/Causality-of-lexically-ambiguous-phrases

Causality of lexically ambiguous phrases

Description of the dataset

The dataset consists on 2 JSON files:

  • SV.json: Subject-Verb phrases
  • VO.json: Verb-Object phrases

Keys

Each of these files is of the form:

{
    phrase_1: {
        "model": array,
        "type": str,
        "A_to_B": float,
        "B_to_A": float,
        "A___B": float
    },
    phrase_2: {
        "model": array,
        "type": str,
        "A_to_B": float,
        "B_to_A": float,
        "A___B": float
    },
    ...
}

The keys phrase_k is used to extract the contexts of the empirical model. For example letter_press_admit_bury will have the following contexts:

  1. (letter, admit)
  2. (letter, bury)
  3. (press, admit)
  4. (press, bury)

(in this order)

The key model is used for denoting the empirical model obtain from AmazonMTurk as follows:

[
    [
        P[output=(0,0) | context 1],
        P[output=(0,1) | context 1],
        P[output=(1,0) | context 1],
        P[output=(1,1) | context 1]
    ],
    [
        P[output=(0,0) | context 2],
        P[output=(0,1) | context 2],
        P[output=(1,0) | context 2],
        P[output=(1,1) | context 2]
    ],
    [
        P[output=(0,0) | context 3],
        P[output=(0,1) | context 3],
        P[output=(1,0) | context 3],
        P[output=(1,1) | context 3]
    ],
    [
        P[output=(0,0) | context 4],
        P[output=(0,1) | context 4],
        P[output=(1,0) | context 4],
        P[output=(1,1) | context 4]
    ]
]

The key type stores the levels of ambiguity of the different words (M corresponding to words with multiple meanings and S to words with multiple senses). For example in the model letter_press_admit_bury with type: "SMMS":

  • letter has multiple senses (S)
  • press has multiple meanings (M)
  • admit has multiple meanings (M)
  • bury has multiple sense (S)

The key A_to_B corresponds to the causal fraction associated with the order A -> B, e.g. S_to_V corresponds to the causal order S -> V

Similarly, B_to_Acorresponds to the causal fraction associated with the order B -> A.

Finally, the key A___B corresponds to the no-signalling fraction.