The associative system of early-learned Hebrew verbs and body parts: a comparative study with American English

2.1 Participants

Fifty one native speakers of Hebrew (34 female, 17 male), classified as mid-high SES, were tested at Ono Academic College, Kiryat Ono, Tel Aviv area, Israel. The American English data reported in Maouene et al. (2008) was collected from 50 native speakers of English (half female, half male) who were traditional-aged undergraduates tested in a large Midwestern college town in the United States. The Hebrew speakers tested were older than the English-speaking participants because many students in Israel perform military service before college.

2.2 Stimuli

The verbs in this study are the 103 early-learned verbs listed on the Hebrew Communicative Developmental Inventory (HCDI) (Maital and Dromi 1997; Maital et al. 2000), given in Table 2. The HCDI, an adaptation of the English MCDI (Fenson et al. 1994), is based on a stratified sample of 253 Israeli mothers reporting on their toddlers aged 12–24-months (Maital et al. 2000). Associative data for the 101 early-learned verbs listed on the English MCDI (Table 2) was previously collected and reported (Maouene et al. 2008). The English version is based on a sample of 1,803 children from 16 to 30 months as per parental report.

To establish translational equivalency between the verbs, we used a modified version of the backward-forward translation method proposed by Brislin (1970) and expanded by Wang et al. (2015). For the forward translation, we asked two native Hebrew speakers with native-like proficiency in English to translate the verbs on the Hebrew CDI into English. For the backward translation, we concomitantly asked two different translators who were native English speakers with native-like proficiency in Hebrew to translate a list of English translations of the Hebrew verbs, based on the English CDI, which was provided by the third author.

In total, the Hebrew and English CDIs had 81 pairs of verbs judged as translational equivalents. Of the remaining verbs, 22 were semantically unique to Hebrew and 20 were semantically unique to English. As a final check, the four translators were asked to provide quality and difficulty rating scores for each translation based on two Likert scales. For each verb, we calculated an average among the raters and then averaged all the averages. For the forward translation (Hebrew to English), on average, 89.3% of the quality of the translational equivalents (TE) found was rated as high to very high (Mean: 4.7 where 5 = high quality), and 74.8% of the TE were found easily to very easily (Mean: 1.5, where 1 = very easy). In the backward translation (English to Hebrew), on average, 93.2% the quality of the TE found was rated as high to very high (Mean: 4.8) and 88.4 of the TE were found easily to very easily (Mean: 1.4).

2.3 Procedure

Hebrew-speaking participants were tested as a group in a computer lab. Participants were given a list of 103 verbs in a random order and asked to write down the body-part term they associated with each verb. Participants were naïve to the goals of the study and free to supply any body part at any level of scale (e.g., fingernails, arms, whole-body). They were provided with the following instructions: את/ה מתבקש/ת לכתוב עבור כל פועל לאיזה איבר גוף הוא מקושר אצלך.

At/a mitbakesh/et lixtov avur kol po’al le-eyze eyvar guf hu mekushar eclex/a ‘Please write next to each verb the body part with which it is associated for you.’ This follows the method used to collect data in Chinese and other languages, where participants were asked to write down their answers.

The English data (Maouene et al. 2008) was collected slightly differently, with participants requested to orally supply the first body part that came to mind in response to each verb, with the data recorded by the experimenter. Subjects heard the following instructions: You will hear a list of verbs. Please tell me the first body part that comes to mind. English-speaking adults were asked to provide oral responses so that this method could also be used with children for a subsequent study (Nesheim et al. 2012).

Hebrew speakers interpreted the instructions as that they could provide multiple parts, and 14% of the responses provided included more than one body part. In that case, we selected the first body part provided. By this method, participants offered 73 unique body parts in total, compared with 61 parts in English. To check whether more parts were provided in Hebrew due to the instructions, we went back through the 86% of responses that were single body-part terms and found all 73 parts provided in the single responses. Thus, we are confident that the ambiguity in the Hebrew instructions did not influence the higher number of body parts in Hebrew.

2.4 Reporting body-part terms

Synonyms (e.g., stomach, tummy, belly) and nested parts (e.g., lip, mouth, head) were counted as unique body-part terms but instances of singular and plural forms of the same noun (e.g., yad ‘hand’, yadayim ‘hands’) were counted as the same term. All body-part terms provided by the participants are reported in the singular form. We use italics for the Hebrew and English responses, with the English gloss in single quotes (e.g., sexel ‘mind’/mind). The body regions are in all capitals (e.g., ecba ‘finger’ and kaf yad ‘palm’ both belong to the YAD ‘HAND’ region). The verbs used and all the body parts provided are given in the Supplementary Materials I and II, with Hebrew script listed for Hebrew terms.

3.1 Hebrew and English convergence relations: entropy and correspondence analyses

3.1.2 Correspondence analysis

Next, we used a correspondence analysis (CA) to compare the dimensions accounting for the overall patterns, the overall shape, the similarity of the verbs with respect to their body-part associations and the similarity of the shared verbs relative to the shared body parts in the two associative systems. The CA is a dimension-reduction technique appropriate to categorical data based on positive and negative correlations and does not pre-determine categories based on body morphology or other types of body-part organization. The data in a high-dimensional set is reduced into a lower-dimensional description while maintaining the variance structure among individual instances. Note that correlations are sensitive to infrequent body parts such that a very infrequent response (e.g., kaved ‘liver’/liver for lishtot ‘drink’/drink in both languages) will be highly salient and load strongly on a dimension clustering verb belonging to a different region (in this example, around pe ‘mouth’/mouth). Thus, in the CA, there is room for divergence at the level of very unique responses.

The CA was performed on a contingency matrix of size m × n, where m is the number of rows (verbs) and n is the number of columns (body parts) for each language. Specifically, the matrix for Hebrew consisted of 103 verbs by 73 body parts and the matrix for English (Maouene et al. 2008) consisted of 101 verbs by 61 body parts. Both matrices were composed of “meaning” vectors created for each verb, comprising the number of participants listing each body part as associated with that verb. For example, the vector for the Hebrew verb lehavi ‘bring’ has the values [36-yad ‘hand’, 8-regel ‘leg’, 2-kaf yad ‘palm’, 1-pe ‘mouth’, 1-moax ‘brain’, 1-rosh ‘head’, 1-gapa ‘limb’] and the vector for the English verb bring has the values [38-hand, 9-arm, 2-finger, 1-brain].

3.1.2.1 Dimensions

In dimension reduction, the CA analyses indicate that the first four dimensions in each language accounted for 41% of the cumulative variance in Hebrew and 34.7% in English (Table 4); 8 dimensions accounted for 66% of the cumulative variance in Hebrew and 60% in English; and 20 dimensions accounted for 89.2% of the cumulative variance in Hebrew and 91.6% in English. This shows that a small and identical number of dimensions in both associative systems account for the majority of the variance. The higher variance in Hebrew is explained by the Hebrew speakers associating more unique body parts with certain verbs (body parts that are mentioned only once, e.g., shablul ‘cochlea’ with lishmo’a ‘hear’). However, it is important to note that English speakers had stronger agreements with one another and thus higher correlational patterns across the dimensions.

Table 4:

The correlations and cumulative variance for the first four dimensions of the CA.

Correlations			Cumulative variance
Dimension		English	Hebrew	English	Hebrew
1	0.9427		0.8389	0.0997	0.1137
2	0.8576		0.7995	0.1904	0.222
3	0.7627		0.7259	0.2711	0.3204
4	0.7209		0.6499	0.3473	0.4084

3.1.2.2 Overall shape structures

Next, we examined the overall shape of the Hebrew and English structures. The CA enables a set of data to be displayed in three-dimensional graphical form by decomposing the chi-squared statistics associated with the contingency matrix into orthogonal factors based on the chi-square distance. Points further away from the centroid (0, 0, 0) contribute more to the dimension whereas points on the centroid contribute very little or nothing. The shape of the CAs exhibited by the first four dimensions in both Hebrew and English present highly similar 3-arm shapes with well-delineated clusters, shown in the 3D space in Figure 1a and b.

Figure 1:

A 3D overall shape structure in (a) Hebrew and (b) English verb–body-parts associations. Dimensions 2–4 are shown. Scores along each dimension are standard deviations.

These 3-arm structures in both languages represent clusters of verbs organized by mouth-related parts; hand- and leg-related parts; and eye-related parts. Each dot represents a verb, and closer distance between dots represents greater agreement in body-part association. Verbs in the same cluster have more similar body-part associations, with verbs higher in the clusters being more strongly associated with that body part. Verbs with different patterns of associations appear in different clusters and dots between clusters indicate associations with two or more body regions.

3.1.2.3 Verb similarity structure within each language

Next, we examined the similarity structure of the verbs with respect to their body-part associations within each language. Each CA points to mainly convergence between the two languages, where the first four dimensions present a coherent and overall shared structure of verbs falling into four clusters. This data compression strongly suggests that within each CA, the two corpora of verbs are each systematically related to a small set of common body parts. The Hebrew semantic space organized by dimensions 2–4 with the 103 verbs identified is shown in Figure 2 (shown from a different perspective than Figure 1a, as these are dynamic views). For the 3D English semantic space, see Maouene et al. (2008: 1205).

Figure 2:

The coordinate space derived from the CA for Hebrew with the complete list of the 103 verbs. Dimensions 2–4 are shown. Scores along each dimension are standard deviations. (For a corresponding figure which includes the complete list of the 101 verbs that make up the English semantic space, see Maouene et al. 2008: 1205).

To make the structure revealed by the Hebrew CA in Figures 1a and 2 more intuitive, we describe some of the verbs and body parts associated with each cluster. The cluster that displays the highest specificity in pairing between body parts and verbs is organized by ozen ‘ear’ and shablul ‘cochlea’, driven by just two verbs, lehakshiv ‘listen’ and lishmo’a ‘hear’ (first dimension, not shown here.) The top-most cluster in Figures 1a and 2 contains verbs associated with eye and some eye parts. Verbs higher in the cluster are more strongly associated by participants with ayin ‘eye’ (e.g., lehabit ‘look’, lir’ot ‘see’). This is different from the English cluster, where a subcluster of head, brain and heart was loading on that dimension too. The cluster on the right side of Figures 1a and 2 shows the verbs associated mostly with legs and hands: the right-most verbs (e.g., lashevet ‘sit’, liv’ot ‘kick’) are more strongly associated with leg parts (e.g., regel ‘leg’, yashvan ‘bottom’); the verbs toward the center in the middle (e.g., letapes ‘climb’, lipol ‘fall’) are less strongly associated with leg parts and have additional associations with hand/arm parts (e.g., yad ‘hand’, katef ‘shoulder’); and going further into the center, the verbs in the mass (e.g., lisxot ‘swim’, la’avod ‘work’) are associated with hand parts (e.g., yad ‘hand’), a subcluster of which are verbs associated with rosh ‘head’, moax ‘brain’, sexel ‘mind’ (e.g., limco ‘find’, lexapes ‘search’) and with lev ‘heart’ (e.g., lisno ‘hate’, le’ehov ‘love’). Finally, the bottom left-most cluster in Figures 1a and 2 consists of verbs (e.g., lelakek ‘lick’, lit’om ‘taste’) strongly associated with mostly mouth parts (e.g., safa ‘lip’, lashon ‘tongue’).

The English CA (Maouene et al. 2008: 1205) also shows a 3-arm structure. In Figure 1b, the cluster on the left side shows verbs (e.g., lick, taste) associated with mouth parts. The middle cluster consists of verbs associated with both hand and leg parts, with the verbs at the top showing the highest correlations with leg parts, the verbs in the middle (e.g., climb, drive) having associations with both hand and leg parts, and the verbs at the lowest part of the cluster (e.g., sweep, throw, hit, knock) associated with hand parts (e.g., hand, arm, fist, knuckle). Finally, the cluster on the right side shows verbs strongly associated with eye (e.g., see, look, cry) and those nearer the center are associated with eye, brain and mind (e.g., think, pretend, wish) and with heart and mind (e.g., love, hate, like).

The similar structures found in the Hebrew and English CAs, as with the earlier similar entropy results, suggest that the principles driving these overall commonalities may be shared between the two associative systems. Note that in both systems, it is not just proximity of localization that determines these clusters, as body parts belonging to the same region can load onto different dimensions (e.g., zeret ‘little finger’ is clustered with ayin ‘eye’ for the verb likro ‘read’ in Hebrew; mouth and kidney are clustered with drink in English). Although we discussed the results of the four dimensions separately, in fact in both languages, they interact and combine in many semantic relationships, including fine-motor versus gross-motor activities (e.g., mouth and eye vs. hand and leg) and goal-directed functionality versus spatial exploration (e.g., hand and mouth vs. eye, hand and leg), among others. This suggests a multifaceted and distributed model of verbs organized by body parts.

3.1.2.4 Shared similarity structures

In the previous section, the CA analyses showed that the verbs clustered by the body parts partitioned the two semantic spaces in similar ways. To bolster this finding and visualize whether the four dimensions align in the two systems, in Figure 3 we show a new CA graph using just the 81 translational equivalent verbs and the 42 shared body parts in Hebrew and English. We see again a 3-arm structure. In this solution, the left cluster is driven by mouth-, lip- and tongue-related verbs; the middle cluster is driven by eye-related verbs; and the right cluster shows verbs associated with leg parts. The two ear-related verbs are included in this space but are not easily visible from the perspective shown (they form the first dimension).

Figure 3:

The 81 translational equivalents clustered by the 42 shared body parts. The Hebrew verbs are shown in green and the English verbs are shown in red. The coordinate space is derived from the CA organized by dimensions 2–4. Scores along each dimension are standard deviations.

Next, we ran a correlational analysis on the 81 shared verbs and the 42 shared body parts vectors. The bivariate correlations for dimensions 1 through 4 yield strong r values that range from 0.79 to 0.99, shown in Table 5.

Table 5:

Bivariate correlations for dimensions 1 through 4.

	English
		Dim 1	Dim 2	Dim 3	Dim 4
Hebrew	Dim 1	0.99a
	Dim 2		0.91a
	Dim 3			0.79a
	Dim 4				0.80a

1. ^aThese correlations are significant at p < 0001.

To further support our claim that verbs overall can be clustered similarly by the shared body parts in both languages, we used a different clustering method, k-means, which is based on the distances to the centroid of the CA from the shared verbs and body parts. Five successive analyses were run, with k = 1, 2, 3, 4, 5. In Figure 4 below, a 3D plot with k = 4 shows a clear 4-arm structure. For both languages, cluster 1 (black and red dots) includes verbs related mostly to mouth parts; cluster 2 (black and red upward triangles) includes verbs mostly related to eye; cluster 3 (black and red squares) includes verbs mostly related to leg parts; and cluster 4 (downward triangles) includes verbs mostly related to hand parts. In this solution, leg-related verbs become separated from hand-related verbs, and head, mind, brain, heart are clustered with hand. Interestingly, k = 5 shows a subcluster of verbs related to hand separate from cluster 4. Most of the verbs in cluster 5 have a distribution that span multiple body regions or have whole-body in the vectors, a point we will return to in the next section. Of note, this algorithm does not output the ear-related verbs as a separate dimension.

Figure 4:

The 3 dimensional CA space of 81 translational equivalents clustered by the 42 shared body parts using the k-means method, with k = 4. The English verbs are in black and the Hebrew verbs are in red. Different clusters are indicated with different markers (circles, squares, upward triangles and downward triangles). In these sets of coordinates, the scores along each dimension are normalized to have the unit measure equal one standard deviation.

Finally, five verbs that differ between Hebrew and English in their main body region association (e.g., feed associated with mouth in English but yad ‘hand’ in Hebrew) appear in different clusters in Figure 4. We address these verbs in Section 3.2.2, which looks specifically at the divergence between the two systems.

To provide supportive statistical evidence, we then ran a series of three chi-square tests on the independence of the k by k cluster membership. We counted the pairs of verbs (e.g., lixtov ‘write’ in Hebrew, write in English) in the same cluster versus distributed independently (the null hypothesis). The reasoning was that if the cluster membership structures for the English and Hebrew verbs were independent, the joint probability distribution would follow the multiplication of the marginal probabilities of English and Hebrew verbs in the clusters. We tested this null hypothesis of English-Hebrew independence with the formula χ 2 = ∑ i , j n 2 ( p i , j − p i p j ) 2 n p i p j which follows the chi-square distribution of the degree of freedom ( k − 1 ) 2 . In all four cases, the null hypothesis was rejected by significantly small p values (p < 0.01). These results suggest that the cluster membership structure of Hebrew and English verbs was tightly correlated. Table 6 reports an example of the structure of these multiple comparisons.

Table 6:

Cluster membership of English and Hebrew verb pairs for k = 4. For the 81 translational equivalent verbs, the number of verb pairs in which the English verb is in cluster i and the Hebrew verb is in cluster j.

English-Hebrew	Hebrew in cluster 1	Hebrew in cluster 2	Hebrew in cluster 3	Hebrew in cluster 4
English in cluster 1	13	0	0	0
English in cluster 2	0	5	1	0
English in cluster 3	0	0	11	1
English in cluster 4	2	0	0	48

In conclusion, the four measures used to examine the verbs and body parts common to both languages (CA, bivariate correlations, k-means clustering and the chi-square tests of independence) show mainly convergence and suggest that the principles driving these overall commonalities may be shared between the two associative systems. Divergence in the verbs and body parts is a point we examine in detail in Section 3.3.2.

3.2 Central body regions

Here we take a more detailed and descriptive look at which verb is associated with which central body region as identified by frequency of mention. We examine the fit of the Hebrew and English verbs to these central regions, examine whether verbs with similar meanings cluster similarly, and report exceptions.

3.3 The body parts

Hereafter we focus on the latent structure created by the body parts. We discuss the congruence and divergence in the body parts, examined with respect to most-frequent and least-frequent terms, and show body maps for easier visualization. The body maps are constructed based on the number of mentions of each body part in the total data set, such that the size of a body-part term is proportionally larger when mentioned by a greater number of participants. Details on the construction of these body maps are given in Supplementary Material IV.

3.3.1 Most-frequent body-part terms

The 14 most-frequent body parts in Hebrew and the 16 in English (those mentioned 1% or more in each data set and comprising the central body regions in Table 7) account for the majority of responses in both languages (Hebrew: 92.3%; English: 93.7%). These most-frequent body-part terms are given in Table 8 with ranking and frequency of occurrence.

Table 8:

Ranking of the most-frequent body-part terms (those mentioned ≧1% of the total number of responses) in Hebrew and in English, by frequency.

Hebrew				English
Rank	Body part		Frequency	Rank	Body part	Frequency
1	יד yad	‘hand’	40.4%	1	hand	32.7%
2	רגל regel	‘leg’	14.1%	2	arm	9.4%
3	פה pe	‘mouth’	8.9%	3	eye	8.7%
4	עין ayin	‘eye’	5.7%	4	leg	8.5%
5	ראש rosh	‘head’	4.8%	5	mouth	8.2%
6	מוח moax	‘brain’	3.8%	6	foot	6.6%
7	לב lev	‘heart’	3.6%	7	finger	3.0%
8	אצבע ecba	‘finger’	1.9%	8	bottom	2.9%
9	אוזן ozen	‘ear’	1.8%	9	whole-body	2.3%
10	ישבן yashvan	‘bottom’	1.8%	10	mind	2.3%
11	כל הגוף kol ha-guf	‘whole-body’	1.4%	11	ear	2.1%
12	שפה safa	‘lip’	1.4%	12	head	1.8%
13	לשון lashon	‘tongue’	1.4%	13	tongue	1.7%
14a	כף יד kaf yad	‘palm’	1.3%	14	heart	1.3%
				15	lip	1.1%
				16b	brain	1.1%

1. ^aGav ‘back’ is the 15th most-frequent term in Hebrew, with 0.9% of the total responses. ^bTooth is the 17th most-frequent term in English, with 0.75% of the total responses. The sum of the percentage of mention of hand plus arm in English is 42.1% and the sum of yad ‘hand’ pluszro'a ‘arm’ (+0.04%) in Hebrew is 40.44%. Similarly, the sum of leg plus foot in English is 15.1% and the sum of regel ‘leg’ plus kaf regel ‘foot’ (+0.02%) in Hebrew is 14.13%. Both of these show high levels of convergence.

Except for three body-part terms (zro’a ‘upper arm’ and kaf regel ‘foot’ in Hebrew; palm in English), the most-frequent terms were translational equivalents, albeit with different rankings. The majority of the most-frequent body parts fall into the same three body areas: upper-limbs (yad ‘hand’/hand, ecba ‘finger’/finger; Hebrew: kaf yad ‘palm’; English: arm); lower-limbs (regel ‘leg’/leg; yashvan ‘bottom’/bottom; English: foot); and head (ayin ‘eye’/eye, pe ‘mouth’/mouth, ozen ‘ear’/ear, safa ‘lip’/lip, lashon ‘tongue’/tongue, moax ‘brain’/brain, rosh ‘head’/head; English: mind). The last two terms are lev ‘heart’/heart and kol ha-guf ‘whole-body’/whole-body. Visually, in both systems, these most-frequent terms span one large-size surface (whole-body), most medium-size surfaces (leg, arm, head), some small-size surfaces (eye, finger, lips, tongue, etc.) and internal organs (brain, mind, heart).

The top most-frequently mentioned body-part term in both languages was yad ‘hand’/hand, accounting for 40% of all responses in Hebrew and 33% of all responses in English and was given more than twice as often as the next most-frequent response in each language. Although the list of most-frequent parts in the two languages has significant overlap, there are also clear differences, starting with the second most-frequent part (Hebrew: regel ‘leg’, 14%; English: arm, 9.4%). Additionally, ayin ‘eye’/eye and kol ha-guf ‘whole-body’/whole-body occur more often in English; moax ‘brain’/brain and lev ‘heart’/heart more often in Hebrew; sexel ‘mind’/mind is most frequently mentioned in English but doesn’t make the list in Hebrew; and kaf yad ‘palm’/palm is most frequently mentioned in Hebrew but below the threshold in English.

There are further differences in how Hebrew and English speakers talk about the limbs, following a common divergence in languages around the world (e.g., approximately one-third of the world’s languages studied do not have separate terms for hand and arm, Brown 2005). English has separate terms for arm and hand and for leg and foot. By contrast, Hebrew speakers use yad ‘hand’ to also refer to the whole arm. Additionally, Hebrew speakers use two additional terms for the upper limbs: zro’a ‘upper arm’ which refers to the area between the shoulder and the elbow and ama ‘forearm’ which refers to the area between the elbow and the wrist. For the lower-limbs, Hebrew speakers gave regel ‘leg’ which refers to the area from the hip joint to the foot and includes the foot, and kaf regel ‘foot’ which refers to the foot only, like leg and foot in English. Our data suggests that the two languages linguistically carve up the limbs in different ways, which may reflect differences in how actions are conceptualized: whereas hand, arm, leg, and foot occur on the most-frequent list in English, in Hebrew, only yad ‘hand’ and regel ‘leg’ are most frequently associated with our list of early-learned action verbs; zro’a ‘upper arm’, ama ‘forearm’ and kaf regel ‘foot’ instead appear on the least-frequent parts list.

The high degree of congruence in the most-frequent Hebrew and English body parts is more easily visualized with body maps, shown in Figure 5, using the percentages of mention given in Table 8. These maps are constructed based solely on the number of mentions of each body part in the total data set—that is, the depicted size of the body-part term is proportionally larger when a greater number of participants have mentioned it. We followed morphological localization in upper-middle-lower body regions as an important organizational principle, but many other principles are also at work here, including function, shape, perspective, naming and emotions. We tried to render that complexity: The maps display both continuities and discontinuities with multi-perspective overlays of internal and external parts, front and back parts, and first and second order perspectives (e.g., bite [one’s] lip vs. bite [another’s] ear), along with some continuity (torso, whole-body). The maps were drawn by Yangfei Song, a bilingual Chinese-English speaker who does not know Hebrew, in close collaboration with the first authors. The artist was directed to map size onto frequency of mention (Table 8) by using one body part (brain) as the standard unit of measurement to draw the other body parts in the homunculus. The original size of the brain was 3 cm by 1.25 cm. The full homunculus was resized to create Figure 5a. The homunculi in Figures 6 and 7 (shown proportionally to each other in 5b and 5c, respectively, and proportional to 5a) were drawn following a similar process. Details on the construction of all the homunculi are given in Supplementary Material IV. Note that our body maps do not reflect any judgment by our speakers as to the exact boundaries of the terms. Rather, they are meant to provide quick and direct visualization of the proportionate size of the body-part terms based on the number of mentions, as an illustration of the complexity of the associations.

Figure 5:

Body maps depicting body parts associated with early-learned verbs in Hebrew (always shown on the left) and English (always shown on the right). All maps are shown here at scale with respect to each other. Figure 5a shows body maps depicting the most-frequent body parts, representing 92.3 and 93.7% of the data in Hebrew and English, respectively. Least-frequent body parts are shown in two maps, Figure 5b and c, which are shown much larger in Figures 6 and 7, respectively. Figure 5b shows body maps depicting the non-integrative least-frequent body parts, representing 7 and 6.2% of the data in Hebrew and English, respectively. Figure 5c shows body maps depicting the integrative least-frequent parts, representing 0.6 and 0.04% of the data in Hebrew and English, respectively. See Supplementary Material IV for details on how these body maps were constructed.

Figure 6:

Enlarged body maps depicting the non-integrative least-frequent pairings of verbs and body parts in Hebrew (left) and English (right). Figure 6 shows the same maps as Figure 5b, larger here for easier viewing. These maps are proportional with respect to each other but not shown to scale with respect to Figure 5. These figures represent 7 and 6.2% of the data in Hebrew and English, respectively. See Supplementary Material IV for details on how these body maps were constructed.

Figure 7:

Enlarged body maps depicting the integrative body parts in Hebrew (left) and English (right). Figure 7 shows the same maps as Figure 5c, here larger for easier viewing. These maps are proportional with respect to each other but not shown to scale with respect to Figure 5. These figures represent 0.6 and 0.04% of the data in Hebrew and English, respectively. See Supplementary Material IV for details on how these body maps were constructed.

4.1 Convergence

Three results characterize large similarities in the clustering of verbs by body parts in the Hebrew and English associative systems: 1) similar internal organizations, as operationalized by comparable levels of entropy; 2) similarities in clustering for the first four dimensions organizing action verbs, contributing to shape similarities of the overall patterns of associations; and 3) strong agreement for the most-frequent body parts and the central body regions categorizing verbs.

These converging results may be due to similar organizing principles. One important principle comes from the organization of the somatopic maps in the neural organization of adults, where body parts are well-represented components of multiple distributed systems in the cortical and subcortical areas. Somatopic maps are reported in the sensory-motor strips (Marshall et al. 1937; Penfield and Boldrey 1937), in the occipitotemporal cortex for visual representation (Orlov et al. 2010) and in the cerebellum (Holmes 1918). These cortical and subcortical regions are connected with our own actions or the actions of others. From a developmental neural perspective, continuity in the somatosensory map has been reported, together with some developmental remapping in infants (Rigato et al. 2014). Further, recent fMRI evidence in newborn macaques shows that the topographic sensorimotor maps are present at birth and that these body maps are mostly indistinguishable from those in older macaques. Further, the embodied perspectives in cognitive neuroscience claim that the circuits used for functions such as planning, sensing, calibrating and selecting an action (in the supplementary sensorimotor area, SMA and premotor cortex, PMC) are either simulated for semantic grounding (Barsalou 1999; Gallese and Lakoff 2005; Glenberg and Kashak 2002) or are related to the word form that becomes associated with association regions related to perception and action through Hebbian learning (Pulvermüller 2005). Our data does not support the opposite perspective, which separates the body from language and takes the view of a wired brain that takes in information from culture and the environment independently of spatial or temporal activation (for a review and discussion of the many embodied and disembodied accounts of semantics, see Meteyard et al. 2012). By this view, less convergence between verbs and body parts should be found across different languages and cultures which is not what we find in our data.

A second organizing principle that contributes to convergence is latent semantics and distributional cues. Studies suggest a connection between early-learned and concrete words and later-learned abstract words, where early-learned concrete words derive from neural activity in domain-specific sensorimotor brain regions (Hazy et al. 2020). As later-learned verbs enter the semantic system, they form new connections with these early pathways, including figurative and more abstract connections. Additionally, later-learned abstract terms may be partially learned through linguistic cues, including distributional statistics, such as lexical co-occurrences (Hazy et al. 2020) and social and inner experiences (Borghi and Binkofski 2014). For example, the body and its parts are involved in primary metaphors, which emerge from early embodied experiences and are often shared cross-culturally; an illustration of this is the early-learned metaphor happy is up, sad is down, which correlates with a child energetically exploring the world while running upright and feeling happy versus lying down when tired or unhappy (Lakoff 2012). Such distributional cues establish important pathways and early neurological wiring during the sensitive period in the experience-dependent brain. Primary metaphors continue to be active in adult metaphoric associations and are building blocks for complex and general conceptual metaphors where the body and its parts can be understood as containers, objects, conduits, etc. (Lakoff and Johnson 1980). Body parts continue to be found overtly in adult figurative expressions and associations (e.g., help is at hand, don’t mind me, I put my foot in my mouth, I heart ice cream). For example, in our data and all the languages studied to date (see Table 1), the prevalence of the hand response in adults could stem from the early concrete association between verbs and body parts in daily experiences, as well as from more abstract and figurative associations resulting from distributional learning. If developmental continuity is at play, we predict we should see similar internal organization and shape similarities in the overall patterns of associations in later-learned and abstract verbs (e.g., imagine, judge, manipulate, gamble). Additionally, while we would continue to see many verbs associated with hand, we would see an increased importance of more abstract body-part terms, such as mind, eye and heart, and potentially more verbs would span multiple body regions.

In addition to neural-developmental-semantic principles, linguistic and perceptual categorization also contribute to the convergence we find. Cognitively, the interaction between categorization and naming would predict maximal similarities around mid-level body-part terms, as those would be accessed faster in our associative task (Rosch et al. 1976; Tversky and Hemenway 1984). In fact, in our data most of the terms provided are at the mid-level of generality, with very few detailed (e.g., tastebud, vocal cord, abs) or very general (e.g., genital, upper-limb, lower-body) terms. Additionally, perceptual categorization, in particular visual, also pushes for convergence. In our data, the most-frequently named parts are mostly surface ones: one large-size surface (whole-body), a range of medium-size surfaces (e.g., leg, head, hand, eye, finger), and very few smaller surface parts or visual details (nails, but no blemishes, pimples, wrinkles, cuticles, etc.). The upper and lower limits of categorization may be due to a confluence of influences, such as labels of the first body parts learned, importance of surface similarities in children, frequency of usage and attention to certain parts, similar cultural frames of attention where some details are focused on more than others, comparable proprioceptive awareness, and sparse time resources in experimental tasks such as ours.

Another organizing principle driving convergence in verb–body-part associations may be having a shared concept of the body organized in parts (Goddard and Wierzbiecka 2016; Wierzbicka 2007). The convergence in our data may be due to Hebrew and English speakers using the label whole-body and conceiving of the body in terms of a mereology, or part-whole relationships (e.g., a fingernail is part of a finger, which is part of a hand). However, note that talking about the whole-body and conceiving of it as a collection of part-whole relationships does not hold in all languages and cultures, as some languages refer to the body as ‘person’, and a body part is a ‘thing that is located’, such that a nose is not part of the head, but a thing located on the head (Goddard and Wierzbiecka 2016; Wierzbicka 2007). Additionally, some cultures delineate limbs at different points on a body map or do not have terms for limbs (e.g., Majid 2010), suggesting that part-whole relationships may not be a primary means of discussing the body in such languages, and specific cultures may differ in the labeling, levels and relationships between parts, or whether they even think of the body as comprising parts (see Enfield et al. 2006).

4.2 Divergence

Although it is striking to find large convergence in the two adult associative systems, we also see noteworthy differences in the body-part terms and low frequency verb–body-parts pairings. Below we discuss 1) the divergence in the number and type of body-part terms provided and 2) the most significant differences in the body-part terms and verb pairings.

First, we find that Hebrew speakers mention more body parts per verb than English speakers overall, and the responses in Hebrew include a larger proportion of internal and integrating body-part terms. These differences between the two associative systems may stem from individual differences within the sample, such as personality (e.g., being detail-oriented or wanting to stand out), particular experiences (e.g., recent surgery or health-related issues) and tolerance for experimental tasks. Divergence may also arise from cross-cultural differences in knowledge of the body, comfort-levels in talking about certain body parts (e.g., genitals), historical traditions and cultural practices shaping somatic awareness and interoceptive sensitivity and accuracy (Ma-Kellams 2014) to internal organs. Finally, preferences for abstraction may also account for the divergence in the integrative body-part terms (e.g., lower-body, genitals, skin), reflecting small but important linguistic differences in using more superordinate terms referring to general body regions.

Some significant and interesting divergences are found in specific body parts and their verb pairings. The YAD ‘HAND’/HAND region in the two languages shows significant differences, particularly based on cultural variations in everyday actions and coded gestures, for example, in dancing (e.g., with the belly, holding hands in circle, focused on the legs), eating (e.g., with three fingers, with the whole hand, with utensils) and praying (e.g., palms joined, bowing, kneeling on the floor). Kaf ha-yad ‘palm’ is especially salient in Hebrew, occurring in the most-frequent list and associated with 39 verbs, compared with palm in English which only occurs with 2 verbs (push, splash). Among Hebrew speakers, there may be a heightened and shared awareness of kaf yad ‘palm’, which is talked about in religious, sacred practices (e.g., blessing, praying) and might increase its salience even in very mundane activities (e.g., knocking, pushing, putting, writing, brushing, cutting).

We also find differences for terms for genitals, with Hebrew speakers mentioning more types of sexual parts and to a greater proportion than do English speakers. These differences may be due to factors including differences in age, cultural taboos or a greater comfort with talking about sex and sexual parts among Hebrew speakers and individual differences (e.g., humor and fatigue, among others). Although speakers of both languages mention pin ‘penis’/penis, Hebrew speakers mention more types of female parts, perhaps due to 67% of the participants being female. The female genital terms provided are pot ‘vulva’, dagdegan ‘clitoris’ and rexem ‘womb’ and the integrating term ever min ‘genital’ which refers to both male and female parts. The terms pot ‘vulva’ and ever min ‘genital’ are used with lidfok (lidpok) ‘knock’ which in slang means ‘shag, have sex.’ Pot ‘vulva’ is also associated with laredet ‘get down, go down’, lelakek ‘lick’, lirkav ‘ride’, lidxof ‘push’ and ligmor ‘finish’, as well as with lexasot ‘cover’ and lishtof ‘wash’. In sexual parts provided in common between the two languages, we find differences in verb associations, for shad ‘breast’/breast (Hebrew: litpos ‘catch’, le’ehov ‘love’; English: buy) and pin ‘penis’/penis (ligmor ‘finish’/finish in both languages, but also in Hebrew, lidfok (lidpok) ‘knock’, lasim ‘put’ and lexabot ‘turn off’). English speakers but not Hebrew mention nipple (associated with feed).

Another difference we find is that Hebrew, but not English, speakers use integrating terms referring to the immaterial parts neshima ‘breath/breathing’ (associated with le’ehov ‘love’, laruc ‘run’, lashir ‘sing’, la’acor ‘stop’) and neshama ‘soul’ (associated with lacet ‘go out’, la’azor ‘help’); both terms share the same consonantal root n-sh-m. These associations suggest a dynamic interaction in Hebrew between physicality (the respiratory system) and cultural, religious and linguistic conventions. The importance of breath and breathing is found in Hebrew in expressions such as kax neshima amuka ‘take a deep breath, relax’. Neshama ‘soul’ also has figurative meaning, such as referring to a person as good-hearted and to express extra familiarity in order to ask for a favor, or conversely, to refer to a person as bad.

Finally, we also see cultural divergence in the verb–body-part associations involving more internal, less observable states and activities, including cognition, emotion and spirituality, which may reflect differences in how these two cultures conceptualize and talk about these. It is important to note that verbs of cognition and emotion in both languages tend to be learned later as compared with other verbs (English: Fenson et al. 1994; Hebrew: Maital and Dromi 1997; Maital et al. 2000). Three body parts associated with verbs of cognition in the two languages occur with different frequency rankings: moax ‘brain’/brain and rosh ‘head’/head are among the most-frequently mentioned terms in both languages, but with a higher ranking in Hebrew; and mind is in the most-frequent list for English only (with sexel ‘mind’ occurring in the least-frequent list). Within the set of verbs of cognition, only limco ‘find’/find and ligmor ‘finish’/finish are included on both early-learned verb lists. In Hebrew, these verbs are more strongly associated with the ROSH ‘HEAD’ region, particularly moax ‘brain’, sexel ‘mind’ and rosh ‘head’, whereas in English, these cognitive verbs are also associated with eye, hand and other parts. The other verbs of cognition differ between the two languages and also differ in the associated parts, with livxor ‘choose’ and lexpes ‘search, look for’ associated primarily with rosh ‘head’ in Hebrew and pretend, think and wish associated predominantly with mind in English, suggesting a more perceptual-emotional dimension of cognition in English. The two cultures also show divergence in some emotional appraisal verbs. Lisno ‘hate’/hate and lexabev ‘like’/like are associated with rosh ‘head’ in Hebrew, suggesting a cognitive component to the emotion, whereas they are associated with mind and heart in English, suggesting more emotion and feeling. However, note that le’ehov ‘love’/love is associated with lev ‘heart’/heart in both languages (plus mind in English only), suggesting both Semitic and Indo-European cultures have somewhat similar conventions regarding “love” (for a review, see Vanhove 2008), though “hate” and “like” may be understood differently. These patterns of associations suggest that the concepts of sexel ‘mind’/mind and rosh ‘head’/head are quite different in each language. Other studies have also suggested more marked cross-linguistic differences for abstract terms (e.g., Goddard 2010).

4.3 Limitations

One of the well-known limits of the associative method is that it cannot by itself disambiguate between overlapping meanings that have different sources (literal, vicarious, metaphoric, etc.). Future work will need to address the complexity of associative interpretations experimentally and directly compare the proportion of concrete uses of body parts as compared with symbolic, figurative uses. Additionally, the associative method used here might not provide the full range of possibilities found in all languages and may need to be modified to incorporate culture-specific notions and terms regarding the body, for instance using ideas of personhood rather than body, and thing rather than part (Wierzbicka 2007).

It could also be that, in some cultures, the term “body part” is more or less restricted in terms of what it encompasses (e.g., excludes internal organs, bones, or very small parts; see Enfield et al. 2006). The divergences we found in our data may be due to the Hebrew speakers providing more body parts, particularly internal organs and detailed parts, than the English speakers. Comparison between languages that provide similar numbers of parts may result in less divergence. To explore cultural divergence more fully, it is important in future work to ask for more than just one body part, to add organs or bones in the definition of what a body part can refer to, and to use other detailed measures of the real world, such as pictures and questions about function, localization and categorization. Future work should also examine languages from cultures that are significantly more different, as the overall patterns we found here may not always hold.