Finding your voice: Voice-specific effects in Tagalog reveal the limits of word order priming
Introduction
A core aim of psycholinguistics is to uncover the architecture of the language faculty in a manner that links performance to linguistic representation. One key method for studying the deployment of linguistic representations during language production is structural priming (Bock, 1986; Branigan & Pickering, 2017), which takes advantage of the fact that speakers tend to re-use constructions they have previously heard. For example, speakers are more likely to use a passive sentence such as the cat is being chased by the dog after hearing another passive sentence like the suspect is being followed by the policeman than after an equivalent active sentence (i.e., the policeman is following the suspect). Since the two passive sentences do not contain open class lexical overlap, a priming effect is typically taken to indicate the presence of an abstract representation of the English passive structure.
While the utility of priming for identifying the representational properties of the system is not in question, the nature and scope of the representations within psycholinguistic theory remain unclear. A key source of evidence in this endeavour is data from typologically-diverse languages, the study of which tests the limits of theoretical approaches built upon a fairly small and non-representative sample of languages. Notably, psycholinguistics has drawn heavily from English and a handful of other well-studied languages (Anand, Chung, & Wagers, 2011; Kidd & Garcia, 2022; Jaeger & Norcliffe, 2009), which are not representative of the linguistic diversity present in natural language (Evans & Levinson, 2009). In this study, we investigated priming in Tagalog, an understudied language spoken primarily in the Philippines. Of particular interest is Tagalog's typologically rare symmetrical voice system (Foley, 2008; Riesberg, Malcher, & Himmelmann, 2019), which is characterised by the presence of more than one basic transitive structure that are equally marked with verbal and nominal morphology without argument demotion (i.e., in the English passive, the agent is demoted to an oblique argument). Across three priming studies, we show that the pattern of priming across voices is different and, crucially, voice-dependent in ways that challenge existing theories of priming. Additionally, the pattern of results also provides empirical evidence bearing upon linguistic theory concerning language change in Austronesian languages, underlining the utility of priming as a method for studying the dynamics of language at the individual and, by implication, the population level.
Tagalog is a Western Austronesian language with around 28 million speakers across the globe (Eberhard, Simons, & Fennig, 2022). In Tagalog, the voice-marking on the verb assigns the thematic role of the subject—the argument marked by ang1 (see 1–4; Himmelmann, 2005). In the agent voice (AV), the verb infix -um- marks the subject as the agent (1, 2). In the patient voice (PV), the infix -in- assigns the subject the patient role (3, 4). Aside from the patient voice, there are also other undergoer voices where the ang-phrase is a recipient, instrument or beneficiary. Non-subject arguments or adjuncts are marked by ng or sa. Tagalog is canonically verb-initial, with a relatively free post-verbal argument order (Schachter, 2015), resulting in both agent-initial (1, 3) and patient-initial (2, 4) orders in the different voices.
In Tagalog's voice system, the change in verb morphology also changes the mapping between syntactic functions and thematic roles. For example, an agent voice patient-initial sentence (2) can potentially prime another patient-initial sentence or an ang-last sentence. In the agent voice, both these orders point to the same structure. However, given a patient voice target, the speaker would have to switch the order of syntactic functions in the prime in order to produce a patient-initial sentence (4), or if the speaker follows the order of syntactic functions of the agent voice prime, a patient voice agent-initial sentence (3) will be produced instead.
There is evidence suggesting that the order of post-verbal arguments in Tagalog depends, to some extent, on the verb's morphological voice: corpus and experimental work show that the agent voice allows a great deal of flexibility, whereas in the patient voice there is a clear (though not absolute) preference for the agent-first sentence (Garcia, Dery, Roeser, & Höhle, 2018; Garcia, Garrido Rodriguez, & Kidd, 2021; Garcia & Kidd, 2020). These usage patterns are driven by two preferences. The first is a general ang-last preference, where the prominent syntactic argument appears clause final (Kroeger, 1993b). The second is a preference to place agents first, following the verb (see Riesberg et al., 2019). In the agent voice, where the ang-phrase marks the agent (1, 2), there is tension between the two preferences, which results in more flexible word order in this voice. In contrast, in the patient voice, the two preferences converge, such that sentences like (3) are more common than (4).
Riesberg et al. (2019) suggested that this tension between ang-last and agent-first ordering may drive language change towards agent-first word order preferences in symmetrical voice languages, with the universal agent-first preference eventually winning out diachronically (i.e., over historical time). This leads to an intriguing hypothesis at the intersections of psycholinguistics, typology, and language change: if a language like Tagalog allows more variation of argument order in the agent voice than in the patient voice, then we should see stronger priming in the agent voice than in the patience voice, due to the countervailing influences of the ang-last and agent-first preference in the agent voice. Note that this is quite different from the inverse frequency effect generally observed in the syntactic priming literature (e.g., see Pickering & Ferreira, 2008): low frequency alternations such as the passive in active-passive alternation prime more than subtler differences such as priming of the double object dative construction in the dative alternation. Instead, the suggestion is that flexibility in the ordering of core arguments differs across voice, and where that flexibility is licensed by competing grammatical preferences, priming may be higher in magnitude and act as a driver of diachronic change. The prediction is consistent with Jäger and Rosenbach's (2008) suggestion that priming at the level of the individual acts as a driver of language change.
There are different theoretical explanations for structural priming phenomena. According to Pickering and Branigan (1998), priming originates from the residual activation that sentences leave in the combinatorial nodes of their structures. In their theory, the base form of the verb (lemma: Levelt, 1989) is linked to syntactic properties such as tense or number inflection, and to the structure the verb is in (e.g., prepositional dative or double object structure). For example, the English prepositional dative prime sentence The racing driver showed the torn overall to the helpful mechanic leaves residual activation in the lemma SHOW, in the combinatorial prepositional dative node, and in the link between the lemma and prepositional dative nodes. This increases the likelihood for another prepositional dative structure to be used. When the prime and target verbs have different lemmas, only the residual activation of the prepositional dative node causes priming. When the prime and target contain the same verb, activation in both the lemma and prepositional dative node result in an additive effect, causing the lexical boost. More importantly, Pickering and Branigan (1998) argue that since it is the unspecified/uninflected form of the verb (lemma) that is linked to the prepositional dative combinatorial node, and not the inflected form of the verb (for tense, aspect or number), varying the inflection of the verb does not affect the magnitude of priming.
Pickering and Branigan (1998) presented data from English that were consistent with this account: participants produced more prepositional datives The patient showed the [object] to the [beneficiary] after reading a prepositional dative prime compared to a double object prime The patient showed the [beneficiary] the [object], regardless of whether the verbs in prime and target were in the same tense (prime: showed – target: showed) or in a different tense (prime: showed – target: shows). They also observed a lexical boost effect, in which priming was larger in magnitude when prime and target shared the same verb.
These results, however, may be specific to English. Chang, Baumann, Pappert, and Fitz (2015) tested the predictions of the account in German-speaking adults. Word order variations in German are sensitive to verbal tense/aspect. Specifically, past tense verbs occur after the subject (in the verb second word order), as in (5). However, in the perfective aspect, the lexical verb moves to the end of the sentence, as in (6).
- (5)
Der Rechtsanwalt schickte den Vertrag an den Klienten
‘The lawyer sent the contract to the client’
- (6)
Der Rechtsanwalt hat den Vertrag an den Klienten geschickt
‘The lawyer has sent the contract to the client’
The residual activation account locates the lexical boost at the level of the lemma. As both sentences have the same verb lemma, it predicts the same pattern of priming (i.e., the same magnitude) across sentences containing verb-second (sentence 5) or verb-final (6) lexical verbs. However, Chang, Dell and Bock's (2006) Dual-path model makes an alternative prediction: The model learns language-specific grammatical representations via a serial recurrent network that attempts to predict the next word in a sentence and updates its knowledge when the model predictions are not met. Thus, abstract knowledge emerges from locally recurring lexical patterns via implicit learning (for other accounts of priming that explain abstract priming via learning processes see Jaeger & Snider, 2013; Reitter, Keller, & Moore, 2011). For German, the prediction is that the model acquires separate structural representations for verb-second and verb-final structures, in addition to general verb-independent structural knowledge (see Chang et al., 2015, p. 11). Thus, priming should be strongest when prime and target share the verb-second and verb-final pattern. This is what the authors found in human adult participants.
The difference in results across English and German reveals the value in crosslinguistic comparisons for theory testing, even in closely related languages. The pattern of results raises an important question concerning grammatical representation in symmetrical voice systems like that of Tagalog. Firstly, do uninflected verbs, which are ungrammatical but do appear in naturalistic speech in some contexts (Garcia & Kidd, 2022), constitute lemmas, and thus, can priming occur at this level? If this is the case, it would suggest that in symmetrical voice systems there is a unitary transitive structure in which voice marking induces valency changes.2 The implication is that the residual activation account would predict that priming would be observed across voice types. For example, if a Tagalog agent voice kick-ng-ang (patient-initial) prime sentence leaves residual activation in the verb lemma KICK, in the ng-ang node, and in the link between KICK and the ng-ang node, it would increase the likelihood for another ng-ang sentence to be used regardless of the voice of the target verb. Note that a ng-ang sentence in the agent voice is patient-initial but agent-initial in the patient voice, so activation of a ng-ang node results in different thematic role order depending on the voice-marking on the verb.
Secondly, and in contrast to the unitary account, each voice may constitute a separate transitive structure, such that there is no priming at the lemma level. The prediction is that priming would be voice-specific, a hypothesis that we interpret to be consistent with the Chang et al. (2006) model, as it learns structure by mapping event roles onto word sequences, which it refines across its development via error-based learning. Thus, it will learn the mapping between Verb-Noun Phrase-Noun Phrase (V-NP-NP) sequences, and because these mappings change depending on voice, the model will induce separate syntactic frames for different voices (effectively treating each voice-marked variant as a separate lexical entry). This is broadly consistent with symmetrical voice analyses of Philippine-type Austronesian languages3 (Himmelmann, 2002, Himmelmann, 2005; Foley, 2008; Riesberg, 2014; for a review see Chen & McDonnell, 2019), and is consistent with on-line parsing data reported by Garcia et al. (2021), who showed a voice-driven asymmetry in children's and adults' ability to predict argument roles from voice-inflected verb plus noun marker combinations.
There is currently one previous priming study on Tagalog, which focused on children but also collected an adult comparison group. In that study, Garcia and Kidd (2020) manipulated the order of agent and patient in descriptions of actions between two animals (prime sentence) as well the voice of the target verb prompts, and found no evidence for priming of word order in their sample of adult participants. However, as the experiment was designed for children, it is unclear if the results were observed because priming is impossible in Tagalog, or if it was due to the fact that any effect was obscured in adults because of the significant methodological changes required to test young children. For example, given that the method involved the description of pictures and was presented to the participants as a study that was to be conducted with children, the adult participants may have ignored the prime sentences. In the current study, we used a different paradigm that is more typical of adult-focused priming studies, and which required the participants to engage with the prime sentence.
In three structural priming experiments, we investigated how voice morphology affects word order priming. Our main aim was to use structural priming to determine the representational nature of the Tagalog transitive structure. Accordingly, we determined whether priming occurred across different voices. This question naturally bears upon different explanations of priming. If, following residual activation accounts (Pickering & Branigan, 1998), priming occurs at the lemma level, we expect priming across voice, which would be consistent with the possibility that the transitive has a unitary representational structure. However, if different voice alternations constitute separate structures, as is assumed by the symmetrical voice analyses of Austronesian voice (Chen & McDonnell, 2019; Foley, 2008; Himmelmann, 2002, Himmelmann, 2005; Riesberg, 2014), and as predicted by the Chang et al. (2006) computational model of sentence production, priming effects are only expected when prime and target verbs are marked with the same voice affixes.
Our study also informs Riesberg et al.'s (2019) proposal that, due to a preference to place agents early in a sentence (i.e., the Universal Agent Hypothesis), Tagalog agent voice sentences have more flexible word order patterns because of the countervailing influence of the ang-last preference. This predicts greater priming of patient-initial order in the agent voice than in the patient voice. Moreover, Riesberg et al. (2019) argue that the universal agent-first preference will eventually win out diachronically. Thus, the results also bear upon Jäger and Rosenbach (2008) argument that priming at the individual level plays a key role in language change via implicit learning.
These predictions were tested across three experiments where we manipulated whether the target verb prompt had the same voice as the verb in the prime sentence. In Experiment 1, we tested whether Tagalog speakers could be primed to produce patient-initial sentences given agent or patient voice targets, while the primes were consistently in the agent voice. In Experiment 2, we used patient voice-marked primes instead of agent voice-marked primes. In our pre-registered Experiment 3, we used both agent voice and patient voice primes while keeping the target sentences in the agent voice.
Section snippets
Participants
We tested a total of 85 Tagalog-speakers from Greater Manila Area (50 females, 35 males) with a median age of 28 years (SD = 4.73, range: 18–35); none reported a present or past diagnosis of a speech or language impairment or psychological / neurological illness. Participants who contributed <20 trials (n = 11) were excluded from the data analysis, as were participants who had completed the experiment previously (n = 6), failed to follow the instructions (n = 3), or scored lower than 60% (n
Experiment 2: patient voice primes
Experiment 2 was largely identical to Experiment 1, with the exception that prime verbs were inflected for the patient voice. This manipulation allowed us to test two possibilities. Firstly, it allowed us to test if priming only occurs with agent voice targets because of its greater word order flexibility in comparison to the patient voice (Garcia et al., 2018; Garcia & Kidd, 2020). On this possibility, priming is possible across voice (i.e., if patient voice sentences primed agent voice
Experiment 3: pre-registered experiment manipulating agent and patient voice primes with agent voice targets
Experiments 1 and 2 showed a clearly different pattern of results that indicate priming in Tagalog is voice-dependent. However, the design of the experiments was not typical in that we manipulated target voice and held prime voice constant.7 In our pre-registered Experiment 3, we tested the opposite relation, holding target voice constant (agent voice) and manipulating prime voice. We chose to use agent over patient voice targets because of
General discussion
In the current paper, we have presented three structural priming studies focusing on Tagalog's symmetrical voice system, a typologically unique feature of Philippine-type Austronesian languages whereby the transitive alternation is syntactically ‘balanced’ (i.e., symmetrical) in the agent and patient voice. Studying Tagalog enabled us to test the effect of voice morphology on priming. Word order priming was conditional on the voice of the target sentence, and the priming strength differed
Conclusion
In the current paper, we present data that demonstrate syntactic priming in Tagalog is voice-specific in two key ways: first, there were no word order priming effects if the verbs of the prime and the target sentence were marked by different voice inflections; second, when the verb of the prime and the target sentence were inflected with the same voice morpheme, the magnitude of word order priming effect was conditional on the voice morpheme, with priming effects being larger for the voice type
CRediT authorship contribution statement
Rowena Garcia: Conceptualization, Methodology, Software, Investigation, Formal analysis, Writing – original draft. Jens Roeser: Conceptualization, Methodology, Data curation, Formal analysis, Visualization, Writing – original draft. Evan Kidd: Conceptualization, Methodology, Writing – original draft.
Acknowledgements
We would like to thank the audience at AMLaP 2021 and X-PPL 2022, our colleagues at MPI's Language Development department, and the four anonymous reviewers for their helpful comments. We are also grateful to the support team of Gorilla, and most of all, to our participants. This work was supported by the Max Planck Society, the Australian Research Council [CE140100041: CI Kidd], and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation, 317633480 – SFB 1287).
References (87)
Antipassive and ergativity in Tagalog
Lingua
(2012)- et al.
Random effects structure for confirmatory hypothesis testing: Keep it maximal
Journal of Memory and Language
(2013) - et al.
The persistence of syntactic priming revisited
Journal of Memory and Language
(2016) Syntactic persistence in language production
Cognitive Psychology
(1986)- et al.
Contrast coding choices in a decade of mixed models
Journal of Memory and Language
(2022) - et al.
Mapping concepts to syntax: Evidence from structural priming in Mandarin Chinese
Journal of Memory and Language
(2012) Learning to order words: A connectionist model of heavy NP shift and accessibility effects in Japanese and English
Journal of Memory and Language
(2009)- et al.
Can thematic roles leave traces of their places?
Cognition
(2003) - et al.
Developmental effects in the online use of morphosyntactic cues in sentence processing: Evidence from Tagalog
Cognition
(2021) - et al.
Syntactic priming persists while the lexical boost decays: Evidence from written and spoken dialogue
Journal of Memory and Language
(2008)
How do people produce ungrammatical utterances?
Journal of Memory and Language
Alignment as a consequence of expectation adaptation: Syntactic priming is affected by the prime’s prediction error given both prior and recent experience
Cognition
The head or the verb: Is the lexical boost restricted to the head verb?
Journal of Memory and Language
Implicit learning of structure across time: A longitudinal investigation of syntactic priming in young English-acquiring children
Journal of Memory and Language
A meta-analysis of syntactic priming in language production
Journal of Memory and Language
The representation of verbs: Evidence from syntactic priming in language production
Journal of Memory and Language
The development of abstract syntax: Evidence from structural priming and the lexical boost
Cognition
How to capitalize on a priori contrasts in linear (mixed) models: A tutorial
Journal of Memory and Language
The lexical boost effect is not diagnostic of lexically-specific syntactic representations
Journal of Memory and Language
Bayesian hypothesis testing for psychologists: A tutorial on the savage–dickey method
Cognitive Psychology
How abstract is syntax? Evidence from structural priming
Cognition
Internally and externally headed relative clauses in Tagalog
Glossa: A Journal of General Linguistics
Widening the net: Challenges for gathering linguistic data in the digital age
Gorilla in our midst: An online behavioral experiment builder
Behavior Research Methods
Serious stats: A guide to advanced statistics for the behavioral sciences
Parsimonious mixed models
The neurophysiology of language processing shapes the evolution of grammar: Evidence from case marking
PLoS One
The persistence of structural priming: Transient activation or implicit learning?
Journal of Experimental Psychology: General
Framing sentences
Cognition
The extended argument dependency model: A neurocognitive approach to sentence comprehension across languages
Psychological Review
An experimental approach to linguistic representation
Behavioral and Brain Sciences
Brms: An R package for Bayesian multilevel models using Stan
Journal of Statistical Software
Advanced Bayesian multilevel modeling with the R package brms
The R Journal
Stan: A probabilistic programming language
Journal of Statistical Software
Linking of thematic roles in derivational word formation
Linguistic Inquiry
Do lemmas speak German? A verb position effect in German structural priming
Cognitive Science
Becoming syntactic
Psychological Review
Structural priming as implicit learning: A comparison of models of sentence production
Journal of Psycholinguistic Research
Western Austronesian voice
Annual Review of Linguistics
Hierarchy in language interpretation: Evidence from behavioural experiments and computational modelling
Language, Cognition and Neuroscience
Ergative analysis for Philippine languages: An analysis
The P-chain: Relating sentence production and its disorders to comprehension and acquisition
Philosophical Transactions of the Royal Society, B: Biological Sciences
The weighted likelihood ratio, sharp hypotheses about chances, the order of a markov chain
The Annals of Mathematical Statistics
Cited by (1)
Philippine Psycholinguistics
2024, Annual Review of Linguistics