionary(path);
  }

  public String checkDic(String c2, SeqClassifierFlags flags) {
    if (cd.getW(c2).equals("1")) {
      return "1";
    } 
    return "0";
  }

View on GitHub (pinned to 1b7edd19c4)