Analogies

An analogical proportion is a statement of the form "A is toB as Cis to D", written "A : B :: C : D". The underlying idea is that the relation between A and B is similar to the one between C and D.

On the one hand, analogical proportions ground analogical inference that has been used in various machine learning tasks such as classification, decision making, and automatic translation, with competitive results. On the other hand, they support analogical extrapolation that can be used to solve hard reasoning tasks, such as IQ tests. It can also be used in data augmentation, especially in learning environments with few labeled samples.

What makes reasoning with analogies special is its ability to process simultaneously similarities and dissimilarities. This characteristic establishes bridges between the two main axes of AI: knowledge representation and reasoning (KRR) and machine learning (ML). Analogical reasoning contributes to a transparent AI as it is close to human reasoning and provides explanations based on examples and counter-examples. The objective of the ANNa project is to provide an online platform to detect, solve, and reason on analogies, with multiple applications in various domains, for instance, in NLP, biomedical sciences, as well as in industry.

Morphological Anlogical Proportions

Morphological analogies are analogical proportions between words, on the morphological features of the words. For example in English, tidy : untidy :: do : undo is a morphological analogy: untidy is tidy with the prefix un-, and undo is do with the prefix un-.In other words, tidy is to untidy as do is to undo.

Learn more

ANNa-MD-Morphological Analogy Detection

We propose ANNa-MD (MD for Morphological analogy Detection), an approach using deep learning to identify morphological analogies.It encodes structural properties of analogical proportions, and relies on a specifically designed embedding model to capture morphological characteristics of words.Our model displays competitive performance on analogy detection over multiple languages.

Try ANNa-MD Online

ANNa-MR-Morphological Analogy Resolution

We propose ANNa-MR (MR for Morphological analogy Resolution), an approach using deep learning to infer the missing element of a morphological analogy. Like ANNa-MD, it encodes structural properties of analogical proportions and relies on a specifically designed embedding model to capture morphological characteristics of words.Our model displays competitive performance on analogy resolution over multiple languages.

Try ANNa-MR Online