Transformers excel on tasks that process well-formed inputs according to some grammar, such as natural language and code. However, it remains unclear how they can process grammatical syntax. In fact, under standard complexity conjectures, standard transformers cannot recognize context-free languages (CFLs), a canonical formalism to describe syntax, or even regular languages, a subclass of CFLs (Merrill et al., 2022). Merrill Sabharwal (2024) show that 𝒪(log n) looping layers (w.r.t. input length n) allows transformers to recognize regular languages, but the question of context-free recognition remained open. In this work, we show that looped transformers with 𝒪(log n) looping layers and 𝒪(n^6) padding tokens can recognize all CFLs. However, training and inference with 𝒪(n^6) padding tokens is potentially impractical. Fortunately, we show that, for natural subclasses such as unambiguous CFLs, the recognition problem on transformers becomes more tractable, requiring 𝒪(n^3) padding. We empirically validate our results and show that looping helps on a language that provably requires logarithmic depth. Overall, our results shed light on the intricacy of CFL recognition by transformers: While general recognition may require an intractable amount of padding, natural constraints such as unambiguity yield efficient recognition algorithms.

Scaling LLMs requires tremendous computational resources, and recent advances in AI have gone hand in hand with massive amounts of capital expenditure. While it is established that scaling up LLMs reliably increases model quality (quantified in terms of loss or downstream evaluations), it is unclear how these quality improvements translate to potential revenue, and whether revenue increases would offset costs of larger-scale training and inference. In this work, we develop an economic model for characterizing the rational behavior of an LLM training firm by combining scaling laws with microeconomic theory. Under our model of firm behavior, LLM quality can be increased with more parameters and training tokens, leading to more potential adoption by consumers, who each have a quality threshold for using the LLM. On the other hand, additional parameters and training tokens both incur additional costs. We analyze the profit maximization problem for this model under compute-bound and data-bound regimes. In the compute-bound regime, optimal model size and token budget track hardware efficiency E (FLOPs/$) at a near-linear rate; total training cost then scales sub-quadratically in E. Data efficiency improvements incentivize larger models and training expenditure. When we are limited to D data, profit-optimal training expenditure scales as D^2/E, i.e, increase with data and decreases with hardware efficiency (as well as data efficiency). Finally, we analyze practical trends in training expenditure: current trends are consistent with our most permissive model variants in the compute-bound regime, but are not profit-optimal in the data-bound regime or assuming hardware advances will stall. Overall, our results provide a theory of profit-optimal LLM training, providing a foundation for engaging critically with industry statements and supporting long-term economic decision making.

Public discourse about artificial intelligence (AI) often uses anthropomorphic language: language that attributes human capabilities and characteristics to the system. This practice has been criticized for setting misleading expectations, inflating claims, and fueling hype around AI, which may distort public understanding of AI and impact policy priorities. We study the effects of anthropomorphic framing by comparing changes in participants' perceptions (N=815) when reading passages with and without anthropomorphic language, designed to reflect realistic public-facing AI discourse. We further examine whether these effects differ across two types of AI technologies – large language models and recommendation systems – and measure changes in perceptions of AI across several dimensions that are prominent in current public discourse. In a separate condition using a text that explicitly discusses the dangers of AI, we show that individuals' views of AI can shift in response to reading a text; yet in the main conditions of the experiment, where we compare anthropomorphic and non-anthropomorphic descriptions, we find that whether the text uses anthropomorphic language does not substantially affect participants' perceptions of AI. Our results indicate that any immediate effects on public opinions of AI are modest, although they leave open the possibility that anthropomorphic language could have an effect in naturalistic settings, or over gradual, continued exposure.

When comparing the linguistic capabilities of language models (LMs) with humans using LM probabilities, factors such as the length of the sequence and the unigram frequency of lexical items have a significant effect on LM probabilities in ways that humans are largely robust to. Prior works in comparing LM and human acceptability judgments treat these effects uniformly across models, making a strong assumption that models require the same degree of adjustment to control for length and unigram frequency effects. We propose MORCELA, a new linking theory between LM scores and acceptability judgments where the optimal level of adjustment for these effects is estimated from data via learned parameters for length and unigram frequency. We first show that MORCELA outperforms a commonly used linking theory for acceptability-SLOR (Pauls and Klein, 2012; Lau et al., 2017)-across two families of transformer LMs (Pythia and OPT). Furthermore, we demonstrate that the assumed degrees of adjustment in SLOR for length and unigram frequency overcorrect for these confounds, and that larger models require a lower relative degree of adjustment for unigram frequency, though a significant amount of adjustment is still necessary for all models. Finally, our subsequent analysis shows that larger LMs' lower susceptibility to frequency effects can be explained by an ability to better predict rarer words in context.(1)

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

In languages without orthographic word boundaries, NLP models perform word segmentation, either as an explicit preprocessing step or as an implicit step in an end-to-end computation. This paper shows that Chinese NLP models are vulnerable to morphological garden path errors: errors caused by a failure to resolve local word segmentation ambiguities using sentence-level morphosyntactic context. We propose a benchmark, ERAS, that tests a model's vulnerability to morphological garden path errors by comparing its behavior on sentences with and without local segmentation ambiguities. Using ERAS, we show that word segmentation models make garden path errors on locally ambiguous sentences, but do not make equivalent errors on unambiguous sentences. We further show that sentiment analysis models with character-level tokenization make implicit garden path errors, even without an explicit word segmentation step in the pipeline. Our results indicate that models' segmentation of Chinese text often fails to account for morphosyntactic context.

Chesi's (this issue) target paper depicts a generative linguistics in crisis, foreboded by Piantadosi's (2023) declaration that "modern language models refute Chomsky's approach to language." In order to survive, Chesi warns, generativists must hold themselves to higher standards of formal and empirical rigor. This response argues that the crisis described by Chesi and Piantadosi actually has little to do with rigor, but is rather a reflection of generativists' limited social ambitions. Chesi ties the fate of generative linguistics to its intellectual merits, but the current success of language model research is social in nature as much as it is intellectual. In order to thrive, then, generativists must do more than heed Chesi's call for rigor; they must also expand their ambitions by giving outsiders a stake in their future success.

Inspired by the concept of the male gaze (Mulvey, 1975) in literature and media studies, this paper proposes a framework for analyzing gender bias in terms of female objectification: the extent to which a text portrays female individuals as objects of visual pleasure. Our framework measures female objectification along two axes. First, we compute an agency bias score that indicates whether male entities are more likely to appear in the text as grammatical agents than female entities. Next, by analyzing the word embedding space induced by a text (Caliskan et al., 2017), we compute an appearance bias score that indicates whether female entities are more closely associated with appearance-related words than male entities. Applying our framework to 19th and 20th century novels reveals evidence of female objectification in literature: we find that novels written from a male perspective systematically objectify female characters, while novels written from a female perspective do not exhibit statistically significant objectification of any gender.

This paper shows that the universal generation problem (Heinz, Kobele, and Riggle 2009) for Optimality Theory (OT, Prince and Smolensky 1993, 2004) is PSPACE-complete. While prior work has shown that universal generation is at least NP-hard (Eisner 1997, 2000b; Wareham 1998; Idsardi 2006) and at most EXPSPACE-hard (Riggle 2004), our results place universal generation in between those two classes, assuming that NP ≠ PSPACE. We additionally show that when the number of constraints is bounded in advance, universal generation is at least NL-hard and at most NPNP-hard. Our proofs rely on a close connection between OT and the intersection non-emptiness problem for finite automata, which is PSPACE-complete in general (Kozen 1977) and NL-complete when the number of automata is bounded (Jones 1975). Our analysis shows that constraint interaction is the main contributor to the complexity of OT: the ability to factor transformations into simple, interacting constraints allows OT to furnish compact descriptions of intricate phonological phenomena.

Deep architectures such as Transformers are sometimes criticized for having uninterpretable "black-box" representations. We use causal intervention analysis to show that, in fact, some linguistic features are represented in a linear, interpretable format. Specifically, we show that BERT's ability to conjugate verbs relies on a linear encoding of subject number that can be manipulated with predictable effects on conjugation accuracy. This encoding is found in the subject position at the first layer and the verb position at the last layer, but distributed across positions at middle layers, particularly when there are multiple cues to subject number.

With the proliferation of online misinformation, fake news detection has gained importance in the artificial intelligence community. In this paper, we propose an adversarial benchmark that tests the ability of fake news detectors to reason about real-world facts. We formulate adversarial attacks that target three aspects of "understanding": compositional semantics, lexical relations, and sensitivity to modifiers. We test our benchmark using BERT classifiers fine-tuned on the LIAR arXiv:arch-ive/1705648 and Kaggle Fake-News datasets, and show that both models fail to respond to changes in compositional and lexical meaning. Our results strengthen the need for such models to be used in conjunction with other fact checking methods.

This paper analyzes three formal models of Transformer encoders that differ in the form of their self-attention mechanism: unique hard attention (UHAT); generalized unique hard attention (GUHAT), which generalizes UHAT; and averaging hard attention (AHAT). We show that UHAT and GUHAT Transformers, viewed as string acceptors, can only recognize formal languages in the complexity class AC(0), the class of languages recognizable by families of Boolean circuits of constant depth and polynomial size. This upper bound subsumes Hahn's (2020) results that GUHAT cannot recognize the DYCK languages or the PARITY language, since those languages are outside AC(0) (Furst et al., 1984). In contrast, the non-AC(0) languages MAJORITY and DYCK-1 are recognizable by AHAT networks, implying that AHAT can recognize languages that UHAT and GUHAT cannot.

Search & Copy (S&C) is a procedural model of vowel harmony in which underspecified vowels trigger searches for targets that provide them with features. In this paper, we seek to relate the S&C formalism with models of phonological locality proposed by recent work in the subregular program. Our goal is to provide a formal description, within the framework of mathematical linguistics, of the range of possible phonological transformations that admit an analysis within S&C. We show that used in its unidirectional mode, all transformations described by an S&C analysis can be modeled by tier-based input strictly local functions (TISL). This result improves the previous result of Gainor et al 2012, which showed that vowel harmony processes can be modeled by subsequential functions. However, non-TISL transformations can be given S&C descriptions in the following ways. Firstly, since TISL functions are not closed under composition, a non-TISL vowel harmony pattern may be obtained by applying two S&C rules sequentially. Secondly, when S&C is used in its bidirectional mode, it has the ability to describe transformations that cannot be modeled by finite-state functions.

LSTM language models have been shown to capture syntax-sensitive grammatical dependencies such as subject-verb agreement with a high degree of accuracy (Linzen et al., 2016, inter alia). However, questions remain regarding whether they do so using spurious correlations, or whether they are truly able to match verbs with their subjects. This paper argues for the latter hypothesis. Using layer-wise relevance propagation (Bach et al., 2015), a technique that quantifies the contributions of input features to model behavior, we show that LSTM performance on number agreement is directly correlated with the model's ability to distinguish subjects from other nouns. Our results suggest that LSTM language models are able to infer robust representations of syntactic dependencies.

Interpretability methods for neural networks are difficult to evaluate because we do not understand the black-box models typically used to test them. This paper proposes a framework in which interpretability methods are evaluated using manually constructed networks, which we call white-box networks, whose behavior is understood a priori. We evaluate five methods for producing attribution heatmaps by applying them to white-box LSTM classifiers for tasks based on formal languages. Although our white-box classifiers solve their tasks perfectly and transparently, we find that all five attribution methods fail to produce the expected model explanations.

By positing a relationship between naturalistic reading times and information-theoretic surprisal, surprisal theory (Hale, 2001; Levy, 2008) provides a natural interface between language models and psycholinguistic models. This paper re-evaluates a claim due to Goodkind and Bicknell (2018) that a language model's ability to model reading times is a linear function of its perplexity. By extending Goodkind and Bicknell's analysis to modern neural architectures, we show that the proposed relation does not always hold for Long Short-Term Memory networks, Transformers, and pre-trained models. We introduce an alternate measure of language modeling performance called predictability norm correlation based on Cloze probabilities measured from human subjects. Our new metric yields a more robust relationship between language model quality and psycholinguistic modeling performance that allows for comparison between models with different training configurations.

This paper analyzes the language-theoretic complexity of Harmonic Serialism (HS), a derivational variant of Optimality Theory. I show that HS can generate non-rational relations using strictly local markedness constraints, proving the “result” of Hao (2017), that HS is rational under those assumptions, to be incorrect. This is possible because deletions performed in a particular order have the ability to enforce nesting dependencies over long distances. I argue that coordinated deletions form a canonical characterization of non-rational relations definable in HS.

This paper situates culminative unbounded stress systems within the subregular hierarchy for functions. While Baek (2018) has argued that such systems can be uniformly understood as input tier-based strictly local constraints, we show here that default-to-oppositeside and default-to-same-side stress systems belong to distinct subregular classes when they are viewed as functions that assign primary stress to underlying forms. While the former system can be captured by input tier-based input strictly local functions, a subsequential function class that we define here, the latter system is not subsequential, though it is weakly deterministic according to McCollum et al.'s (2018) non-interaction criterion. Our results motivate the extension of recently proposed subregular language classes to subregular functions and argue in favor of McCollum et al.'s definition of weak determinism over that of Heinz and Lai (2013).