Transitivity, Space, and Hand: The Spatial Grounding of Syntax

1 Introduction

A central issue in cognitive psychology is how the mind processes language. Work by cognitive grammarians suggests the possibility that syntax, a crucial part of language, may involve an inherent spatial representation or prototypical temporal sequencing (Croft, 1991; Langacker, 2001; Talmy, 2000). While this research may seem non-central to psycholinguistic interests, connections between space and language have been made by psychologists, mainly in the context of metaphor (Boroditsky, 2001; Lakoff & Johnson, 1980), semantics (Zwaan & Yaxley, 2003), or explicitly spatial language (Levinson, 1996). One area in cognitive psychology where the connection between space and syntax has been addressed more directly is in studies of visual scene production (e.g., drawing) and recognition tasks (Chatterjee, 2001; Chatterjee, Southwood, & Basilico, 1999; Maas & Russo, 2003), which have found evidence for a spatial bias in verb representations such that participants recognize scenes with action motion trajectories moving left-to-right faster than those moving right-to-left. Interestingly, this lateral asymmetry is similar to findings from research on the relation between space and number (e.g., Bueti & Walsh, 2009; Dehaene, Bossini, & Giraux, 1993; Hubbard, Piazza, Pinel, & Dehaene, 2005; Ishihara, Keller, Rossetti, & Prinz, 2008), the general conclusion from this research being that smaller quantities (e.g., low numbers, small grip aperture) prime the left side of space, and larger quantities (e.g., high numbers, large grip aperture) prime the right side. In this study, we build on these various lines of research to further explore the relationship between syntax and space.

The relation between language and space is closely related to questions of embodiment, namely the role of sensorimotor representations in language processing. Indeed, related lines of research have found evidence for the embodiment of language in motor processes (Glenberg & Kaschak, 2002; Hauk, Johnsrude, & Pulvermuller, 2004; Zwaan & Taylor, 2006) and visuospatial representations (e.g., Engelen, Bouwmeester, de Bruin, & Zwaan, 2011; Stanfield & Zwaan, 2001). For example, Glenberg, Sato, and Cattaneo (2008) had participants transfer several hundred beans one by one from one jar to another, moving either in a direction toward or away from the body and later had these participants make plausibility judgments of sentences with implied directional transfer. Participants were slower to respond to sentences with implied motion in the direction they had been moving the beans, indicating that overworked motor effectors were interfering with semantic processing. Other studies have shown that hand- and foot-based verbs modulate motor evoked potentials measured in the hands and feet, respectively (Buccino et al., 2005), that processing hand and foot action sentences differentially facilitates manual and pedal responses (Scorolli & Borghi, 2007), and that there is a benefit for responding to concrete words with the hands and abstract relations with the mouth (Granito, Scoroli, & Borghi, 2015). In addition to these motor embodiment effects, a number of studies have shown that language is also embodied in visuospatial processing (Stanfield & Zwaan, 2001; Zwaan, Stanfield, & Yaxley, 2002; Zwaan & Yaxley, 2003). Most relevant to this study is the study by Zwaan and Yaxley (2003), in which participants made semantic relatedness judgments to word pairs that were or were not related and which (among related pairs) were presented in either canonical or non-canonical vertical positions relative to one another. For example, one item (coffee and foam) presented coffee either below or above foam in canonical or non-canonical conditions, respectively. When word pairs were presented to the left visual field (and hence the right hemisphere), there was an effect of position, but when presentation was to the right visual field (left hemisphere), position had no effect.

Reminiscent of the work in embodied cognition, cognitive grammarians have argued that syntax reflects iconic properties of the world. Langacker (2001) proposed that word order, and thus the conceptual evolution of a linguistic utterance over time, is a central component in that utterance's meaning. Some important factors to consider about word order are how something is expressed (the surface structure used), the conceptual order, and the real-world ordering of events. Most languages tend to order words in such a way that these different factors line up. For example, of the six possible orderings of the main syntactic elements of a transitive sentence (i.e., subject (S), verb (V) and object (O)), SVO and SOV account for the basic word orders of roughly 90% of the world's languages (Dryer, 2006). These orderings may not be accidental but rather may iconically reflect the causal flow of events (Croft, 1991). Along similar lines, these orderings may be the product of constraints of language production (MacDonald, 2013), such that subjects may be accessed from memory and hence produced first because they are more salient figures in an action. Generally, these various accounts predict that the notion of time is an important aspect in linguistic meaning and is therefore tied into the very grammatical structure of a sentence. This provides a serendipitous marriage between the notion of language with those of magnitude and space (discussed further below). Relatedly, Talmy (2000) introduced the concept of force dynamics, in which all interactions are processed in a manner akin to how physical forces act upon one another. In any sentence there are agonists and antagonists, taking the roles of agents, patients, subjects, objects, etc. Verbs express motion, be it actual, metaphorical, or unrealized. This motion is a result of the struggle to action between the antagonist and agonist. This can apply to not only concrete but also abstract concepts, such as the verb want, which, according to Talmy's analysis, implies psychological pressure or force toward some goal. Thus, force dynamics evoke Lakoff and Johnson's (1980) proposal of the grounding of abstract concepts in the concrete as well as the more general notion of grounded cognition, where language comprehension is embodied in sensorimotor processes (e.g., Glenberg & Kaschak, 2002; Zwaan & Taylor, 2006). However, Talmy's proposal (and what we will argue for in this study) is that syntax is also in some sense grounded.

One of the first studies to show that there may indeed be an empirically observable relation between syntax and space came from Chatterjee, Maher, Gonzalez Rothi, and Heilman (1995). These authors described the case of a left-handed patient (WH) suffering from a right hemisphere stroke that left him agrammatic. In one of the experiments from that study WH listened to transitive sentences in the active or passive voice and selected between two pictures that either accurately or inaccurately depicted the action. The stimuli described circle and square cartoon characters, effectively eliminating animacy plausibility cues. While WH performed at chance for both active and passive sentences in this experiment, if the lateral direction of movement of the action was controlled for, it was found he performed at ceiling. Specifically, he performed accurately on active sentences when they described left-to-right action, and on passive sentences when they described right-to-left action. Thus, WH was using a combination of spatial and temporal information to reconcile the pictures with the sentences. Building upon the results from this patient study, Chatterjee et al. (1999) showed that when drawing images, healthy English-speaking participants tended to place the agent of the action on the left side of the picture. In addition, they were faster in matching sentences to images when the image depicted the agent of a transitive action on the left and the patient on the right. Maas and Russo (2003) tested whether the effects Chatterjee and colleagues had reported so far were an effect of the orthographic direction of the language spoken by participants (i.e., the left-to-right direction of English), comparing the performance of Italian and Arabic speakers on a sentence-picture matching task similar to the one used in Chatterjee et al.'s experiments. They found that Italian speakers responded more quickly to images with the agent on the left, and Arabic speakers to images depicting agents on the right, suggesting that indeed orthographic direction drives this effect. However, in addition to this finding, both language groups showed a benefit to responding to images with action moving in a left-to-right direction (e.g., push and pull with agents on the left and right sides, respectively), which may suggest that in addition to orthography, the neural lateralization of language is also contributing to the effect. Specifically, Chatterjee (2001) proposed that individuals with left-lateralized language may show facilitation in responding to left-to-right flowing action as a result of left hemispheric regions processing such types of motion (Kinsbourne, 1987). Indeed, left hemispheric regions are involved in launching contraversive saccades (Herter & Guitton, 2004) and manual movements (McNeil & Pedelty, 1995) toward the right region of space. Left-to-right motion processing may recruit occipital-temporal regions and inferior and superior parietal cortices in the left hemisphere (Wu, Morganti, & Chatterjee, 2008). While the studies by Chatterjee et al. and Maas and Russo suggest native language and language lateralization together lead to interactions between space and verbal processing, Altmann, Saleem, Kendall, Heilman, and Gonzales Rothi (2006), using a larger n (18) of verbs than Maas and Russo (4: give, take, push, and pull), failed to find an agent spatial bias or a verb directionality effect when testing both Arabic and English speakers. It is notable also that Chatterjee's experiments employed a low n of 12 verbs, leaving open the possibility that this spatial effect is driven by an idiosyncratic group of concrete verbs. Another problem is that all these studies employed methods involving either scene recognition (i.e., sentence-picture matching) or production (i.e., drawing), making it possible that language comprehenders do not automatically form these spatial representations but rather only do so when the task involves relating language to visual scenes.

In addition to these various hand-language and language-space connections, there is a clear relationship between space and hand. For example, the Simon effect (the faster responding to a stimulus that shares some task-irrelevant dimension with a response) has typically been elicited using tasks involving stimuli presented at discrete locations that match the lateralization of manual responses (e.g., Iani, Baroni, Pellicano, & Nicoletti, 2011; Rubichi & Nicoletti, 2006; Simon & Rudell, 1967; Simon & Small, 1969). Mental representations of space are impacted by our ability to manipulate surrounding stimuli, and prosthetic extensions of the hand that bring distant regions of space within reach alter the neural receptive field associated with those spatial regions (Colby & Goldberg, 1999; Iriki, Tanaka, & Iwamura, 1996). Indeed, some studies have shown that processing stimuli in near-hand space leads to greater contributions of the magnocellular versus parvocellular visual pathways (Chan, Peterson, Barense, & Pratt, 2013; Gozli, West, & Pratt, 2012). This effect has also been shown to occur in language, where semantic judgments are degraded in near-hand space (Davoli, Du, Montana, Gaverick, & Abrams, 2010). These studies thus suggest that the relation between hand and space is an important factor to consider when exploring the connection between linguistic and spatial processing.

Many studies have found an intimate connection between magnitude and space (i.e., the spatial-numerical association of response codes; SNARC; Dehaene et al., 1993; Hubbard et al., 2005). The association between number and space can be observed in numerous ways. For example, processing mathematical operation signs produce spatial biases such that addition primes the right and subtraction the left (Pinhas, Shaki, & Fischer, 2014). Bodily movement toward the left or right primes small or large numbers, respectively (Anelli, Lugli, Borghi, & Nicoletti, 2014; Hartmann, Grabherr, & Mast, 2012). Some studies have implicated the importance of the hand in the SNARC effect. For example, the typical SNARC effect disappears when the hands are crossed (Wood, Nuerck, & Willmes, 2006), and differences between finger counting habits both within (Fischer, 2008) and across (Domahs, Moeller, Huber, Wilmes, & Nuerk, 2010) cultures modulate the SNARC effect. It should be noted, however, that this effect has also been elicited via responding with the feet (Schwarz & Müller, 2006), suggesting that a combination of finger counting habits and response lateralization contributes to the SNARC effect. The SNARC is also modulated to some degree by the orthographic direction of the native language (Dehaene et al., 1993). Furthermore, over-learned ordinal sequences (e.g., letters of the alphabet and months) also produce a SNARC-like effect (Gevers, Reynvoet, & Fias, 2003). Together, these findings suggest that experience with finger counting and exposure to a left-to-right orthography over the course of development may lead to the left-small/right-large SNARC effect, but others have shown that similar associations exist in infants (de Hevia, Girelli, Addabbo, & Cassia, 2014) and domestic chicks (Rugani, Vallortigara, Priftis, & Regolin, 2015).

Other effects related to the SNARC that go beyond written sequences may also prove to be important in language processing. First, Ishihara et al. (2008) showed that judgments of stimulus onset (short vs. long) are made faster to the left than to the right, respectively (the spatial-temporal association of response codes; STEARC). This seems especially relevant to language, given the fact that the speech signal unfolds over (and thus is dependent on) time. In this sense, whether a language is written from left-to-right, right-to-left, or top-to-bottom, the STEARC effect would predict that words at the beginning of a sentence should evoke a more leftward bias than words appearing farther down the speech stream, based on their association with relative moments in time. Xuan, Zhang, He, and Chen (2007) also observed that more of something (e.g., 7 dots vs. 3 dots; bright lights vs. dim lights; big squares vs. small squares) is judged to last longer than less of something, an issue that may be important with regards to the relative saliency of entities in a discourse model. Finally, Nuerk, Iverson, and Wilmes (2004) found that congruent linguistic markedness1 leads to facilitation of responses, whereas incongruency produces interference. For example, in a parity judgment task (typical to the SNARC literature) when even numbers are mapped onto right-hand responses and odd numbers onto the left, there will be facilitation, as both even and right are the less marked words of the pairs even and odd and left and right (the linguistic markedness of response codes; MARC effect; see also Roettger & Domahs, 2015).

While these findings may seem a bit far removed from the study of syntax, as we already mentioned, language is necessarily processed through various mediums relying on space and time. These interactions could potentially create SNARC-like effects in language processing and may elucidate some of the spatial biases of language that we reviewed above. Specifically, the SNARC literature would predict that any word appearing at the start of a sentence should show a left-side bias and any word at the end should show a right-side bias (if we assume a sentence as being roughly comparable to a left-to-right mental number line). The reasons for this (in English) are that beginning words appear early in the speech stream and in a leftward position in the text. This might similarly apply to the subjects and objects (or agents and patients) of a sentence, as they usually occur in predictable sentence locations. Thus, research from a number of areas, including cognitive grammar, visual scene recognition, embodied cognition, and numerical processing, all suggest that syntax may have an underlying lateral spatial component. Furthermore, research on the relation between hand and space suggests that use of the hands may change the nature of spatial processing, for example via altering the weights of the visual processing pathways or the contribution of the left and right hemispheres. If such spatial representations do exist, they might be explained differently, depending on the theoretical framework of interpretation. A cognitive grammarian invoking Talmy's force dynamics might explain that successfully parsing a syntactic structure involves depicting entities acting upon one another. However, from the perspective of the magnitude processing literature, language might epiphenomenally interact with space due to developmental priming from reading (a visuospatial process in itself). The purpose of the subsequent four experiments is first to provide evidence for a connection between syntax, space, and hand (E1) and then to determine the nature of this effect (E2, E3, E4), possibly allowing us to tease apart the contributions of the various theoretical frameworks. Transitive sentences are an ideal tool for studying the underlying spatial nature of language, as they require two participants occurring in a predictable sequence. Similarly, they can be passivized to reverse this sequence. Thus, we tested various aspects of transitive sentences. In E1 we compare the lateral bias of subjects and objects. In E2 we ask whether the final word of a sentence shows a stronger rightward lateral bias than other words in the sentence. In E3 we manipulate sentence voice to see if passivization leads to a reversal in the lateral bias of agents and patients. Finally, in E4 we determine the importance of the hand in the results of the previous experiments.

2 Experiment 1

Experiment 1 was designed to extend findings from other research lines and establish them in a simple reading paradigm. Our main interest was to investigate the possible spatial bias of different parts of transitive sentences: the subject and the object. We wondered if participants would show a preference for identifying the subject of a sentence when it appeared on the left side of space and identifying the object of a sentence when it appeared on the right. For testing this hypothesis, we developed a design inspired by techniques used in the Simon and SNARC effects literature (Hubbard et al., 2005; Simon, 1969), using a button box with left- and right-lateralized buttons, and with participants responding to probes appearing in stimulus-response matching or mismatching regions of space after they read a sentence presented word by word at the center of the screen.

2.1 Methods

2.1.1 Participants

Forty-eight2 participants (45 female, 3 male, M_age = 20.5) took part in this study for extra credit toward a psychology class. All participants were right-handed (self-reported), native speakers of English with normal or corrected-to-normal visual acuity (self-reported). Participants gave their informed consent to participate in this research under the guidelines of the University of South Carolina Institutional Review Board.

2.1.2 Procedure

Eighty transitive sentences were created, each consisting of two male or female human characters, with an equal number of all-female or all-male items (see Fig. 1 for a sample item and Table A1 for all stimuli used). The purpose of the gender manipulation was to prevent participants from forming any unconscious associations between gender and agency (e.g., male dominance associated with syntactic role dominance). Sentences were an average of 7.1 words long (range: 5–10). We equated word length of subject (M_length = 5.41) and object (M_length = 5.58) names within sentences such that on average possible probe words of a given sentence were of the same length (t = 0.7, p > .05). Sentences were presented one word at a time in the center of a computer monitor. Each trial began with a fixation cross lasting 1-s. This was followed by the sentence, 500 ms for each word with no delay between words. Following the final word, another fixation cross was presented for 1,500 ms. After this, a probe (either the subject or object of the previous sentence) appeared on either the left or right side of the screen. Participants were instructed to identify whether the probe was the subject or object of the previously read sentence and press a corresponding button on a PST Serial Response Box with a left button press when encountering the subject and a right button press for the object, and on the other half of trials button assignment was switched. Participants were told that the subject of the sentence was the person who appeared first in the sentence. Participants had 4,500 ms to respond to the probe before advancing to the next trial. Failures to respond were coded as incorrect. Items were divided into two blocks for the different button assignments, each beginning with eight practice trials presenting items in all possible conditions (Male vs. Female, Subject vs. Object, Left vs. Right side presentation). Participants could only advance into the experimental blocks after achieving at least 75% accuracy. Button assignment order was counterbalanced across subjects. Sentence presentation within block was randomized (see Fig. 1 for schematic representation of a sample trial and Appendix for a list of all items used across the experiments).

2.2 Results

For this and subsequent experiments we are primarily interested in response times to correct probe recognition. Therefore, before conducting data analyses we removed all incorrect responses from the dataset (21.18%). In addition, we reasoned that participants were not able to fully process and generate responses for items faster than 300 ms, whereas responses slower than 3,000 ms reflected failures to attend to the task. Thus, responses that did not fall in between these limits were removed from the dataset as well (0.19%). In all, then, 21.37% of responses were removed, and log-transformed response times were analyzed.

Data analyses were conducted in the lme4 package (Bates, Maechler, & Bolker, 2011) in R (v.3.1.1; R Development Core Team, 2014), using mixed-effects modeling with stimulus Side (Left vs. Right), response Hand (LHand vs. RHand), and Word type (Subject vs. Object) as fixed effects, as well as all possible interactions between the three factors, and random intercepts and slopes both by-participants and by-items3 (Baayen, Davidson, & Bates, 2008; Barr, Levy, Scheepers, & Tily, 2013). The reason for including random slopes for participants and items is that there may be variability in the sensitivity of different participants and items to the fixed effects (Barr et al., 2013). We compared the full model with a model that included all two-way interactions and main effects and found that the full model was a better fit, χ²(1) = 7.82, p = .005.4 Full model results are reported in Table 1 and shown in Fig. 2.

Table 1. Fixed effects for the max model in Experiment 1. Baseline level for Word is Object, for Side is Left, and for Hand is Left

Coefficient estimates
β	Condition	Est.	SE	t	p
β0	(Intercept)	6.655	0.032	206.382	<0.001***
β1	Subject	−0.1	0.018	−5.661	<0.001***
β2	RightSide	−0.00006	0.018	−0.003	≤0.997
β3	RHand	−0.054	0.018	−3.061	≤0.002**
β4	Subject × RightSide	0.046	0.025	1.832	≤0.067.
β5	Subject × RHand	0.113	0.025	4.534	<0.001***
β6	RightSide × RHand	0.014	0.025	0.546	≤0.585
β7	Subject × RightSide × RHand	−0.099	0.035	−2.798	≤0.005**

Note

• ***p < .001, **p < .01, *p < .05; borderline effects p < .1 and >.05.

To get a better understanding of the three-way interaction, we ran a series of post hoc comparisons, adjusting p values using the Bonferroni method. Due to the large number of possible contrasts, we only report in the text those that were of theoretical interest (see Table 2 for all contrasts). First, there was a benefit for responding to Subjects on the Left using the LHand as opposed to Objects on the Left with the LHand, β = 0.1, SE = 0.018, t = 5.66, p < .001. In addition, responses to Subjects on the Left with the LHand were faster than to Objects on the Right with the LHand, β = −0.1, SE = 0.018, t = −5.66, p < .001. Responses to Subjects on the Left with the LHand were faster than responses to Subjects on the Left with the RHand, β = −0.06, SE = 0.018, t = −3.33, p = .02. Responses to Subjects on the Left with the LHand were also faster than responses to Objects on the Right with the RHand, β = −0.06, SE = 0.018, t = −3.37, p = .02. Responses to Subjects on the Right with the RHand were faster than to Objects on the Left with the LHand, β = 0.081, SE = 0.018, t = 4.51, p < .001. All other contrasts were not statistically significant.

Table 2. Pairwise comparisons for the three-way interaction between Side (Left-L, Right-R), Word (Subject-Sub, Object-Ob), and Hand (Right Hand-RH, Left Hand-LH) in E1, each cell reporting coefficient difference estimates, SE, t ratios, and p values

Condition	Value	LSubH	LObRH	LSubRH	RObLH	RSubLH	RObRH	RSubRH
LObLH	Est.	.1	.054	.041	<.001	.054	.041	.081
	SE	.018	.018	.018	.018	.018	.018	.018
	t	5.65	3.06	2.29	.003	3.05	2.29	4.51
	p	<.001	.06	.62	1	.06	.62	<.001
LSubLH	Est.		−.046	−.059	−.1	−.046	−.06	−.02
	SE		.018	.018	.018	.018	.018	.018
	t		−2.61	−3.33	−5.66	−2.58	−3.37	−1.11
	p		.26	.02	<.001	.28	.02	1
LObRH	Est.			−.013	.054	<−.001	−.014	.026
	SE			.018	.018	.018	.018	.018
	t			−.75	3.06	−.01	−.77	1.49
	p			1	.06	1	1	1
LSubRH	Est.				.041	−.013	<−.001	.04
	SE				.018	.018	.018	.018
	t				2.29	−.75	−.02	2.23
	p				.62	1	1	.72
RObLH	Est.					.054	.041	.08
	SE					.018	.018	.018
	t					3.05	2.29	4.52
	p					.06	.62	<.001
RSubLH	Est.						−.014	.026
	SE						.018	.018
	t						−.78	1.47
	p						1	1
RObRH	Est.							.04
	SE							.018
	t							2.25
	p							.68

3 Experiment 2

In E2 we test whether the spatial effect found in E1 can be observed when no direct attention is drawn to the fact that the probe words are the subjects and objects of the previously read sentences. In addition, we compare recognition of subject and object to the final word in the sentence. van Dijck and Fias (2011) reported a SNARC-like effect for items in working memory, such that earlier appearing items in a list have a left-side advantage and later items a right-side advantage. Given the importance of working memory in language processing, it may be the case that the difference we observed between left- and right-side processing of subject and object probe words is merely related to order in the sentence (i.e., the order words are introduced into working memory). If this syntax-space effect is just driven by serial word order, then the final word should show the strongest right-side bias. In this task, then, participants judged whether a probe word was in the sentence they had just read (a go/no-go task). Finding a syntax-space effect using such a task will provide further evidence that language comprehenders naturally generate spatial schemata when encountering transitive sentences and that these schemata follow a predictable pattern with the tendency of subjects to be placed to the left of objects. In addition, not finding an effect of the final word of the sentence will provide evidence that the syntax-space effect is specific for the thematic roles in the sentence and not necessarily just the order words appear in a sentence.

3.2 Results

As we did in E1, before conducting data analyses, we removed all incorrect probe responses from the dataset (3.33%). Next, we removed trials in which participants made incorrect response to the comprehension question (4.39%).5 One participant was removed entirely due to only achieving around chance on the comprehension questions (57%). Finally, we trimmed out extreme responses faster than 300 ms and slower than 3,000 ms (0.29%). In all, then, 8.01% of responses were removed. For the below analyses we looked at log-transformed response times.

As in E1, data analyses were conducted in the lme4 package (Bates et al., 2011) in R (v.3.1.1; R Development Core Team, 2014), using mixed-effects modeling with stimulus position (Side) with levels Left and Right, between-subjects factor Hand with levels LHand and RHand, and word type (Word) with levels Subject, Object, and Final as fixed effects, as well as all possible interactions between the three factors. Items were treated as random effects, as well as participants, both with the interaction between Word and Subject as random slopes6 (Baayen et al., 2008). Full model results are reported in Table 3. Statistical analyses followed the same procedure as in E1, starting with the maximum model, and removing coefficient by coefficient to find the model of best fit. The three-way interaction was significant, χ²(2) = 7.12, p = .03 (see Fig. 3).

Table 3. Fixed effects for the max model in Experiment 2. Baseline levels are Object, Left Side, and Left Hand

Coefficient estimates
β	Condition	Est.	SE	t	p
β0	(Intercept)	6.694	0.029	230.867	<0.001***
β1	Final	0.007	0.025	0.27	≤0.787
β2	Subject	−0.001	0.025	−0.007	≤0.994
β3	RightSide	−0.047	0.025	−1.852	≤0.064.
β4	RHand	0.072	0.042	1.726	≤0.086.
β5	Final × RightSide	0.038	0.035	1.091	≤0.276
β6	Subject × RightSide	0.095	0.036	2.685	≤0.007**
β7	Final × RHand	−0.013	0.037	−0.346	≤0.73
β8	Subject × RHand	0.038	0.037	1.01	≤0.313
β9	SideRight × RHand	0.027	0.036	0.741	≤0.459
β10	Final × SideRight × RHand	−0.033	0.052	−0.641	≤0.521
β11	Subject × SideRight × RHand	−0.137	0.052	−2.633	≤0.009**

Note

• ***p < .001, **p < .01, *p < .05; borderline effects p < .1 and >.05.

We explored the three-way interaction by dividing the datasets into the LHand and RHand participant groups, then reran the analyses for each group with just the fixed effects of Word, Side, and the Word by Side interaction. There was an interaction for the LHand group, χ²(2) = 6.96, p = .03, but not the RHand group, χ²(2) = 1.56, p = .46. Since the interaction was not significant for the RHand group, we further tested the main effects. Word was not significant, χ²(2) = 0.99, p = .61, but Side was, χ²(1) = 4.5, p = .03, with faster Right Side responses than Left Side.

Regarding the Word by Side interaction on the LHand group, we ran pairwise comparisons using a Bonferroni adjustment. Only one contrast was significant, finding that Right-Side Object responses were faster than Right-Side Subject responses, β = −0.091, SE = 0.023, t = −4.025, p < .001. Crucially, there were no differences between responses to Final and Object words on the Right Side, β = −0.043, SE = 0.022, t = 1.92, p = .83 (see Table 4 for all contrasts).

Table 4. Pairwise comparisons for the two-way interaction between Side (Left-L, Right-R) and Word (Subject, Object, Final) in E2 for the LHand group, each cell reporting coefficient difference estimates, SE, t ratios, and p values

Condition	LObject	LSubject	RFinal	RObject	RSubject
LFinal	.007	.009	.008	.051	−.04
	.022	.022	.022	.022	.022
	.31	.412	.385	2.362	−1.811
	1	1	1	.276	1
LObject		.002	.002	.045	−.047
		.023	.023	.023	.023
		.091	.067	1.963	−2.032
		1	1	.75	.637
LSubject			−.001	.043	−.049
			.022	.022	.022
			−.023	1.905	−2.173
			1	.857	.451
RFinal				.043	−.048
				.022	.023
				1.92	−2.134
				.828	.497
RObject					−.091
					.023
					−4.025
					<.001

4 Experiment 3

Both E1 and E2 found effects of a spatial bias for actors in a transitive sentence, although they differed somewhat in nature, possibly due to the involvement of response laterality. However, E1 suggested a left-hand, left-side penalty for objects, and E2 found that object responses were faster on the right side than the left, showing some cross-experiment concordance. The question remains whether this is an effect of Subject- and Object-hood (i.e., order of mention) or Agent- and Patient-hood (i.e., role in the transitive sentence). The most obvious way to test this is to see whether passivization of transitive sentences reverses the effect we have so far observed. If it does lead to agents now being responded to slower than patients on the left side and/or with the left hand, we can conclude that any relation between syntax and space is related to order of mention (subject before object) and not the underlying causal structure of the event. On the other hand, if there is no reversal (agents are still responded to faster than patients on the left side and/or with the left hand despite being the grammatical objects of the passive sentences), we can conclude that transitive sentences evoke a kind of action space with placeholders for agents and patients. Thus, in E3 we add a further level of complexity to the design: sentence voice. In addition, we reintroduce response laterality into the design (the reason for doing this is described in the Methods section).

4.2 Results

Following the same procedures from E1 and E2, we removed all trials with incorrect probe responses (6.16%), those with incorrect responses to comprehension questions (17.09%), and those with extreme responses (0.32%), a total of 23.57% of trials. In addition, we removed five participants' data due to chance performance (M = 54%) on the comprehension questions. Statistical analyses followed the same procedure as in previous experiments, starting with the maximum model, and removing coefficient by coefficient to find the model of best fit7 (see Table 5 for the maximum model). It was found that the four-way interaction was unnecessary, χ²(1) = 1.6, p = .64. We next proceeded to remove those coefficients that were most theoretically interesting to this study. Removing the three-way interaction between Side, Voice, and Word resulted in a significant reduction in variance accounted for, χ²(1) = 4.41, p = .04 (see Fig. 4). Next we removed the Word, Hand, and Voice interaction. This too explained a significant amount of variance, χ²(1) = 3.98, p = .05 (see Fig. 5). Removing other three-way interactions did not result in any significant changes: removing the Side, Word, and Hand interaction, χ²(1) = 1.69, p = .19; removing the Side, Hand, and Voice interaction, χ²(1) < 1.

Table 5. Fixed effects for the max model in Experiment 3. Baseline levels are Agent, Left Side, Left Hand, and Active Voice

Coefficient Estimates
β	Condition	Est.	SE	t	p
β0	(Intercept)	6.909	0.025	277.731	<0.001***
β1	RightSide	0.03	0.021	1.377	≤0.169
β2	Patient	0.107	0.021	5.023	<0.001***
β3	RHand	−0.021	0.022	−0.929	≤0.353
β4	Passive	0.097	0.022	4.469	<0.001***
β5	RightSide × Patient	−0.05	0.028	−1.67	≤0.095.
β6	RightSide × RHand	−0.072	0.03	−2.347	≤0.019*
β7	Patient × RHand	−0.081	0.03	−2.631	≤0.009**
β8	RightSide × Passive	−0.02	0.031	−0.652	≤0.515
β9	Patient × Passive	−0.224	0.03	−7.437	<0.001***
β10	RHand × Passive	−0.021	0.031	−0.673	≤0.501
β11	RightSide × Patient × RHand	0.054	0.043	1.257	≤0.209
β12	RightSide × Patient × Passive	0.077	0.042	1.824	≤0.068.
β13	RightSide × RHand × Passive	0.019	0.043	0.443	≤0.658
β14	Patient × RHand × Passive	0.074	0.043	1.734	≤0.083.
β15	RightSide × Patient × RHand × Passive	−0.028	0.06	−0.466	≤0.641

Note

• ***p < .001, **p < .01, *p < .05; borderline effects p < .1 and >.05.

Next, we analyzed the interaction effects using pairwise comparisons. For each interaction we report adjusted p values using the Bonferroni method, reporting only the theoretically significant results in the text and the rest of the comparisons in the respective tables. Regarding the interaction between Voice, Word, and Side, there was a significant advantage for responses to Agents over responses to Patients on the Left in the Active Voice, β = −0.067, SE = 0.015, t = −4.42, p < .001, but this difference was not significant on the Right in the Active Voice, β = −0.044, SE = 0.015, t = −2.9, p = .11. In the Passive Voice, responses to Patients were faster than those to Agents on both the Left, β = 0.13, SE = 0.015, t = 7.84, p < .001, and Right, β = 0.079, SE = 0.015, t = 5.25, p < .001 (see Table 6 for all contrasts).

Table 6. Pairwise comparisons for the three-way interaction between Side (Left-L, Right-R), Word (Agent-A, Patient-P), and Voice (Active-Act, Passive-Pass) in E3, each cell reporting coefficient difference estimates, SE, t ratios, and p values

Condition	Value	RAAct	LPAct	RPAct	LAPass	RAPass	LPPass	RPPass
LAAct	Est.	.006	−.067	−.037	−.086	−.07	.033	.01
	SE	.015	.015	.015	.015	.015	.015	.015
	t	.41	−4.42	−2.45	−5.59	−4.57	2.18	.63
	p	1	<.001	.4	<.001	<.001	.82	1
RAAct	Est.		−.073	−.044	−.092	−.076	.027	.004
	SE		.015	.015	.015	.015	.015	.015
	t		−4.89	−2.9	−6.07	−5.04	1.8	.23
	p		<.001	.11	<.001	<.001	1	1
LPAct	Est.			.03	−.019	−.003	.1	.077
	SE			.015	.015	.015	.015	.015
	t			1.98	−1.28	−.2	6.67	5.08
	p			1	1	1	<.001	<.001
RPAct	Est.				−.049	−.032	.071	.047
	SE				.015	.015	.015	.015
	t				−3.21	−2.15	4.69	3.1
	p				.04	.88	<.001	.05
LAPass	Est.					.017	.12	.096
	SE					.015	.015	.015
	t					1.08	7.84	6.25
	p					1	<.001	<.001
RAPass	Est.						.103	.079
	SE						.015	.015
	t						6.83	5.25
	p						<.001	<.001
LPPass	Est.							−.024
	SE							.015
	t							−1.56
	p							1

Now, we turn to the other significant three-way interaction, Voice, Word, and Hand. Focusing first on the Active Voice, there was an advantage for responding to Agents over Patients on the LHand, β = −0.082, SE = 0.015, t = −5.39, p < .001, but no such advantage for the RHand, β = −0.028, SE = 0.015, t = −1.82, p = 1. For the Passive Voice, there was an advantage for responses to Patients over Agents on both the LHand, β = 0.103, SE = 0.015, t = 6.7, p < .001, and the RHand, β = 0.096, SE = 0.016, t = 6.18, p < .001 (see Table 7 for all contrasts).

Table 7. Pairwise comparisons for the three-way interaction between Hand (Right Hand-RH, Left Hand-LH), Word (Agent-A, Patient-P), and Voice (Active-Act, Passive-Pass) in E3, each cell reporting coefficient difference estimates, SE, t ratios, and p values

Condition	Value	PassALH	ActPLH	PassPLH	ActARH	PassARH	ActPRH	PassPRH
ActALH	Est.	−.086	−.082	.016	.056	−.019	.028	.077
	SE	.015	.015	.015	.016	.016	.015	.016
	t	−5.68	−5.39	1.05	3.59	−1.2	1.84	4.98
	p	<.001	<.001	1	.01	1	1	<.001
PassALH	Est.		.004	.103	.143	.068	.115	.164
	SE		.015	.015	.016	.016	.015	.016
	t		.28	6.7	9.08	.43	7.47	10.55
	p		1	<.001	<.001	<.001	<.001	<.001
ActPLH	Est.			.098	.139	.063	.111	.16
	SE			.015	.015	.015	.015	.015
	t			6.6	9.06	4.14	7.16	10.22
	p			<.001	<.001	.001	<.001	<.001
PassPLH	Est.				.04	−.035	.012	.061
	SE				.015	.015	.016	.016
	t				2.62	−2.27	.79	3.91
	p				.25	.65	1	.003
ActARH	Est.					−.075	−.028	.021
	SE					.015	.015	.016
	t					−4.91	−1.82	1.35
	p					<.001	1	1
PassARH	Est.						.047	.096
	SE						.015	.016
	t						3.06	6.18
	p						.06	<.001
ActPRH	Est.							.049
	SE							.015
	t							3.22
	p							.04

Finally, since this experiment involved responding to False probes, we analyzed these separately to test the possibility that responding to probes unassociated with any linguistic context will also interact with manual or spatial processes. For this analysis the maximal model included the variables Hand, Side, and Voice, with participants and items as random factors with maximal random slope structure.8 As before we proceeded to remove interaction effects one by one, comparing the change in variance with the previous model (see Table 8). The only significant effect was a main effect of Hand, with RHand responses slower than LHand, χ²(1) = 11.85, p < .001, likely indicating participants' preference to respond with their right hand to indicate correct responses and their left hand to indicate incorrect ones.

Table 8. Fixed effects for the max model of False probe responses in Experiment 3. Baseline levels are Left Hand, Left Side, and Active Voice

Coefficient Estimates
β	Condition	Est.	SE	t	p
β0	(Intercept)	6.866	0.023	294.224	<0.001***
β1	RightSide	−0.025	0.014	−1.783	≤0.075.
β2	RHand	0.013	0.014	0.918	≤0.359
β3	Passive	−0.0002	0.014	−0.02	≤0.984
β4	RightSide × RHand	0.02	0.02	0.995	≤0.32
β5	RightSide × Passive	0.015	0.02	0.755	≤0.451
β6	RHand × Passive	0.007	0.02	0.329	≤0.742
β7	RightSide × RHand × Passive	0.001	0.028	0.046	≤0.963

Note

• ***p < .001, **p < .01, *p < .05; borderline effects p < .1 and >.05.

5 Experiment 4

So far we have provided evidence for a syntax-space effect (E1), which is still present when response laterality is removed (E2) but is nevertheless only present in active sentences, suggesting that the saliency of the agent drives the effect (E2 and E3). While response laterality was removed in E2, responses were still made manually. This final experiment, therefore, asks whether this effect remains when participants must make vocal responses using a design similar to what was used in E3. We hypothesize that, given the previously observed hand interactions, indeed the syntax-space effect is intimately connected with the laterality of the effector, and thus no effect will be present in E4.

5.2 Results

Following the same procedures used previously, we removed all trials with incorrect probe responses (6.25%), incorrect responses to the comprehension question (19.05%), and those with extreme responses (2.87%), a total of 28.17% of trials. In addition, we removed seven participants' data due to chance performance (M = 46%) on the comprehension questions. Note that questions in E3 and E4 varied between active and passive voice. Thus, half of the questions appeared in the opposite voice to the one presented in the item (e.g., a passive voiced question following an active voiced item). This may account for the relatively low accuracy on the comprehension questions in these two experiments compared to in E2. Log-transformed response times to probes were analyzed.

Statistical analyses followed the same procedure as in previous experiments, starting with the maximum model, and removing coefficient by coefficient to find the model of best fit9 (see Table 9 for the maximum model). It was found that the three-way interaction between Voice, Word, and Side was unnecessary, χ²(1) = 0.65, p = .42, so we proceeded to remove the two-way interaction coefficients from the model (see Fig. 6). The Side by Word interaction did not account for a significant amount of variance, χ²(1) = 0.05, p = .82. However, removing the Word by Voice interaction did account for a significant amount of variance, χ²(1) = 24.23, p < .001. The Side and Voice interaction was also unnecessary, χ²(1) = 0.96, p = .33.

Table 9. Fixed effects for the max model in Experiment 4. Baseline levels are Agent, Left Side, and Active Voice

Coefficient Estimates
β	Condition	Est.	SE	t	p
β0	(Intercept)	6.933	0.029	236.326	<0.001***
β1	RightSide	−0.0001	0.017	−0.004	≤0.997
β2	Patient	0.045	0.017	2.582	≤0.01**
β3	Passive	0.042	0.018	2.415	≤0.016*
β4	RightSide × Patient	−0.018	0.025	−0.731	≤0.465
β5	RightSide × Passive	0.002	0.025	0.098	≤0.922
β6	Patient × Passive	−0.1	0.025	−4.052	<0.001***
β7	RightSide × Patient × Passive	0.028	0.035	0.808	≤0.419

Note

• ***p < .001, **p < .01, *p < .05; borderline effects p < .1 and >.05.

Next, we present post hoc comparisons of the interaction between Word and Voice, using the appropriate Bonferroni method of p value adjustment. Responses to Active Agents were on average faster than those to Passive Agents, β = −0.044, SE = 0.012, t = −3.52, p = .003, and Active Patients, β = −0.036, SE = 0.012, t = −2.92, p = .02, whereas responses to Passive Patients were faster than those to Passive Agents, β = 0.05, SE = 0.012, t = 4.07, p < .001, and Active Patients, β = 0.042, SE = 0.012, t = 3.45, p < .001 (see Table 10 for all contrasts).

Table 10. Pairwise comparisons for the two-way interaction between Word (Agent-Ag, Patient-Pat) and Voice (Active-Act, Passive-Pass) in E4, each cell reporting coefficient difference estimates, SE, t ratios, and p values

Condition	Value	PassAg	ActPat	PassPat
ActAg	Est.	−.044	−.036	.006
	SE	.012	.012	.012
	t	−3.52	−2.92	.512
	p	.003	.02	1
PassAg	Est.		−.008	.05
	SE		.012	.012
	t		−.62	4.07
	p		1	<.001
ActPat	Est.			.042
	SE			.012
	t			3.45
	p			<.001

As we did in E3, we analyzed the False probe responses only, using similar analyses as with the Present probe responses. The maximal model included the factors Side and Voice and the interaction, and participants and items as random factors with maximal random slope structure.10 None of the effects were statistically significant (see Table 11 for the maximal model).

Table 11. Fixed effects for the max model of False Probe responses in Experiment 4. Baseline levels are Left Side and Active Voice

Coefficient Estimates
β	Condition	Est.	SE	t	p
β0	(Intercept)	6.977	0.027	262.253	<0.001***
β1	RightSide	−0.011	0.012	−0.92	≤0.358
β2	Passive	0.004	0.012	0.356	≤0.722
β3	RightSide × Passive	0.019	0.017	1.113	≤0.266

Note

• ***p < .001, **p < .01, *p < .05; borderline effects p < .1 and >.05.

6 General discussion

The goal of this study has been to investigate the possibility that the actors in transitive sentences come with lateral spatial biases, as follows from claims that the syntactic frame of sentences is spatial in nature (e.g., Chatterjee, 2001; Talmy, 2000). We tested this idea using a novel technique inspired by tasks used in the magnitude-space and Simon effect literature (e.g., Dehaene et al., 1993; Gevers et al., 2003; Rubichi & Nicoletti, 2006). This new design may prove to be useful in the future for researchers interested in further exploring the relationship between language, space, and hand. E1 had participants read transitive sentences and judge whether probe words presented to the left and right side of the computer monitor were the subjects or objects of the sentences they had read, making lateralized button presses. E2 employed a go/no-go task, with participants responding to subjects, objects, and the final words of the sentences with a central button press. E3 combined strengths of the designs from E1 and E2 with the introduction of the variable of sentence voice. Finally, E4 removed button pressing from the design, repeating E3 with vocal responses. All experiments except E4 found similar interactions among the variables of word and space. The involvement of the hand indicated stronger syntax-space effects either when responding toward the left side of space or when right hemisphere motor regions were engaged (Zwaan & Yaxley, 2003) and perhaps greater involvement of more dorsal visual pathways (Gozli et al., 2012). Specifically, E1 found a penalty associated with left-side object responses made with the left hand. The contributions of the hands in this experiment may have reflected a preference for assigning the subject to the left hand and the object to the right, potentially because this hand-word assignment mirrors the proposed syntactic spatial frame. However, E2 also found a spatial effect dependent upon response hand, showing that the spatial effect associated with subjects and objects does not depend on response laterality but the specific hand responding. Results also showed the final words show no facilitation or penalty on either side. E3 found that voice and word interacted separately with space and hand. Both of these interactions can be described as showing spatial effects for the active voice, such that patients have a left-side penalty. E4 only found a word-voice interaction. It should be pointed out that while the interaction between word and voice observed in E4 was not the main interest of this study, it does verify the validity and reliability of the paradigm we have developed. If in using this design we were able to replicate well-established psycholinguistic findings (e.g., Ferreira, 2003), the design is probably also suitable to detect other non-canonical effects (see Table 12 for summary of findings). We consider the implications of all of these findings below.

First, to get a better understanding of what the findings from these experiments can tell us about the relation between transitivity, space, and hand, let us break down the task of comprehending a transitive sentence. From a strictly formalistic syntactic view, there are two distinct thematic roles, and there is an action originating from one (the subject or agent) and terminating at the other (the object or patient). Knowledge of the event crucially depends on understanding how it unfolds in time. Langacker (2001) argued that when word order corresponds with conceptual order the sentence should be easier to understand, and indeed, difficulty in understanding passive sentences may be a result of the formation of shallow representations, based on these conflicting orders (Ferreira, Bailey, & Ferraro, 2002). Difficulty with passives may also involve the movement of syntactic elements within the sentence's deep structure to their position in the surface structure (Chomsky, 1957). The idea of conflicting temporal representations (e.g., between the fleeting memory of words in the sentence and the imagery formed of the event) might result in a neutral spatial preference. On the other hand, when the two factors align, a clear preference is formed for agents on the left and patients on the right. This was the case in E3, in which the passive voice resulted in no spatial bias, whereas the active voice did result in such a bias. The spatial bias fits in with those observed in previous research looking at scene production and recognition (e.g., Chatterjee, 2001; Chokron & De Agostini, 2000; Maas & Russo, 2003) and potentially even with the findings from the magnitude-space literature (e.g., Bueti & Walsh, 2009; Dehaene et al., 1993), all of which seems to be related to orthographic directionality (although speakers of languages with right-to-left orthographies will have to be tested using our paradigm in the future to confirm this intuition). However, it should be noted that orthography may not be the driving factor here. de Hevia et al. (2014) showed preverbal infants prefer magnitudes increasing from left-to-right over right-to-left, and Rugani et al. (2015) reported that the domestic chick (Gallus gallus) also shows evidence for a left-to-right mental number line. In addition, Maas and Russo (2003) found that orthographic direction seems to have a combinatorial effect with the overall preference to process actions as moving from left-to-right, exaggerating the effect in speakers of Romance languages and reducing it in Arabic speakers. In sum, agents are leftward and patients rightward due to the order within which they canonically occur, both superficially and conceptually. Crucially, we show that this generalization holds for a wide range of verbs (E2 and E3 used a total of 160 different transitive verbs) varying in imageability. However, when a noncanonical order is used, there is no preference, and given the various bodies of evidence just reviewed, we would not expect to find one. In addition, this syntax-space effect only seems to emerge when there is a direct interaction with space (via the hands in our experiments).

It is intriguing to note that the spatial effect seems to disappear (or is at least reduced) on the right (both hand and side). This can be explained in at least a couple of ways. One possible explanation involves language laterality and is based on the assumption that the spatial representations evoked by transitive sentences in right-handed people are neurally right lateralized. When we respond with our right hands (or vocally), we are involving left hemispheric motor regions to a greater extent than right hemispheric ones. In this case, the spatial representation that was formed is not directly accessed. However, when left-hand responses are made, the spatial representation facilitates (or interferes with) responding, based simply on the priming of these structures. In line with this explanation are reports that hemispatial neglect, characterized by a deficit in attention to the left side of space, typically results from damage to the right parietal and temporal cortices (Karnath & Rorden, 2012), homologs of left-hemispheric regions implicated in language processing and more specifically in the processing of transitive syntactic frames (Den Ouden, Fix, Parrish, & Thompson, 2009; Shetreet, Palti, Friedmann, & Hadar, 2007; Thompson et al., 2007). Furthermore, the motor cortex engages regions in the parietal lobes involved with spatial planning (Colby & Goldberg, 1999). This explanation is also in line with the findings of Zwaan and Yaxley (2003), who reported that the canonical positioning of stimuli impacted semantic judgment response times only when the stimuli were presented to the right hemisphere (via the left visual field). Under this explanation, the non-linguistic right-hemisphere homologs, interacting with right-hemisphere motor regions, take a supporting role in processing the spatial-syntactic frame through direct activation from the linguistic stimuli. Also, not necessarily alternative to the above explanations, but possibly playing a contributing role are the observations that the involvement of the hands alters how stimuli are processed (Chan et al., 2013; Gozli et al., 2012). Specifically, when stimuli are processed in near-hand space, there is greater contribution of dorsal visual pathways. All of these various studies taken together point toward the involvement of right parietal structures via left-handed responses in contributing to the syntax-space effect.

Of course, it is also possible that using the hands generally enhances spatial attention and that this attention is flexible, depending on such factors as the positioning of the hands in space. If this is indeed the case, the various hand-related effects reported in E1 and E3 were not actually hand effects, but rather were spatial effects in the sense that responding required interacting with discrete locations in space. Indeed, in the experiments in which responses were lateralized we found the strongest results (recall the small number of significant pairwise comparisons in E2). This would suggest that perhaps it is not necessarily the hands that are important for eliciting this effect. Instead, it is having a spatial dimension in the response (no matter what the motor effector) that is really crucial. According to this explanation, using different fingers of one hand to press lateralized buttons, using the feet to press pedals, or even nodding toward different locations may produce similar effects. The fact that SNARC effects have been elicited when using pedal responses lends some support to this notion (Schwarz & Müller, 2006). That there may be separate contributions of neural and response lateralization suggests the need for a more careful manipulation of the positioning of the hands as well as comparisons with other motor effectors. Indeed, studies looking at SNARC (Wood et al., 2006) and STEARC (Vallesi, Binns, & Shallice, 2008) have found that crossing the hands eliminates the respective effects, indicating a role for both hand and space. Similarly, Rubichi and Nicoletti (2006), finding larger Simon effects in the dominant-hand controlled hemifield, showed that crossing hands shifts the effect over to the opposite hemifield. This proposal of course would raise innumerable possible variations on designs for testing the nature of this syntax-space effect.

Finally, both of these suggestions may be partially correct, as we have reported in these experiments that while all three elements of language, hand, and space interact, they do not do so in a very straightforward way. For example, E3 found two separate three-way interactions, which suggests that both the hand and space contribute to language processing separately. Indeed, it has been suggested that praxis and language overlap to some degree, both conceptually and neuroanatomically (Kobayashi & Ugawa, 2013); the hands provide a ready tool for the deaf in signing, a language as complex as any spoken language (Emmorey, 1999); manual gesturing facilitates speech production (Rauscher, Krauss, & Chen, 1996); it has been suggested that the human capacity for language evolved out of a gestural protolanguage (Corballis, 2010; Rizzolatti & Arbib, 1998); at an ontological level, gestural babbling precedes vocal babbling in the human infant (Petitto & Marentette, 1991; Volterra & Iverson, 1995); and (gesturing aside) the importance of the hands in language has found a resurgence in such recent technological advances as texting and online chatting, the immediacy of which emulate that of spoken conversation (Corballis, 2011).

The relation between the hands and spatial processing seems noncontroversial. Use of the hands guides our understanding of space, though certainly one can lack hands and still have a concept of space. Similarly, while some researchers have noted the importance of the hands even in vocal speech production (Rauscher et al., 1996), it is clear also that language can and does exist without the hands. The connection of language and space is perhaps more speculative, though even this area has its proponents (e.g., Chatterjee, 2001; Croft, 1991; Langacker, 2001; Talmy, 2000), and indeed the rare co-occurrence of aphasia and neglect has been noted following left-hemispheric insults (Suchan & Karnath, 2011). Reading is clearly a visuospatial task, and its effects on development are of critical importance. However, speakers from non-literate cultures may not show any connections between syntax and space when tested empirically, and Arabic or Hebrew speakers may show the reverse of the effects we have shown here (or perhaps a right-side agent/left-side patient advantage on the left hand only). Evidence from both populations will be necessary to further elucidate the syntax-space effect. In addition, given the importance of semantics in spatial processing (e.g., Bergen, Lindsay, Matlock, & Narayanan, 2010; Zwaan & Yaxley, 2003), there are likely to be some differences between the spatial effect elicited by different transitive verbs, whether additive or interactive. For example, verbs producing different causal inferences (i.e., implicit causality) might show reversed or reduced syntax-space effects, as well as action verbs denoting action flowing in opposite directions (e.g., push/pull as per Chatterjee et al., 1999; Maas & Russo, 2003). The results of these experiments thus leave us to consider the three distinct yet intertwining (and perhaps overlapping) systems of language, space, and hand.