yes-no-classifiers

Yes/No answer polarity classifiers (ONNX) for 43 languages, trained on TigreGotico/yes-no-multilingual (200 samples/language, 8,600 total).

Used by little_questions to detect whether a statement is an affirmative, negative, or uncertain answer.

Classes

yes, no, maybe

Models

A multilingual model (yesno_svm_cal_multilingual_0.9.0.onnx) covers all 43 languages and ships bundled with little_questions. Per-language models achieve 90-96% macro F1 on their own language; the multilingual model achieves 84%.

Per-language models

File
yesno_svm_cal_AN_0.9.0.onnx
yesno_svm_cal_AR_0.9.0.onnx
yesno_svm_cal_BG_0.9.0.onnx
yesno_svm_cal_CA_0.9.0.onnx
yesno_svm_cal_CS_0.9.0.onnx
yesno_svm_cal_DA_0.9.0.onnx
yesno_svm_cal_DE_0.9.0.onnx
yesno_svm_cal_EL_0.9.0.onnx
yesno_svm_cal_EN_0.9.0.onnx
yesno_svm_cal_ES_0.9.0.onnx
yesno_svm_cal_ET_0.9.0.onnx
yesno_svm_cal_EU_0.9.0.onnx
yesno_svm_cal_FA_0.9.0.onnx
yesno_svm_cal_FIL_0.9.0.onnx
yesno_svm_cal_FI_0.9.0.onnx

Inference

import onnxruntime as rt, numpy as np

sess = rt.InferenceSession("yesno_svm_cal_multilingual_0.9.0.onnx")
inp = np.array(["Yes, of course!"], dtype=object)
label, probs = sess.run(None, {"input": inp})
print(label[0])   # yes