-
AI systems are currently used in a wide variety of applications,
with several levels of societal impact, and are expected to be soon
deployed in safety-critical fields, e.g., autonomous driving. Hence,
a natural need for ethical accountability of such systems is gaining
importance. A central issue lies in designing systems whose
decisions are transparent [
6
], i.e., they must be easily interpretable by
humans, as users must be able to suitably weight and trust their
assistance. Deep neural networks are clearly problematic in this
regard: their high non-linearity, despite allowing for
state-of-theart performances in several challenging problems also amplifies
the epistemological opaqueness of the decision-flow and limits its
interpretability. The concept of transparency of a machine learning
model spans multiple definitions, focusing on diferent aspects, from
the simplicity of the model, e.g., the number of nodes in a decision
tree, to the intuitiveness of its parameters and computations [
4
]. In
this context, an important capability of an AI system is the ability
of providing post-hoc explanations in terms of evidences supporting
the provided decisions: although they usually do not formally
elucidate how a model works, post-hoc explanations often have the
nice property of being quite intuitive, conveying useful information
also to end-users without any AI or machine learning expertise [
8
].
In semantic inference tasks (e.g., text classification), an
explanation model producing post-hoc explanations should hence be able
to trace back connections between the output categories and the
semantic and syntactic properties of the input texts. Such models
should have three desired properties: semantic transparency,
informativeness w.r.t. the system decision and efectiveness in enabling
auditing processes against the system.
In this work we focus on a specific post-hoc mechanism which
is to provide, along with the prediction, a comparison with one or
more other examples, namely landmarks, that share task-relevant
linguistic properties with the input. From an argument theory
perspective, this corresponds to supporting decisions through an
“argument by analogy” schema [
9
]: a user exposed to such a kind of
argument will endow a diferent level of trust into the machine
decision according to the linguistic plausibility of the analogy. In
fact, he/she will implicitly gauge the evidence from the linguistic
properties shared between the input sentence (or its parts) and the
one used for comparison as well their importance with respect to
the output decision. Let us consider, for example, the following
prediction in question classification (QC) [
7
]: "What is the capital
of Zimbabwe?" refers to a Location. We would like the system to
motivate its decision with an argument such as: ...since it recalls me
of "What is the capital of California?" which also refers to a Location.
Notice that a decision explaining task is quite diferent from
relevance ranking, and semantic similarity plays here a minor role:
clear and trustful analogies may exist between training examples
that are semantically diferent but such that their properties imply
similar causal relationships between the input and the decision.
Recent work has been inspired by eforts in improving model’s
interpretability in image processing tasks, in particular by the
Layerwise Relevance Propagation (LRP) [
3
]. In LRP, the classification
decision of a deep neural network is decomposed backward across
the network layers and evidence about the contribution to the final
decision brought by individual input fragments (i.e., pixels of the
input image) is gathered. We propose here to extend the LRP
application to a linguistically motivated network architecture, known as
Kernel-Based Deep Architecture (KDA) [
5
], which frames semantic
information captured by linguistic Tree Kernel [2] methods within
the neural-based learning paradigm. The result is a mechanism that,
for each system’s prediction such as in question classification,
generates an argument-by-analogy explanation based on real training
examples, not necessarily similar to the input.
We also propose here a novel approach to evaluate numerically
the interpretability of any explanation-enriched model applied in
semantic inference tasks. By defining a specific audit process, we
derive a synthetic metric, i.e. Auditing Accuracy, that takes into
account the properties of transparency, informativeness and
efectiveness. The evaluation of the proposed methodology shows the
meaningful impact of LRP-based explanation models: users faced
with explanations are systematically oriented to accept (or reject)
the system decisions, so that post-hoc judgments may even help in
improving the overall application accuracy.
This work has been accepted for publication at the 2019 Conference
on Empirical Methods in Natural Language Processing (EMNLP2019
[1]).