Universal NER: A Gold-Standard Multilingual Named Entity Recognition Benchmark

We introduce Universal NER (UNER), an open, community-driven project to develop gold-standard NER benchmarks in many languages. The overarching goal of UNER is to provide high-quality, cross-lingually consistent annotations to facilitate and standardize multilingual NER research. UNER v1 contains 18 datasets annotated with named entities in a cross-lingual consistent schema across 12 diverse languages. In this paper, we detail the dataset creation and composition of UNER; we also provide initial modeling baselines on both in-language and cross-lingual learning settings. We release the data, code, and fitted models to the public.

Analyzing Cognitive Plausibility of Subword Tokenization

Subword tokenization has become the de-facto standard for tokenization, although comparative evaluations of subword vocabulary quality across languages are scarce. Existing evaluation studies focus on the effect of a tokenization algorithm on the performance in downstream tasks, or on engineering criteria such as the compression rate. We present a new evaluation paradigm that focuses on the cognitive plausibility of subword tokenization. We analyze the correlation of the tokenizer output with the response time and accuracy of human performance on a lexical decision task. We compare three tokenization algorithms across several languages and vocabulary sizes. Our results indicate that the UnigramLM algorithm yields less cognitively plausible tokenization behavior and a worse coverage of derivational morphemes, in contrast with prior work.

Emptying the Ocean with a Spoon: Should We Edit Models?

We call into question the recently popularized method of direct model editing as a means of correcting factual errors in LLM generations. We contrast model editing with three similar but distinct approaches that pursue better defined objectives: (1) retrieval-based architectures, which decouple factual memory from inference and linguistic capabilities embodied in LLMs; (2) concept erasure methods, which aim at preventing systemic bias in generated text; and (3) attribution methods, which aim at grounding generations into identified textual sources. We argue that direct model editing cannot be trusted as a systematic remedy for the disadvantages inherent to LLMs, and while it has proven potential in improving model explainability, it opens risks by reinforcing the notion that models can be trusted for factuality. We call for cautious promotion and application of model editing as part of the LLM deployment process, and for responsibly limiting the use cases of LLMs to those not relying on editing as a critical component.

Scope is all you need: Transforming LLMs for HPC Code

With easier access to powerful compute resources, there is a growing trend in the field of AI for software development to develop larger and larger language models (LLMs) to address a variety of programming tasks. Even LLMs applied to tasks from the high-performance computing (HPC) domain are huge in size (e.g., billions of parameters) and demand expensive compute resources for training. We found this design choice confusing - why do we need large LLMs trained on natural languages and programming languages unrelated to HPC for HPC-specific tasks? In this line of work, we aim to question design choices made by existing LLMs by developing smaller LLMs for specific domains - we call them domain-specific LLMs. Specifically, we start off with HPC as a domain and propose a novel tokenizer named Tokompiler, designed specifically for preprocessing code in HPC and compilation-centric tasks. Tokompiler leverages knowledge of language primitives to generate language-oriented tokens, providing a context-aware understanding of code structure while avoiding human semantics attributed to code structures completely. We applied Tokompiler to pre-train two state-of-the-art models, SPT-Code and Polycoder, for a Fortran code corpus mined from GitHub. We evaluate the performance of these models against the conventional LLMs. Results demonstrate that Tokompiler significantly enhances code completion accuracy and semantic understanding compared to traditional tokenizers in normalized-perplexity tests, down to ~1 perplexity score. This research opens avenues for further advancements in domain-specific LLMs, catering to the unique demands of HPC and compilation tasks.

Advising OpenMP Parallelization via a Graph-Based Approach with Transformers

There is an ever-present need for shared memory parallelization schemes to exploit the full potential of multi-core architectures. The most common parallelization API addressing this need today is OpenMP. Nevertheless, writing parallel code manually is complex and effort-intensive. Thus, many deterministic source-to-source (S2S) compilers have emerged, intending to automate the process of translating serial to parallel code. However, recent studies have shown that these compilers are impractical in many scenarios. In this work, we combine the latest advancements in the field of AI and natural language processing (NLP) with the vast amount of open-source code to address the problem of automatic parallelization. Specifically, we propose a novel approach, called OMPify, to detect and predict the OpenMP pragmas and shared-memory attributes in parallel code, given its serial version. OMPify is based on a Transformer-based model that leverages a graph-based representation of source code that exploits the inherent structure of code. We evaluated our tool by predicting the parallelization pragmas and attributes of a large corpus of (over 54,000) snippets of serial code written in C and C++ languages (Open-OMP-Plus). Our results demonstrate that OMPify outperforms existing approaches, the general-purposed and popular ChatGPT and targeted PragFormer models, in terms of F1 score and accuracy. Specifically, OMPify achieves up to 90% accuracy on commonly-used OpenMP benchmark tests such as NAS, SPEC, and PolyBench. Additionally, we performed an ablation study to assess the impact of different model components and present interesting insights derived from the study. Lastly, we also explored the potential of using data augmentation and curriculum learning techniques to improve the model's robustness and generalization capabilities.

MPI-rical: Data-Driven MPI Distributed Parallelism Assistance with Transformers

Automatic source-to-source parallelization of serial code for shared and distributed memory systems is a challenging task in high-performance computing. While many attempts were made to translate serial code into parallel code for a shared memory environment (usually using OpenMP), none has managed to do so for a distributed memory environment. In this paper, we propose a novel approach, called MPI-rical, for automated MPI code generation using a transformer-based model trained on approximately 25,000 serial code snippets and their corresponding parallelized MPI code out of more than 50,000 code snippets in our corpus (MPICodeCorpus). To evaluate the performance of the model, we first break down the serial code to MPI-based parallel code translation problem into two sub-problems and develop two research objectives: code completion defined as given a location in the source code, predict the MPI function for that location, and code translation defined as predicting an MPI function as well as its location in the source code. We evaluate MPI-rical on MPICodeCorpus dataset and on real-world scientific code benchmarks and compare its performance between the code completion and translation tasks. Our experimental results show that while MPI-rical performs better on the code completion task than the code translation task, the latter is better suited for real-world programming assistance, in which the tool suggests the need for an MPI function regardless of prior knowledge. Overall, our approach represents a significant step forward in automating the parallelization of serial code for distributed memory systems, which can save valuable time and resources for software developers and researchers. The source code used in this work, as well as other relevant sources, are available at: https://github.com/Scientific-Computing-Lab-NRCN/MPI-rical

Incorporating Context into Subword Vocabularies

Most current popular subword tokenizers are trained based on word frequency statistics over a corpus, without considering information about co-occurrence or context. Nevertheless, the resulting vocabularies are used in language models' highly contextualized settings. We present SaGe, a tokenizer that tailors subwords for their downstream use by baking in the contextualized signal at the vocabulary creation phase. We show that SaGe does a better job than current widespread tokenizers in keeping token contexts cohesive, while not incurring a large price in terms of encoding efficiency or domain robustness. SaGe improves performance on English GLUE classification tasks as well as on NER, and on Inference and NER in Turkish, demonstrating its robustness to language properties such as morphological exponence and agglutination.

Lost in Space Marking

We look at a decision taken early in training a subword tokenizer, namely whether it should be the word-initial token that carries a special mark, or the word-final one. Based on surface-level considerations of efficiency and cohesion, as well as morphological coverage, we find that a Unigram LM tokenizer trained on pre-tokenized English text is better off marking the word-initial token, while one trained on raw text benefits from marking word ends. Our findings generalize across domains.

UniMorph 4.0: Universal Morphology

The Universal Morphology (UniMorph) project is a collaborative effort providing broad-coverage instantiated normalized morphological inflection tables for hundreds of diverse world languages. The project comprises two major thrusts: a language-independent feature schema for rich morphological annotation and a type-level resource of annotated data in diverse languages realizing that schema. This paper presents the expansions and improvements made on several fronts over the last couple of years (since McCarthy et al. (2020)). Collaborative efforts by numerous linguists have added 67 new languages, including 30 endangered languages. We have implemented several improvements to the extraction pipeline to tackle some issues, e.g. missing gender and macron information. We have also amended the schema to use a hierarchical structure that is needed for morphological phenomena like multiple-argument agreement and case stacking, while adding some missing morphological features to make the schema more inclusive. In light of the last UniMorph release, we also augmented the database with morpheme segmentation for 16 languages. Lastly, this new release makes a push towards inclusion of derivational morphology in UniMorph by enriching the data and annotation schema with instances representing derivational processes from MorphyNet.

Learning to Parallelize in a Shared-Memory Environment with Transformers

In past years, the world has switched to many-core and multi-core shared memory architectures. As a result, there is a growing need to utilize these architectures by introducing shared memory parallelization schemes to software applications. OpenMP is the most comprehensive API that implements such schemes, characterized by a readable interface. Nevertheless, introducing OpenMP into code is challenging due to pervasive pitfalls in management of parallel shared memory. To facilitate the performance of this task, many source-to-source (S2S) compilers have been created over the years, tasked with inserting OpenMP directives into code automatically. In addition to having limited robustness to their input format, these compilers still do not achieve satisfactory coverage and precision in locating parallelizable code and generating appropriate directives. In this work, we propose leveraging recent advances in ML techniques, specifically in natural language processing (NLP), to replace S2S compilers altogether. We create a database (corpus), Open-OMP, specifically for this goal. Open-OMP contains over 28,000 code snippets, half of which contain OpenMP directives while the other half do not need parallelization at all with high probability. We use the corpus to train systems to automatically classify code segments in need of parallelization, as well as suggest individual OpenMP clauses. We train several transformer models, named PragFormer, for these tasks, and show that they outperform statistically-trained baselines and automatic S2S parallelization compilers in both classifying the overall need for an OpenMP directive and the introduction of private and reduction clauses. Our source code and database are available at: https://github.com/Scientific-Computing-Lab-NRCN/PragFormer.