In law, lore, and everyday life, loopholes are commonplace. When people exploit a loophole, they understand the intended meaning or goal of another person, but choose to go with a different interpretation. Past and current AI research has shown that artificial intelligence engages in what seems superficially like the exploitation of loopholes, but this is likely anthropomorphization. It remains unclear to what extent current models, especially Large Language Models (LLMs), capture the pragmatic understanding required for engaging in loopholes. We examined the performance of LLMs on two metrics developed for studying loophole behavior in humans: evaluation (ratings of trouble, upset, and humor), and generation (coming up with new loopholes in a given context). We conducted a fine-grained comparison of state-of-the-art LLMs to humans, and find that while many of the models rate loophole behaviors as resulting in less trouble and upset than outright non-compliance (in line with adults), they struggle to recognize the humor in the creative exploitation of loopholes in the way that humans do. Furthermore, only two of the models, GPT 3 and 3.5, are capable of generating loopholes of their own, with GPT3.5 performing closest to the human baseline.

The dominant paradigm for high-performance models in novel NLP tasks today is direct specialization for the task via training from scratch or fine-tuning large pre-trained models. But does direct specialization capture how humans approach novel language tasks? We hypothesize that human performance is better characterized by flexible inference through composition of basic computational motifs available to the human language user. To test this hypothesis, we formulate a set of novel fragmentary text completion tasks, and compare the behavior of three direct-specialization models against a new model we introduce, GibbsComplete, which composes two basic computational motifs central to contemporary models: masked and autoregressive word prediction. We conduct three types of evaluation: human judgments of completion quality, satisfaction of syntactic constraints imposed by the input fragment, and similarity to human behavior in the structural statistics of the completions. With no task-specific parameter tuning, GibbsComplete performs comparably to direct-specialization models in the first two evaluations, and outperforms all direct-specialization models in the third evaluation. These results support our hypothesis that human behavior in novel language tasks and environments may be better characterized by flexible composition of basic computational motifs rather than by direct specialization.

Recent causal probing literature reveals when language models and syntactic probes use similar representations. Such techniques may yield “false negative” causality results: models may use representations of syntax, but probes may have learned to use redundant encodings of the same syntactic information. We demonstrate that models do encode syntactic information redundantly and introduce a new probe design that guides probes to consider all syntactic information present in embeddings. Using these probes, we find evidence for the use of syntax in models where prior methods did not, allowing us to boost model performance by injecting syntactic information into representations.

Prior work has shown that structural supervision helps English language models learn generalizations about syntactic phenomena such as subject-verb agreement. However, it remains unclear if such an inductive bias would also improve language models’ ability to learn grammatical dependencies in typologically different languages. Here we investigate this question in Mandarin Chinese, which has a logographic, largely syllable-based writing system; different word order; and sparser morphology than English. We train LSTMs, Recurrent Neural Network Grammars, Transformer language models, and Transformer-parameterized generative parsing models on two Mandarin Chinese datasets of different sizes. We evaluate the models’ ability to learn different aspects of Mandarin grammar that assess syntactic and semantic relationships. We find suggestive evidence that structural supervision helps with representing syntactic state across intervening content and improves performance in low-data settings, suggesting that the benefits of hierarchical inductive biases in acquiring dependency relationships may extend beyond English.

Transformer-based language models pre-trained on large amounts of text data have proven remarkably successful in learning generic transferable linguistic representations. Here we study whether structural guidance leads to more human-like systematic linguistic generalization in Transformer language models without resorting to pre-training on very large amounts of data. We explore two general ideas. The “Generative Parsing” idea jointly models the incremental parse and word sequence as part of the same sequence modeling task. The “Structural Scaffold” idea guides the language model’s representation via additional structure loss that separately predicts the incremental constituency parse. We train the proposed models along with a vanilla Transformer language model baseline on a 14 million-token and a 46 million-token subset of the BLLIP dataset, and evaluate models’ syntactic generalization performances on SG Test Suites and sized BLiMP. Experiment results across two benchmarks suggest converging evidence that generative structural supervisions can induce more robust and humanlike linguistic generalization in Transformer language models without the need for data intensive pre-training.

Humans can learn structural properties about a word from minimal experience, and deploy their learned syntactic representations uniformly in different grammatical contexts. We assess the ability of modern neural language models to reproduce this behavior in English and evaluate the effect of structural supervision on learning outcomes. First, we assess few-shot learning capabilities by developing controlled experiments that probe models’ syntactic nominal number and verbal argument structure generalizations for tokens seen as few as two times during training. Second, we assess invariance properties of learned representation: the ability of a model to transfer syntactic generalizations from a base context (e.g., a simple declarative active-voice sentence) to a transformed context (e.g., an interrogative sentence). We test four models trained on the same dataset: an n-gram baseline, an LSTM, and two LSTM-variants trained with explicit structural supervision. We find that in most cases, the neural models are able to induce the proper syntactic generalizations after minimal exposure, often from just two examples during training, and that the two structurally supervised models generalize more accurately than the LSTM model. All neural models are able to leverage information learned in base contexts to drive expectations in transformed contexts, indicating that they have learned some invariance properties of syntax.

While state-of-the-art neural network models continue to achieve lower perplexity scores on language modeling benchmarks, it remains unknown whether optimizing for broad-coverage predictive performance leads to human-like syntactic knowledge. Furthermore, existing work has not provided a clear picture about the model properties required to produce proper syntactic generalizations. We present a systematic evaluation of the syntactic knowledge of neural language models, testing 20 combinations of model types and data sizes on a set of 34 English-language syntactic test suites. We find substantial differences in syntactic generalization performance by model architecture, with sequential models underperforming other architectures. Factorially manipulating model architecture and training dataset size (1M-40M words), we find that variability in syntactic generalization performance is substantially greater by architecture than by dataset size for the corpora tested in our experiments. Our results also reveal a dissociation between perplexity and syntactic generalization performance.

Targeted syntactic evaluations have yielded insights into the generalizations learned by neural network language models. However, this line of research requires an uncommon confluence of skills: both the theoretical knowledge needed to design controlled psycholinguistic experiments, and the technical proficiency needed to train and deploy large-scale language models. We present SyntaxGym, an online platform designed to make targeted evaluations accessible to both experts in NLP and linguistics, reproducible across computing environments, and standardized following the norms of psycholinguistic experimental design. This paper releases two tools of independent value for the computational linguistics community: 1. A website, syntaxgym.org, which centralizes the process of targeted syntactic evaluation and provides easy tools for analysis and visualization; 2. Two command-line tools, ‘syntaxgym‘ and ‘lm-zoo‘, which allow any user to reproduce targeted syntactic evaluations and general language model inference on their own machine.

We investigate the extent to which the behavior of neural network language models reflects incremental representations of syntactic state. To do so, we employ experimental methodologies which were originally developed in the field of psycholinguistics to study syntactic representation in the human mind. We examine neural network model behavior on sets of artificial sentences containing a variety of syntactically complex structures. These sentences not only test whether the networks have a representation of syntactic state, they also reveal the specific lexical cues that networks use to update these states. We test four models: two publicly available LSTM sequence models of English (Jozefowicz et al., 2016; Gulordava et al., 2018) trained on large datasets; an RNN Grammar (Dyer et al., 2016) trained on a small, parsed dataset; and an LSTM trained on the same small corpus as the RNNG. We find evidence for basic syntactic state representations in all models, but only the models trained on large datasets are sensitive to subtle lexical cues signaling changes in syntactic state.

State-of-the-art LSTM language models trained on large corpora learn sequential contingencies in impressive detail, and have been shown to acquire a number of non-local grammatical dependencies with some success. Here we investigate whether supervision with hierarchical structure enhances learning of a range of grammatical dependencies, a question that has previously been addressed only for subject-verb agreement. Using controlled experimental methods from psycholinguistics, we compare the performance of word-based LSTM models versus Recurrent Neural Network Grammars (RNNGs) (Dyer et al. 2016) which represent hierarchical syntactic structure and use neural control to deploy it in left-to-right processing, on two classes of non-local grammatical dependencies in English—Negative Polarity licensing and Filler-Gap Dependencies—tested in a range of configurations. Using the same training data for both models, we find that the RNNG outperforms the LSTM on both types of grammatical dependencies and even learns many of the Island Constraints on the filler-gap dependency. Structural supervision thus provides data efficiency advantages over purely string-based training of neural language models in acquiring human-like generalizations about non-local grammatical dependencies.

Neural language models have achieved state-of-the-art performances on many NLP tasks, and recently have been shown to learn a number of hierarchically-sensitive syntactic dependencies between individual words. However, equally important for language processing is the ability to combine words into phrasal constituents, and use constituent-level features to drive downstream expectations. Here we investigate neural models’ ability to represent constituent-level features, using coordinated noun phrases as a case study. We assess whether different neural language models trained on English and French represent phrase-level number and gender features, and use those features to drive downstream expectations. Our results suggest that models use a linear combination of NP constituent number to drive CoordNP/verb number agreement. This behavior is highly regular and even sensitive to local syntactic context, however it differs crucially from observed human behavior. Models have less success with gender agreement. Models trained on large corpora perform best, and there is no obvious advantage for models trained using explicit syntactic supervision.