Table 3.

Analysis of the etymology of the terms related to interpretability

ID	Word	Etymology		ML Definition
1	Interpretability, Interpretable	From late Latin interpretabilitis from Latin interprĕtor, interprĕtāri (to interpret)	To interpret, comment, explain, expose, illustrate, to translate	To translate, expose, and comment on the generation process of one or multiple ML systems outcomes, making the overall process understandable by a human
2	Explainability, Explainable	From 1600 use of explain + -able adapted from Latin explāno, explānāre	To explain, clarify, expose, illustrate, state clearly	To indicate with precision, to illustrate what features or high-level concepts were used by the ML system to generate predictions for one or multiple inputs. In intelligent agent systems: possibly iterative process of symbolic knowledge manipulation to make it interpretable
3	Transparency, Transparent	Medieval Latin adaptation of the words trans (on the other side) and pārĕo, pārēre (to appear, to show)	To see through	A transparent ML system has a non-opaque output-generation process where the role of the individual components, the learned paradigms, and the overall behavior of the model are known and can be simulated by a human user
4	Intelligibility, Intelligible	From Latin intellegibilis, intellegibilis, II class adjective	To understand, comprehend, decipher	An intelligible ML system is an understandable system with inherent interpretability
5	Accountability, Accountable.	From 1770 use of accountable + -ity, adapted from Old French acont derived from Latin compŭto, compŭtāre, which has multiple meanings including to count, to estimate, to judge and to believe.	Used from the 1610s with the sense of “rendering an account", meaning providing a statement answering for conduct.	An accountable ML system is expected to justify its outcomes and behavior
6	Reliability, Reliable	From Scottish of the 1560s “raliabill", derived from Old French relier a derivation of the Latin rĕlĭgo, rĕlĭgāre (meaning to tie, to bind).	From the 1570s used with the sense of to depend, to trust, typically used in the expression “to rely on something/someone".	To be consistently good and be worthy of trust
7	Auditability, Auditable	From Latin noun auditŭs, auditŭs	The sense of hearing, the act of hearing, audition. Used in the sense of official audience, judicial hearing or examination.	An “auditable" ML system should provide information on how to perform an official audience of the model. For example, this can be done by providing extra documentation and functionalities.
8	Liability, liable	From Anglo-French liable, derived from Latin lĭgo, lĭgāre (to tie, to bind)	Legal responsibility for acts.	Legal liability of a product implementing ML, particularly in the case where something goes wrong
9	Robustness, Robust	From French robuste, derived from Latin robustus, robustum.	The literal meaning is oaken, made of oak. Used in the figurative sense of strong, vigorous and resistant.	Robust ML systems are resistant, secure and reliable. Providing consistent results also in case of adversarial attacks, variations in the dataset, domain shifts, and outliers