Measuring language lateralisation with different language tasks: a systematic review

LI strength, reproducibility and robustness

Across the papers reviewed here, verbal fluency tasks are consistently reported as yielding the strongest laterality when compared to other receptive and expressive tasks within studies (Baciu et al., 2005; Deblaere et al., 2002; Van der Haegen, Cai & Brysbaert, 2012; Dodoo-Schittko et al., 2012; Harrington, Buonocore & Farias, 2006; Jensen-Kondering et al., 2012; Niskanen et al., 2012; Ocklenburg, Hugdahl & Westerhausen, 2013; Ramsey et al., 2001; Vikingstad et al., 2000; Zaca, Jarso & Pillai, 2013). Studies included in the forest plots produced here report a wide spread of LI values for verbal fluency tasks, ranging from 0.05 to 0.94 (see Fig. 2).

A number of studies have compared the use of different verbal fluency paradigms. When a frontal region of interest (ROI) is used, semantic fluency is reported as less strongly lateralising than verb generation or phonemic fluency (Kleinhans et al., 2008; Ruff et al., 2008; Sanjuan et al., 2010a). Conversely, two studies using combined frontal and temporoparietal ROIs reported no differences in their strength of laterality (Rutten et al., 2002; Tailby et al., 2014). This suggests that these tasks may differ in the extent of lateralisation they induce across different language areas (see following section, effect of region of interest). Interestingly, multiple studies report that LIs from generation tasks can vary substantially depending on methodological choices made when calculating laterality, such as the threshold chosen (Dodoo-Schittko et al., 2012), the use of normalisation, smoothing and clustering techniques (Baciu et al., 2005), and the activity measure used (Harrington, Buonocore & Farias, 2006).

Effect of region of interest

Verbal fluency tasks tend to induce the strongest laterality in frontal ROIs (Gaillard et al., 2003; Niskanen et al., 2012; Ocklenburg, Hugdahl & Westerhausen, 2013; Partovi et al., 2012a; Partovi et al., 2012b; Propper et al., 2010; Szaflarski et al., 2008; Vernooij et al., 2007; Vikingstad et al., 2000; Vingerhoets et al., 2013; Zaca, Jarso & Pillai, 2013). Although they can induce strong laterality in temporoparietal ROIs (Harrington, Buonocore & Farias, 2006; Jensen-Kondering et al., 2012; Stippich et al., 2003), this may not be significantly greater than other tasks within this ROI (Zaca, Jarso & Pillai, 2013). However, this may depend on the particular fluency task used; Jensen-Kondering et al. (2012) reported that while phonemic fluency and verb generation yielded the strongest lateralisation for a frontal ROI, the strongest laterality for a temporoparietal ROI was seen with semantic fluency, consistent with a role for such areas in semantic cognition.

Effect of baseline task

Although the majority of studies using verbal fluency tasks employed a passive baseline task such as fixation, a number used active baselines such as finger tapping or silent word repetition (see Fig. 2 and Table 2). Dodoo-Schittko et al. (2012) reported that an active baseline task which required subvocal manipulation of the order of syllables within a pseudoword yielded significantly stronger laterality for a verb generation task compared to the use of a passive resting baseline. This is consistent with the idea of subtraction of bilateral activity related to speech-motor planning (Poeppel, 2014; Price, 2012b).

LI strength, reliability and robustness

Mean LIs reported from sentence generation studies are illustrated in Fig. 3. High mean LIs of between 0.74 and 0.89 have been reported for sentence generation, both when sentences are pre-learnt prior to scanning (e.g., Stippich et al., 2003), and when they are actively generated during the scan (e.g., Tzourio-Mazoyer et al., 2016). However, other studies have reported more modest laterality estimates of between 0.48 and 0.65, again with both variants of the task (Mazoyer et al., 2014; Partovi et al., 2012a; Partovi et al., 2012b). Thus, it does not appear to be the case that strength of laterality differs according to whether sentences are generated spontaneously during the scanning session, or learnt prior to scanning. Two studies within our search measured laterality for semantic fluency and sentence generation within the same participants (Partovi et al., 2012a; Stippich et al., 2003); however, these studies reported differences in strength of laterality between the tasks in different directions. Further, they used the version of the sentence generation task in which sentences are learnt prior to scanning, so task demands were not well matched. Thus, there is currently insufficient data with these tasks to evaluate predictions of stronger laterality for sentence over word processing. Partovi et al. (2012a) report good reproducibility for sentence generation, however their analysis simply looked at significant differences in group means over repeated testing, and not at reproducibility of individual participant’s LIs.

Effect of ROI

There is mixed evidence as to whether sentence generation yields differences in the laterality measured from frontal versus temporoparietal ROIs. While Partovi et al. (2012a), Partovi et al. (2012b) and Stippich et al. (2003) reported equivalent strength of laterality across both, Tzourio-Mazoyer et al. (2016) found significantly stronger laterality in frontal than temporal areas. Interestingly, in contrast to the exclusively right handed samples of the former studies, this latter study used a mixed handedness sample, with an overrepresentation of left handers. This suggests the possibility that greater regional heterogeneity may characterise the atypical profiles of language lateralisation that are more often found within atypical handedness samples.

Effect of baseline

When sentences are learnt prior to scanning, studies generally employ simple cross-fixation or rest as a baseline (Partovi et al., 2012a; Partovi et al., 2012b; Stippich et al., 2003). Conversely, the two studies using spontaneous generation of sentences during the scanning session used an active linguistic baseline, in which participants covertly generated the months of the year (Mazoyer et al., 2014,Tzourio-Mazoyer et al., 2016). A comparison of these baselines in terms of the language processes isolated by each contrast is given in Table 3. As can be seen, the active baseline subtracts out activity related to speech motor planning to leave those processes specific to the construction of novel sentences, such as syntactic and lexico-semantic processing; conversely, the contrast with rest results in poor isolation of such language processes. This highlights the need to consider carefully the functions one wishes to isolate when choosing a suitable baseline, and the implications this will have for interpretation of measured laterality in relation to linguistic processes.

LI strength, reliability and robustness

The majority of studies using passive listening tasks reported very weak average LIs (see Fig. 4 for mean LI values), indicating bilateral activation; indeed, passive listening is often the most weakly lateralising task when compared to others (Binder et al., 2008; Harrington, Buonocore & Farias, 2006; Miro et al., 2014; Ocklenburg, Hugdahl & Westerhausen, 2013; Tzourio-Mazoyer et al., 2015). A notable exception to this was presented by Thivard et al. (2005) who reported a mean laterality index of 0.72 within a frontal ROI for a passive story listening task, stronger than that seen in this ROI for a semantic fluency task (0.51). We also note that high test–retest correlations have been reported for speech listening (Razafimandimby et al., 2007).

Effect of ROI

Studies are inconsistent as to whether stronger laterality is found for speech listening within a frontal or a temporoparietal ROI. Harrington, Buonocore & Farias (2006) reported that temporoparietal LIs were stronger and more reliable than frontal LIs, whereas other studies have reported weaker and more variable LIs for a temporal compared to a frontal ROI (Miro et al., 2014; Ocklenburg, Hugdahl & Westerhausen, 2013; Thivard et al., 2005). In general, posterior language areas appear to be poorly lateralised for receptive speech listening tasks, although this may depend on the baseline task employed (see paragraph below).

Effect of baseline task

The varying levels of asymmetry reported for speech listening tasks may in part be attributable to the baseline used by different studies. Binder et al. (2008) found that changing the baseline from rest to tone listening raised the average LI for word listening from 0.1 to 0.52. In this regard it is interesting that the two studies reporting near-zero average LI values for speech listening employed rest as a baseline task (Miro et al., 2014; Ocklenburg, Hugdahl & Westerhausen, 2013). Conversely, Thivard et al. (2005) and Harrington, Buonocore & Farias (2006) both used backwards speech listening as a baseline and reported stronger laterality measurements for speech listening. This effect of baseline is consistent with the idea of bilateral early auditory processing that must be subtracted out by a non-linguistic auditory stimulus in order to reveal asymmetry for higher-level ‘central language processes’ (Peelle, 2012; Poeppel, 2014; Price, 2012).

LI strength and effect of ROI

Our search identified only two papers investigating lateralisation of text reading. Both studies used the same covert (silent) text reading task with a baseline of covert reading of text composed of pronounceable non-words. Backes et al. (2005) reported moderately strong laterality (LI = 0.59) using a combined frontal-temporoparietal ROI, whereas (Deblaere et al., 2002) reported weak laterality (LI = 0.21) using a global LI (see Fig. 4). This supports the hypothesis that global LIs are generally weaker than regional LIs.

LI strength, reliability and robustness

Phonemic judgement tasks yield relatively strong laterality, with reported LI values ranging from 0.41 to 0.84 (see Fig. 3). However, when compared to other tasks, phonemic judgement is often reported as more weakly lateralising (Baciu et al., 2005; Niskanen et al., 2012; Seghier et al., 2004). Phonemic judgement may be superior to other tasks however in terms of robustness and reproducibility. Morrison et al. (2016) reported that a rhyming decision task demonstrated greater reliability than a word generation task, yielding reproducible dominance classifications in 100% of participants, and average test-retest correlations for LI values of 0.9 and above. Furthermore, such reproducibility of LIs obtained with rhyming decision was more robust against changes in the activity measure used for LI calculation.

Effect of ROI

Rhyming decision tasks yield particularly strong laterality when a frontal or a combined frontal-temporoparietal ROI is used (Baciu et al., 2005; Clements et al., 2006; Cousin et al., 2007; Niskanen et al., 2012). For example, Cousin et al. (2007) identified a particularly strong leftward asymmetry for the inferior frontal gyrus during rhyme detection. Thus, frontal ROIs may be optimal for yielding the strongest laterality with this task.

Effect of baseline task

All studies within our search using phonemic judgement were found to employ an active perceptual decision baseline task on either non-linguistic material (e.g., line orientation matching) or nonsense words or characters (e.g., nonsense word font matching). This subtracts out non-linguistic working memory processes (see Table 2), as well as basic visual processing.

LI strength, reproducibility and robustness

The strength of laterality reported for semantic decision tasks is quite variable, ranging from near-zero to around 0.8 (see Fig. 4 for mean LI values). This may depend on the type of semantic decision required. Tasks which require judgement of the semantic relatedness of two words appear to yield relatively strong laterality, ranging from 0.59 to 0.84 (Bethmann et al., 2007; Fernandez et al., 2001; Häberling, Steinemann & Corballis, 2016; Seghier et al., 2004). In contrast, category membership tasks with single words appear to give much lower LIs, ranging from 0.03 to 0.52 (Deblaere et al., 2002; Hund-Georgiadis et al., 2002; Hund-Georgiadis, Lex & von Cramon, 2001; Ramsey et al., 2001; Van Oers et al., 2010). This suggests that it may be the process of integrating and comparing across semantic representations for different concepts that is strongly lateralising; conversely a simple lexical look-up to determine the category membership of a single concept may not be strongly lateralised.

Jansen et al. (2006) reported very low reproducibility for synonym decision LIs across a range of different LI calculation methods within Broca’s area; much higher reproducibility was found however when a temporoparietal ROI was used. Conversely, Harrington, Buonocore & Farias (2006) reported high test-retest correlations for an abstract/concrete semantic decision task across both frontal (IFG) and temporoparietal ROIs. This discrepancy between the two studies could be due to the differences in the tasks they used.

Effect of ROI

The majority of studies report no significant differences in the magnitude of laterality found within temporoparietal and frontal ROIs for semantic decision tasks (Bethmann et al., 2007; Häberling, Steinemann & Corballis, 2016; Harrington, Buonocore & Farias, 2006; Hund-Georgiadis et al., 2002; Ramsey et al., 2001; Van Oers et al., 2010). Some studies have reported differences across ROIs, however these can be in opposite directions (Fernandez et al., 2001; Szaflarski et al., 2008). As discussed, some evidence suggests that LIs calculated from temporoparietal ROIs for semantic decision may be more reproducible than those calculated from frontal ROIs (Jansen et al., 2006).

Effect of baseline tasks

Semantic decision laterality is also strongly influenced by the baseline task used. Binder et al. (2008), Hund-Georgiadis et al. (2002) and Hund-Georgiadis, Lex & Von Cramon (2001) manipulated the baseline and found that the use of an active perceptual decision task as opposed to passive rest yielded a large increase in the strength and consistency of semantic decision laterality. Binder et al. (2008) argued that resting baselines are unsuitable for subtraction with semantic decision, since they allow for the activation of conceptual language representations as the participant ‘day dreams’ and engages in ‘inner speech’. An active perceptual decision baseline interrupts such ongoing conceptual processing and engages the same executive and attentional processes as the language paradigm. This subtraction is shown in Table 4, which illustrates the better isolation of semantic processes that this baseline provides compared to the contrast with rest. Baseline tasks that engage linguistic processing themselves may result in reduced laterality; for example, Deblaere et al. (2002) suggested their finding of weak laterality for semantic decision may have been due to a vowel decision baseline task. However, it should be noted that Binder et al. (2008) reported identical laterality strength (a mean LI of 0.62) for semantic decision using either a baseline of tone decision or phoneme decision. Overall, this evidence suggests that baseline tasks used for semantic decision must be active, sufficiently engaging and challenging so as to prevent ‘day-dreaming’, and ideally involve material from a non-linguistic domain e.g., symbols or tones.

Sentence comprehension

Sentence comprehension tasks require some judgement about the content of a spoken or written sentence. Syntactic and semantic processing are often confounded (see Table 2); for example, the task may require participants to decide if two sentences with different grammatical constructions have the same meaning. However, they are noteworthy among other tasks in the extent of their syntactic processing requirements. Laterality of syntax has been a subject of debate, with some authors arguing for a bilateral involvement in syntax (e.g., Hund-Georgiadis, Lex & Von Cramon, 2001), but others arguing for a left dominance (Friederici, 2011; Tyler et al., 2011; Wright, Stamatakis & Tyler, 2012). At a more general level, multiple models would predict left lateralisation for the sentence-level processing engaged by this task, without making specific claims about lateralisation of syntactic processing per se (e.g., Peelle, 2012; Poeppel, 2014; Price, 2012).

LI strength, reliability and robustness

Multiple studies report strong laterality for sentence comprehension tasks (Harrington, Buonocore & Farias, 2006; Jensen-Kondering et al., 2012; Kennan et al., 2002; Niskanen et al., 2012; Sanjuan et al., 2010b; Vassal et al., 2016), with LI values ranging from 0.55 to 0.88 (see Fig. 3). Studies which compare sentence comprehension laterality measures to those of other tasks suggest that it can outperform semantic decision, phoneme decision, story listening and naming tasks in terms of the strength of laterality, although this can depend on the ROI (Harrington, Buonocore & Farias, 2006; Niskanen et al., 2012).

Effect of ROI

Evidence appears inconsistent as to the effect of ROI on the laterality obtained with sentence comprehension tasks. Studies have reported both stronger laterality for frontal than temporoparietal ROIs (Jensen-Kondering et al., 2012; Niskanen et al., 2012) and vice versa (Harrington, Buonocore & Farias, 2006; Sanjuan et al., 2010b). In terms of reliability of lateralisation, Harrington, Buonocore & Farias (2006) reported very high reproducibility for a visual sentence comprehension task across both frontal and temporoparietal ROIs, with test-retest correlations above 0.9. In contrast, auditory sentence comprehension yielded more reliable lateralisation within a temporoparietal than a frontal ROI. Modality of the stimuli may thus affect which ROI is optimal.

Effect of baseline task

Generally, active baselines are employed for sentence comprehension tasks. A notable exception is seen in Harrington, Buonocore & Farias (2006) who used passive listening to backwards speech as a baseline for auditory sentence comprehension. This passive baseline may explain the weaker laterality they reported as compared to other studies (mean LI of around 0.45). Interestingly, Sanjuan et al. (2010a); Sanjuan et al. (2010b) who also reported a relatively low level of laterality compared to other studies used phoneme decision as a baseline. As previously discussed in relation to a study by Deblaere et al. (2002) using semantic decision, it is possible that the use of such a baseline with high linguistic processing demands may lower the strength of the laterality seen.

LI strength, reliability and robustness

Studies using naming as a language activation task report a wide variety of LI values, ranging from 0.08 to 0.96 (see Fig. 3). This can partly be explained by variation in the nature of the naming task used. Zero to moderate laterality has been reported for picture naming tasks, which are often the least lateralising when compared to other tasks (Deblaere et al., 2002; Harrington, Buonocore & Farias, 2006; Jansen et al., 2006; Van Oers et al., 2010; Vikingstad et al., 2000). However, the lateralising ability of naming tasks can be increased by the addition of sentence comprehension demands. A naming from written or auditory description task known as ‘responsive naming’ requires comprehension of a question describing an object in order to generate the required name. This has been reported to yield strong laterality across both frontal and temporal ROIs, in the range of 0.65 to 0.96 (Gaillard et al., 2002; Niskanen et al., 2012). This increase in laterality with the addition of sentence-level processing is consistent with models of language which predict an increase in laterality for connected versus unconnected language i.e., structured sentences versus single words (Peelle, 2012; Poeppel, 2014).

Picture naming also shows poor reliability in laterality measurement. Jansen et al. (2006) reported that picture naming did not determine dominance reproducibly in about a third of participants. Rutten et al. (2002) similarly reported a failure to find significant test-retest correlations for naming LIs. Significant test-retest correlations were reported by Harrington, Buonocore & Farias (2006) for a picture naming task at around the same level as those seen for semantic decision; however, reproducibility of naming laterality was lower than that seen for verb generation and sentence comprehension.

Effect of ROI

Naming tasks do not appear to favour one ROI over another in laterality measurement. We found two studies which reported differences in the laterality measured from frontal and temporoparietal ROIs, however this difference was in opposite directions (Harrington, Buonocore & Farias, 2006; Brennan et al., 2007). The majority of studies instead report highly similar strength of laterality across frontal and temporoparietal ROIs (Gaillard et al., 2002; Niskanen et al., 2012; Van Oers et al., 2010; Vikingstad et al., 2000). Furthermore, Rutten et al. (2002) reported similar levels of reproducibility of naming LIs for both regions.

Effect of baseline task

The baselines used with picture naming provide interesting evidence on the processes underlying its laterality. Deblaere et al. (2002) reported near zero laterality for picture naming using a baseline which required participants to name the position of the intersection of four lines (e.g., up, down, left, right). This task involves engagement in similar semantic and word retrieval processes (see Table 2), which may explain the weak laterality measured. However, Brennan et al. (2007) reported strong laterality for a picture naming task with a number counting baseline. This would subtract out speech production and word retrieval processes for an automated speech sequence, predicted to involve bilateral activity (Price, 2012b; Poeppel, 2014). This subtraction would thus isolate spontaneous non-automated retrieval and word generation processes which may engage left hemisphere language systems, increasing measured laterality.