What causes "%s: Splicing in DUMMYTAG for %s" in stanfordnlp/CoreNLP?

Accept the repair: DUMMYTAG is spliced automatically and the tree is usable downstream

%s: Splicing in DUMMYTAG for %s

Error message

%s: Splicing in DUMMYTAG for %s

What it means

A warning from ArabicTreeNormalizer.normalizeWholeTree handling malformed phrasal nodes whose children include bare leaves directly under a non-preterminal (Arabic treebanks contain nodes like '/' with no preterminal). The normalizer logs '<class>: Splicing in DUMMYTAG for <node>' and inserts a synthetic 'DUMMYTAG' preterminal above the leaf so the tree structure becomes valid. This is automatic repair with a warning, not a failure.

Solutions

Accept the repair: DUMMYTAG is spliced automatically and the tree is usable downstream
Pre-scan for leaves under non-preterminals and insert a real POS tag instead, so output has no DUMMYTAG nodes
Filter out or hand-fix the malformed nodes flagged by the warning before parsing/training
If downstream code chokes on DUMMYTAG, post-process the normalized tree replacing DUMMYTAG with an appropriate tag

Example fix

// before
// raw ATB tree: (NP /. ) leaf directly under NP -> DUMMYTAG spliced
// after: pre-insert a real tag
// (NP (PUNC .)) via tree surgery before normalizeWholeTree
for (Tree p : tree) {
  if (!p.isPreTerminal() && !p.isLeaf()) {
    for (Tree c : p.children()) {
      if (c.isLeaf()) p.setChildren(new Tree[]{ tf.newTreeNode("PUNC", Collections.singletonList(c)) });
    }
  }
}

Defensive patterns

Strategy: fallback

Validate before calling

// detect leaves directly under non-preterminals
boolean hasBareLeaves = false;
for (Tree n : tree) {
  if (!n.isPreTerminal() && !n.isLeaf()) {
    for (Tree c : n.children()) if (c.isLeaf()) hasBareLeaves = true;
  }
}

Try / catch

try {
  Tree out = normalizer.normalizeWholeTree(tree, tf);
  if (out.toString().contains("DUMMYTAG")) log.warn("DUMMYTAG present; replace with real tag for training");
} catch (Exception e) {
  log.warn("Normalization failed", e);
}

Prevention

Pre-insert real POS tags where ATB trees omit preterminals
Post-scan normalized trees for DUMMYTAG if you train models on them
Use consistent tree-conversion tooling to avoid malformed structures

When it happens

Trigger: Running normalizeWholeTree on trees where a phrasal node has a Leaf child directly (missing POS layer) — seen in raw ATB trees and trees produced by some segmentation/tokenization tools.

Common situations: Arabic Treebank preprocessing for the Arabic parser; trees converted from other corpora formats that omit preterminals under certain punctuation or elided nodes.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/2e1fc0974cc53c5b. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/trees/international/arabic/ArabicTreeNormalizer.java:175

        }

      } else if (t.isPreTerminal()) {

        if (t.value() == null || t.value().isEmpty()) {
          log.warn(String.format("%s: missing tag for %s",this.getClass().getName(),t.pennString()));
        } else if(t.label() instanceof HasTag) {
          ((HasTag) t.label()).setTag(t.value());
        }

      } else { //Phrasal nodes

        // there are some nodes "/" missing preterminals.  We'll splice in a tag for these.
        int nk = t.numChildren();
        List<Tree> newKids = new ArrayList<>(nk);
        for (int j = 0; j < nk; j++) {
          Tree child = t.getChild(j);
          if (child.isLeaf()) {
            log.warn(String.format("%s: Splicing in DUMMYTAG for %s",this.getClass().getName(),t.toString()));
            newKids.add(tf.newTreeNode("DUMMYTAG", Collections.singletonList(child)));

          } else {
            newKids.add(child);
          }
        }
        t.setChildren(newKids);
      }
    } //Every node in the tree has now been processed

    //
    // Additional processing for specific phrasal annotations
    //

    // special global coding for moving PRD annotation from constituent to verb tag.
    if (markPRDverb) {
      TregexMatcher m = prdVerbPattern.matcher(tree);
      Tree match = null;

View on GitHub (pinned to 1b7edd19c4)