stanfordnlp/CoreNLP · error · ForwardPropagationException

SentimentCostAndGradient: Tree not correctly binarized:...

Error message

SentimentCostAndGradient: Tree not correctly binarized:...

What it means

After classifying a node, the code checks tree structure/labels (e.g. binarization sanity around top-level constituents and CoreLabel types) and throws ForwardPropagationException with a descriptive error buffer when the tree violates model assumptions, here: not correctly binarized / too many top-level constituents.

Solutions

Binarize trees before passing them to the sentiment model (use the same preprocessing as ReadSentimentDataset/convertTrees)
Check the root has exactly one or two children as the model expects
Validate tree structure programmatically before training

Example fix

// before
(ROOT (S (NP ...) (VP ...) (PP ...))) // 3+ branches
// after
(ROOT (S (NP ...) (VP (VP ...) (PP ...)))) // binarized

Defensive patterns

Strategy: validation

Validate before calling

if (root.children().length > 2) throw new IllegalStateException("Tree not correctly binarized: root has " + root.children().length + " children");

Try / catch

try { forwardPropagate(tree); } catch (ForwardPropagationException e) { if (e.getMessage().contains("not correctly binarized")) { binarizeAndRetry(tree); } else throw e; }

Prevention

Binarize all trees at dataset load time
Assert binary branching across the whole tree before training
Keep parser output out of the training pipeline until binarized

When it happens

Trigger: Calling forwardPropagate on a tree with more than two children at the root or multiple top-level constituents, i.e. a tree that was never binarized for the sentiment model.

Common situations: Feeding raw parser output or non-binarized treebank trees directly into SentimentTraining/ExternalEvaluate pipelines.

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/43a26fb7fef61a97. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/sentiment/SentimentCostAndGradient.java:540

      if (model.op.useTensors) {
        SimpleTensor tensor = model.getBinaryTensor(leftCategory, rightCategory);
        SimpleMatrix tensorIn = NeuralUtils.concatenate(leftVector, rightVector);
        SimpleMatrix tensorOut = tensor.bilinearProducts(tensorIn);
        nodeVector = NeuralUtils.elementwiseApplyTanh(W.mult(childrenVector).plus(tensorOut));
      } else {
        nodeVector = NeuralUtils.elementwiseApplyTanh(W.mult(childrenVector));
      }
    } else {
      StringBuilder error = new StringBuilder();
      error.append("SentimentCostAndGradient: Tree not correctly binarized:\n   ");
      error.append(tree);
      error.append("\nToo many top level constituents present: ");
      error.append("(" + tree.value());
      for (Tree child : tree.children()) {
        error.append(" (" + child.value() + " ...)");
      }
      error.append(")");
      throw new ForwardPropagationException(error.toString());
    }

    SimpleMatrix predictions = NeuralUtils.softmax(classification.mult(NeuralUtils.concatenateWithBias(nodeVector)));

    int index = getPredictedClass(predictions);
    if (!(tree.label() instanceof CoreLabel)) {
      log.info("SentimentCostAndGradient: warning: No CoreLabels in nodes: " + tree);
      throw new AssertionError("Expected CoreLabels in the nodes");
    }
    CoreLabel label = (CoreLabel) tree.label();
    label.set(RNNCoreAnnotations.Predictions.class, predictions);
    label.set(RNNCoreAnnotations.PredictedClass.class, index);
    label.set(RNNCoreAnnotations.NodeVector.class, nodeVector);
  } // end forwardPropagateTree

}

View on GitHub (pinned to 1b7edd19c4)