# Elementwise Language Representation

## TL;DR

This paper introduces elementwise embedding, a novel language representation technique that aligns character-level elements across semantic units within transformer models, enabling longer sequence processing and improved performance with fewer parameters.

## Contribution

The paper presents a new elementwise embedding method that generalizes transformer embeddings to all language levels without architectural changes, improving efficiency and robustness.

## Key findings

- Outperforms subword models in patent classification
- Enables processing longer sequences with same complexity
- Reduces embedding parameters to 0.005% of traditional models

## Abstract

We propose a new technique for computational language representation called elementwise embedding, in which a material (semantic unit) is abstracted into a horizontal concatenation of lower-dimensional element (character) embeddings. While elements are always characters, materials are arbitrary levels of semantic units so it generalizes to any type of tokenization. To focus only on the important letters, the $n^{th}$ spellings of each semantic unit are aligned in $n^{th}$ attention heads, then concatenated back into original forms creating unique embedding representations; they are jointly projected thereby determining own contextual importance. Technically, this framework is achieved by passing a sequence of materials, each consists of $v$ elements, to a transformer having $h=v$ attention heads. As a pure embedding technique, elementwise embedding replaces the $w$-dimensional embedding table of a transformer model with $256$ $c$-dimensional elements (each corresponding to one of UTF-8 bytes) where $c=w/v$. Using this novel approach, we show that the standard transformer architecture can be reused for all levels of language representations and be able to process much longer sequences at the same time-complexity without "any" architectural modification and additional overhead. BERT trained with elementwise embedding outperforms its subword equivalence (original implementation) in multilabel patent document classification exhibiting superior robustness to domain-specificity and data imbalance, despite using $0.005\%$ of embedding parameters. Experiments demonstrate the generalizability of the proposed method by successfully transferring these enhancements to differently architected transformers CANINE and ALBERT.