{"url":"/dataset/morph-call","name":"Morph Call","full_name":null,"description_markdown":"Morph Call is a suite of 46 probing tasks for four Indo-European languages that fall under different morphology: Russian, French, English, and German. The tasks are designed to explore the morphosyntactic content of multilingual transformers which is a less studied aspect at the moment.\r\n\r\nThe tasks are divided into four groups:\r\n\r\n* [Morphosyntactic Features](https://github.com/morphology-probing/morph-call/tree/main/data/morphosyntactic_features): probe the encoder for the occurrence of the morphosyntactic properties.\r\n* Masked Token: analogous to [Morphosyntactic Features](https://github.com/morphology-probing/morph-call/tree/main/data/morphosyntactic_features) with the exception that the target word is replaced with a tokenizer-specific mask token.\r\n* [Morphosyntactic Values](https://github.com/morphology-probing/morph-call/tree/main/data/morphosyntactic_values): is a group of k-way classification tasks for each feature where *k* is the number of values that the feature can take.\r\n* [Perturbations](https://github.com/morphology-probing/morph-call/tree/main/data/perturbations): tasks test the encoder sensitivity to syntactic and inflectional sentence perturbations.\r\n\r\n## Probing Methods\r\n\r\n* [Supervised probing](https://github.com/morphology-probing/morph-call/tree/main/probing) involves training a Logistic Regression classifier to predict a property. The performance is used as a proxy to evaluate the model knowledge.\r\n* [Neuron-level Analysis](https://github.com/fdalvi/NeuroX) [Durrani et al., 2020] allows retrieving a group of individual neurons that are most relevant to predict a linguistic property.\r\n* [Contextual Correlation Analysis](https://github.com/johnmwu/contextual-corr-analysis/tree/master) [Wu et al., 2020] is a representation-level similarity measure that allows identifying pairs of layers of similar behavior. \r\n\r\n## Usage\r\nWe provide an [example](https://github.com/morphology-probing/morph-call/blob/main/examples/case-category-masks-probing.ipynb) of the experiment on **Masked Token** task (Case, German).\r\n\r\n```\r\nbash\r\nme@my-laptop:~$ python3 probe.py --help\r\nINFO: Showing help with the command 'probe.py -- --help'.\r\n\r\nNAME\r\n    probe.py - configure the experiment AND perform probing\r\n\r\nSYNOPSIS\r\n    probe.py <flags>\r\n\r\nDESCRIPTION\r\n    configure the experiment AND perform probing\r\n\r\nFLAGS\r\n    --results_path=RESULTS_PATH\r\n        Type: Optional[str]\r\n        Default: None\r\n        path to a folder to store the probing results and the model intermediate activations\r\n    --model_architecture=MODEL_ARCHITECTURE\r\n        Type: typ...\r\n        Default: 'bert multilingual'\r\n    --model_is_finetuned=MODEL_IS_FINETUNED\r\n        Type: bool\r\n        Default: False\r\n        if to perform the experiment on the fine-tuned model\r\n    --model_finetuned_path=MODEL_FINETUNED_PATH\r\n        Type: Optional[str]\r\n        Default: None\r\n        (only if model_is_finetuned is True) path to store the fine-tuned model\r\n    --model_finetuned_config_google_url=MODEL_FINETUNED_CONFIG_GOOGLE_URL\r\n        Type: Optional[]\r\n        Default: None\r\n        (only if model_is_finetuned is True) the url of the fine-tuned model config if to be downloaded\r\n    --model_finetuned_model_google_url=MODEL_FINETUNED_MODEL_GOOGLE_URL\r\n        Type: Optional[]\r\n        Default: None\r\n        (only if model_is_finetuned is True) the url of the fine-tuned model weights if to be downloaded\r\n    --model_is_random=MODEL_IS_RANDOM\r\n        Type: bool\r\n        Default: False\r\n        if to perform the random initialization of the model\r\n    --layers_to_probe=LAYERS_TO_PROBE\r\n        Type: List\r\n        Default: 'all'\r\n        (either \"all\" or list w. possible numbers from 0 to 11) -- model layers to probe. e.g.: [1, 3, 11], or \"all\"\r\n    --train_n_sentences=TRAIN_N_SENTENCES\r\n        Type: int\r\n        Default: 1500\r\n        number of sentences used to train the probing classifier\r\n    --test_n_sentences=TEST_N_SENTENCES\r\n        Type: int\r\n        Default: 1000\r\n        number of sentences used to evaluate the probing classifier\r\n    --dev_n_sentences=DEV_N_SENTENCES\r\n        Type: int\r\n        Default: 0\r\n        DEPRECATED\r\n```","description_withheld":null,"homepage":"https://github.com/morphology-probing/morph-call","introduced_date":"2021-04-26","introduced_date_note":null,"introduced_by":{"paper":"/paper/morph-call-probing-morphosyntactic-content-of","title":"Morph Call: Probing Morphosyntactic Content of Multilingual Transformers","first_author":"Vladislav Mikhailov","url":null},"license":null,"modalities":[{"name":"Texts","url":"/datasets/modality/texts"}],"tasks":[{"name":"Interpretable Machine Learning","url":"/task/interpretable-machine-learning","datasets_with_task":"/datasets/task/interpretable-machine-learning"}],"languages":[{"name":"English","url":"/datasets/language/english"},{"name":"French","url":"/datasets/language/french"},{"name":"German","url":"/datasets/language/german"},{"name":"Russian","url":"/datasets/language/russian"}],"variants":["Morph Call"],"data_loaders":[],"num_papers_in_archive":2,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[],"papers_with_a_benchmark_row":[],"syntology_totals":{"read_at":"2026-09-24T18:15:14+00:00","papers_with_samples":0,"samples_harvested":0,"samples_ran":0,"samples_unverified":0,"pointer_only_for_licence":0,"papers_with_no_sample_that_ran":0,"note":"the per-paper counts above, summed; not a rate"},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks."}