[WRONG augmentation] pos_list.append(token.annotation_layers["pos"][0]._value) ->KeyError: 'pos'

[code-chendl]

I run the following code in Colab:

#pip3 install textattack[tensorflow]

from textattack.augmentation import CLAREAugmenter

augmenter = CLAREAugmenter(pct_words_to_swap=0.2, transformations_per_example=5)
s = "I'd love to go to Japan but the tickets are 500 dollars"
augmenter.augment(s)

and got the following error

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-9-7c7856df0718> in <module>()
      3 augmenter = CLAREAugmenter(pct_words_to_swap=0.2, transformations_per_example=5)
      4 s = "I'd love to go to Japan but the tickets are 500 dollars"
----> 5 augmenter.augment(s)

6 frames
/usr/local/lib/python3.7/dist-packages/textattack/shared/utils/strings.py in zip_flair_result(pred, tag_type)

    235         word_list.append(token.text)
    236         if "pos" in tag_type:
--> 237             pos_list.append(token.annotation_layers["pos"][0]._value)
    238         elif tag_type == "ner":
    239             pos_list.append(token.get_tag("ner"))

KeyError: 'pos'

System Information

Python==3.7.0

Name: textattack
Version: 0.3.8
Summary: A library for generating text adversarial examples
Home-page: https://github.com/QData/textattack
Author: QData Lab at the University of Virginia
Author-email: [email protected]
License: MIT
Location: d:\tools\anaconda3\envs\textattack\lib\site-packages
Requires: bert-score, click, datasets, editdistance, filelock, flair, jieba, language-tool-python, lemminflect, lru-dict, more-itertools, nltk, num2words, numpy, OpenHowNet, pandas, pinyin, pycld2, PySocks, scipy, terminaltables, torch, tqdm, transformers, word2number

[Idris-MAROUF]

In the strings.py file, change the line 237 which is "pos_list.append(token.annotation_layers["pos"][0]._value)" to "pos_list.append(token.annotation_layers["upos"][0]._value)", restart runtime and it should work

[code-chendl]

In the strings.py file, change the line 237 which is "pos_list.append(token.annotation_layers["pos"][0]._value)" to "pos_list.append(token.annotation_layers["upos"][0]._value)", restart runtime and it should work

Thanks! The problem was successfully solved in this way.

[code-chendl]

As for the second issue proposed in https://github.com/QData/TextAttack/issues/713#issue-1574652467, I replace self._enptb_to_universal with an array with the relevant universal POS tags, in the \textattack\transformations\word_swaps\word_swap_inflections.py file, as the following code.

        self._enptb_to_universal = {
          #-----dictionary with new tags---------
            "PUNCT": ".", 
            "CCONJ": "CONJ",
            "SCONJ": "CONJ",
            "PROPN": "NOUN",
            "PART": "PRT",
            "AUX": "VERB",
            "SYM": "NOUN",
            "INTJ":"X", 
           #----original dictionary below----------
            "JJRJR": "ADJ",
            "VBN": "VERB",
            "VBP": "VERB",
            "JJ": "ADJ",
            "VBZ": "VERB",
            "VBG": "VERB",
            "NN": "NOUN",
            "VBD": "VERB",
            "NP": "NOUN",
            "NNP": "NOUN",
            "VB": "VERB",
            "NNS": "NOUN",
            "VP": "VERB",
            "TO": "VERB",
            "MD": "VERB",
            "NNPS": "NOUN",
            "JJS": "ADJ",
            "JJR": "ADJ",
            "RB": "ADJ",
        }

(Mapping info: https://github.com/slavpetrov/universal-pos-tags and https://zhuanlan.zhihu.com/p/427520069.)

[AdvancedQxj]

In the strings.py file, change the line 237 which is "pos_list.append(token.annotation_layers["pos"][0]._value)" to "pos_list.append(token.annotation_layers["upos"][0]._value)", restart runtime and it should work

When modifying code in this way, although the PSO attack method can run, the effect is poor.