sapienzanlp/mcl-wic

Semeval-2021 Multilingual and Cross-lingual Word-in-Context Task

mcl-wic's Issues

Wrong indexes

Hi!
It seems to me that some indexes of the target tokens are wrong.
I found problem in the next instances:
ru-ru: 0, 4
en-en: 2, 3
en-ru: 13
And may be the other pairs of languages have the same problem

Examples:
trial.en-en.2, resolution, NOUN, en, 17, en, 8, 'Ability to detect vertical movements , variations and dislocations of ground areas to a resolution of a few centimetres ;',
'Although Iraq was fully implementing all United Nations resolutions , it continued to suffer the devastating effects of the sanctions .'

trial.ru-ru.4, год, NOUN, ru, 3, ru, 12, 'В мае 1987 года на вершине горы Окубо в 4,1 км к западу от пусковой площадки был построен новый телеметрический центр.',
'Так , более 100 000 пуэрториканцев приняли участие в демонстрации 14 июля прошлого года в знак протеста против присоединения .'

Regarding the mode of communication

We wanted to know what is the main channel of communication for the task(We had few queries about the task) as we are unable to find the google group mentioned on the official SemEval page. Also, we wanted to know when will the training dataset be released.

Labels wrongly include numbers

Hi,
It seems to me that some of the labels for the Chinese and Arabic multilingual trial data are wrong, because they include numbers in addition to the label e.g.
trial.zh-zh.0 3F

List of all ids where this is the case:

• trial.ar-ar.0 2T
• trial.ar-ar.1 3T
• trial.ar-ar.2 2T
• trial.ar-ar.3 3T
• trial.ar-ar.4 2F
• trial.ar-ar.5 3F
• trial.ar-ar.6 2T
• trial.ar-ar.7 3T
• trial.ar-ar.8 2T
• trial.ar-ar.9 3T
• trial.zh-zh.0 3F
• trial.zh-zh.1 3F
• trial.zh-zh.2 2T
• trial.zh-zh.3 3T
• trial.zh-zh.4 2T
• trial.zh-zh.5 3T
• trial.zh-zh.6 2R
• trial.zh-zh.7 2R
• trial.zh-zh.8 3R
• trial.zh-zh.9 3R

POS tags with extra trailing suffixes among the training instances

Hi!
There are 4 instances in the training data collection, where extra string follow the POS tags.
These are instances at indices 2000, 2001, 3396 and 3397 in the list of training instances.
Can one just safely remove these trailing suffixes from these POS tags or are these instances in a pending status where updates could still happen in a later release?