Part of Speech for Telugu (pos_mtg)

Description

A Part of Speech classifier predicts a grammatical label for every token in the input text. Implemented with an averaged perceptron architecture.

Predicted Entities

  • ADJ
  • ADP
  • ADV
  • AUX
  • CCONJ
  • DET
  • NOUN
  • NUM
  • PART
  • PRON
  • PROPN
  • PUNCT
  • VERB
  • X

Live Demo Open in Colab Download

How to use

document_assembler = DocumentAssembler()
  .setInputCol("text")
  .setOutputCol("document")

sentence_detector = SentenceDetector()
  .setInputCols(["document"])
  .setOutputCol("sentence")

pos = PerceptronModel.pretrained("pos_mtg", "te")
  .setInputCols(["document", "token"])
  .setOutputCol("pos")

pipeline = Pipeline(stages=[
  document_assembler,
  sentence_detector,
  posTagger
])

example = spark.createDataFrame(pd.DataFrame({'text': ["ఆయన వస్తున్నారా , లేదా ?"]}))

result = pipeline.fit(example).transform(example)
val document_assembler = DocumentAssembler()
        .setInputCol("text")
        .setOutputCol("document")

val sentence_detector = SentenceDetector()
        .setInputCols(["document"])
.setOutputCol("sentence")

val pos = PerceptronModel.pretrained("pos_mtg", "te")
        .setInputCols(Array("document", "token"))
        .setOutputCol("pos")

val pipeline = new Pipeline().setStages(Array(document_assembler, sentence_detector, pos))

val result = pipeline.fit(Seq.empty["ఆయన వస ,  ?"].toDS.toDF("text")).transform(data)

import nlu
text = [""ఆయన వసతుననారా , లేదా ?""]
token_df = nlu.load('te.pos.mtg').predict(text)
token_df

Results

+------------------------+--------------------------------+
|text                    |result                          |
+------------------------+--------------------------------+
|ఆయన వస్తున్నారా , లేదా ?|[PRON, VERB, PUNCT, VERB, PUNCT]|
+------------------------+--------------------------------+

Model Information

Model Name: pos_mtg
Compatibility: Spark NLP 2.7.5+
License: Open Source
Edition: Official
Input Labels: [sentence, token]
Output Labels: [pos]
Language: te

Data Source

The model was trained on the Universal Dependencies data set.

Benchmarking

|              | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|---------|
| ADJ          | 0.50      | 0.40   | 0.44     | 5       |
| ADP          | 0.75      | 0.43   | 0.55     | 7       |
| ADV          | 0.78      | 0.68   | 0.72     | 31      |
| CCONJ        | 0.00      | 0.00   | 0.00     | 1       |
| DET          | 0.89      | 0.89   | 0.89     | 18      |
| INTJ         | 0.00      | 0.00   | 0.00     | 0       |
| NOUN         | 0.82      | 0.76   | 0.79     | 171     |
| NUM          | 0.83      | 0.42   | 0.56     | 12      |
| PART         | 0.00      | 0.00   | 0.00     | 2       |
| PRON         | 0.88      | 0.93   | 0.91     | 122     |
| PROPN        | 0.69      | 0.86   | 0.77     | 21      |
| PUNCT        | 0.99      | 0.99   | 0.99     | 165     |
| SCONJ        | 0.71      | 1.00   | 0.83     | 5       |
| VERB         | 0.84      | 0.92   | 0.88     | 161     |
| accuracy     |           |        | 0.87     | 721     |
| macro avg    | 0.62      | 0.59   | 0.59     | 721     |
| weighted avg | 0.87      | 0.87   | 0.86     | 721     |