Language Modeling Teaches You More Syntax than Translation Does: Lessons Learned Through Auxiliary Task Analysis

Abstract

Recent work using auxiliary prediction task classifiers to investigate the properties of LSTM representations has begun to shed light on why pretrained representations, like ELMo (Peters et al., 2018) and CoVe (McCann et al., 2017), are so beneficial for neural language understanding models. We still, though, do not yet have a clear understanding of how the choice of pretraining objective affects the type of linguistic information that models learn. With this in mind, we compare four objectives---language modeling, translation, skip-thought, and autoencoding---on their ability to induce syntactic and part-of-speech information. We make a fair comparison between the tasks by holding constant the quantity and genre of the training data, as well as the LSTM architecture. We find that representations from language models consistently perform best on our syntactic auxiliary prediction tasks, even when trained on relatively small amounts of data. These results suggest that language modeling may be the best data-rich pretraining task for transfer learning applications requiring syntactic information. We also find that the representations from randomly-initialized, frozen LSTMs perform strikingly well on our syntactic auxiliary tasks, but this effect disappears when the amount of training data for the auxiliary tasks is reduced.

References in corpus (2)

• Deep Learning Scaling is Predictable, Empirically
• What you can cram into a single vector: Probing sentence embeddings for linguistic properties

Cited by in corpus (15)

• Sentence Encoders on STILTs: Supplementary Training on Intermediate Labeled-data Tasks
• Do Attention Heads in BERT Track Syntactic Dependencies?
• Correlating neural and symbolic representations of language
• Analyzing analytical methods: The case of phonology in neural models of spoken language
• Designing and Interpreting Probes with Control Tasks
• Emergence of Separable Manifolds in Deep Language Representations
• Can You Tell Me How to Get Past Sesame Street? Sentence-Level Pretraining Beyond Language Modeling
• Investigating Multilingual NMT Representations at Scale
• Dissecting Contextual Word Embeddings: Architecture and Representation
• Detecting and Exorcising Statistical Demons from Language Models with Anti-Models of Negative Data
• Raw-to-End Name Entity Recognition in Social Media
• Syntax Representation in Word Embeddings and Neural Networks -- A Survey
• Conditional probing: measuring usable information beyond a baseline
• Analyzing and Interpreting Neural Networks for NLP: A Report on the First BlackboxNLP Workshop
• Evaluating the Representational Hub of Language and Vision Models