Our data set consists of twelve musicians, who follow higher music education in Belgium and agreed to participate in an eye-tracking experiment. Except one, all study at LUCA School of Arts and were selected on the basis of their musical abilities as judged by the chamber music coordinator of the institution. One participant was found through social media and studies at the Royal Conservatory of Brussels. Four clarinet-violin-piano trios were formed making sure that no musician had ever played chamber music with any of the partners before. Two of the trios were all female, one included a male musician, and one included two male musicians. Ages ranged from 18–28 years (Mean age = 23 years). None of the musicians had ever played the composition chosen for the recording session. They consented in writing to taking part in the study and to the use of still images and audiovisual recordings for scientific purposes. At the end of the session each musician received a voucher.

The recording session took place in the concert hall of the institution. The musicians were positioned the way they would be in a natural condition (Fig. 1), i.e. the clarinetist stood inside the wing of the piano, while the violinist stood next to where the pianist was seated. Clarinetist and violinist were faced toward each other and the height of their music stands was lowered just enough so that the participants could see each other’s head. The distance between the musicians was such that gazes at different large areas of the body could be distinguished (i.e., head, torso, legs, arms…). Smaller parts (for instance, eyes and mouth) could not be detected separately.

Each musician in the trio wore a binocular mobile eye tracker (Tobii Pro Glasses 2, sampling rate 50 Hz). For those who normally wore prescription glasses the eye tracker was fitted with lenses with approximately the same strength as the participants’ own. Two external cameras (frame rate 50 fps) captured the overall interaction. One was positioned in the seating area of the hall, filming a frontal view of the trio; the other filmed the back of the trio from the balcony above the stage. An audio recorder (TASCAM DR-2d) was placed on one of the front seats and guaranteed a reasonable sound quality.

The musical excerpt was taken from the last movement of Milhaud’s Suite for violin, clarinet and piano (measure 1 to 103 of the Vif-section). The musical parts carried a metronome marking of 120 bpm. At this speed the excerpt lasts about 2 minutes in performance. The marking could inform the participants about the envisaged performance tempo, however the researchers gave no instructions as to what tempo was expected. The musicians were also told not to use a metronome. The music was deemed appropriate for the study of individual differences and interactional dynamics, as the instruments are treated more or less as equal partners through an almost equal share in the melody and through passages that combine the melody with countermelodic material, rather than accompaniment patterns.

Upon arrival at the concert hall, the participants were briefly introduced to each other. The researcher explained the schedule for the session, handed over the musical parts and guided the clarinetist and violinist to an individual practice room. The pianist remained in the concert hall. The musicians were allowed to practice for half an hour, after which they were assembled for the eye-tracked rehearsal. The rehearsal followed a pre-determined schedule that alternated between uninterrupted runs through the musical fragment and rehearsal periods during which the participants were expected to work collaboratively on the fragment. The schedule was organised as follows: rehearsal period 1 (30’) – run-through – rehearsal period 2 (30’) – run-through – run-through. Except for the individual practice, the entire session was recorded with mobile eye trackers, cameras, and audio device. The eye trackers were recalibrated before each run-through. At the very end, each participant filled out a post-performance questionnaire. On the one hand, the questionnaire obtained information about the participants’ experience of the equipment and procedure. On the other, it aimed to collect possible data points for analysis by enquiring about the difficulties in the musical excerpt and by asking where in the score participants thought they had looked at a partner. Gaze will be analysed in the light of these responses at a later stage of our research. It is well worth noting that all participants, with one exception, considered the amount of individual practice time either adequate or too long. Regarding the amount of rehearsal time, participants stated either that no additional time was needed or that additional rehearsing on another day could be useful if they were to study the full musical piece. It therefore seems that the musical excerpt was not too difficult for the participants. The questions were formulated in Dutch and English. However, quite a few participants were not native English or Dutch speakers. Hence, after participants completed the form a brief one-to-one discussion followed with the researcher, mostly to ensure that the questions and answers were clear to both participant and researcher.

Prior to annotation, the gaze data of each trio member were exported as video files. These were synchronised with each other, with one of the external camera recordings and the audio recording in Adobe Premier Pro (but only the eye-tracked data and audio recording are of importance for the current publication). Synchronisation was enabled through the claps that were executed at the beginning and end of each run-through and rehearsal period. In the resulting quadvid, the audio of the eye-tracking videos and the external camera recording was disabled, leaving only the sound from the audio recorder. The synchronised data were exported at 25 fps and imported into the editing tool ELAN (42) to be annotated manually. The procedure thus far followed that of researchers in conversation analysis (see for instance (19, 20)).

Whenever the gaze cursor approached one of the partners, the annotator checked the location of the cursor frame by frame in order to determine the start and end of a partner-gaze. Partner-gazes were annotated as such when the gaze cursor fell onto the partner, including (parts of) the instrument, i.e. the clarinet, the violin and bow, and the keys of the piano. The annotated data set did not include moments where the gaze cursor fell near the partner. We will investigate these moments separately (in a later stage of our research), since these moments were clearly marked by a gaze shift toward the partner and therefore could be relevant for the study of gaze in ensemble interaction. Equally excluded from the annotated data set were 20 instances where it was unclear whether the gaze cursor pointed toward the partner due to overlap. These cases concerned some of the violinists. Due to the particular posture of violinists (who hold the instrument at the left side of the body) and their particular position in the trio (at the right side of the pianist), it was at times hard to distinguish gazes at the scroll of the violin and the violinist’s left hand from gazes at the pianist, who was seated ‘behind’ the scroll and left hand.

Musical bars were annotated manually on an additional tier in ELAN by listening to the audio recording. This included checking each bar in an initial annotation in order to eliminate traces of sound that belonged to the previous bar. Since ensemble playing always involves some asynchrony, this means that in our annotations the new bar started once all three musicians had arrived there. We note that asynchronies were overall hard to detect by ear so, for our research purposes, this procedure seemed adequate.

After analysis of the score, score characteristics were annotated on additional tiers, using the audio recordings, in a similar fashion as described above. The score analysis identified structural features (e.g. section endings, phrase endings, and smaller phrase segments), role allocation (indicating which instrument held the melody, countermelody, or accompaniment), role switches (moments where role allocation changed), rests, entrances and exits.

Gazing at the partner in a trio constellation can occur in six directions, in this particular study between violin and clarinet, between clarinet and piano, and between violin and piano, each time in two directions. As regards the number of partner-gazes in all four run-throughs in total (Fig. 2), gazes occurred in both directions between violin and clarinet in all trios. However, in trios 1 and 4 the violinist looked more often at the clarinetist than vice versa. In trios 2 and 3, the clarinetist looked more often at the violinist than vice versa. Regardless of the specific gaze direction, in each trio the highest amount of partner-gazes happened between the violinist and the clarinetist.

Much less gazing at the partner can be seen in the interactions that involve the pianist. In trio 3, no one looked at the pianist and the pianist did not look at anyone. To be sure that, indeed, there was no visual interaction at all with the pianist in this trio, we checked whether gazes near (as opposed to on) the partner occurred. This was not the case. In trio 1, the clarinetist only looked five times at the pianist across all run-throughs. The pianist, again, did not look at anyone (nor gaze near the partner). There were no gazes at the pianist by the violinist (although overlaps occurred four times). In trios 2 and 4, gazes happened in all six directions. In both trios, gazes between violinist and pianist occurred only sporadically. Compared to that, the pianist received more than sporadic visual attention from the clarinetist in trio 2. In trio 4, both pianist and clarinetist looked more than sporadically at each other.

When we view each run-through separately, we see that gazing at the partner occurred least often in the first run-through (Fig. 3, 4, 5, 6). In terms of the amount of gaze directions that were represented in this first run-through, most trios (1, 2 and 4) reveal that gazes occurred in fewer directions than in the other run-throughs. Also, pianists never looked at anyone during the first run-through. As is the case with the amount of gaze directions, the first run-through tended to contain a lesser amount of partner-gazes than the other run-throughs. More specifically, this was the case in trio 1 (in all directions), in trio 2 (in all directions, except from violin to clarinet), in trio 3 (except from violin to clarinet), and in trio 4 (in all directions).

While still attending to each gaze direction within each trio, we consider the possibility of a tendency across all four run-throughs. Some instances of the last run-through contain the highest number of partner-gazes, namely in trio 1 (from violin to clarinet), trio 2 (from clarinet to violin), and trio 4 (from violin to clarinet). A steady increase of partner-gazes across run-throughs can be observed in trio 1 (from violin to clarinet) and – if one considers the total amount of gazes by a musician to both partners – in trio 2 (by the clarinetist and by the pianist). Finally, there are no musicians, whose gazing at the partner decreases across run-throughs and there are no musicians either, whose last run-through reveals the lowest number of partner-gazes.

As regards the amount of partner-gazes for all run-throughs in total, it was found that most visual interaction occurred between clarinetists and violinists, some between clarinetists and pianists, and no or only sporadic visual interaction between violinists and pianists. A possible explanation could lie in the particular musical excerpt used. Although the three instrumentalists overall are treated as equal partners (through an almost equal share in playing the melody and through passages that are polyphonic in nature), exchanges of the melody happen at a quicker pace between the violinist and clarinetist than with the pianist. According to our analysis, the violinist and clarinetist take over the melody each 12 times, whereas the pianist does so only 4 times but for a longer stretch of time.

Another plausible interpretation relates to the (natural) set-up of the musicians. Violinists and clarinetists only have to look up from the score in order to see each other, whereas a slight turn of the head to the right is needed for the clarinetists and pianists, and a turn of the body for the violinists and pianists. The same remark applies to the study by King and Ginsborg (26), where (relatively) little observable gazing at the partner was found. In their study a head turn was necessary for the singers and pianists to see each other within central vision, similarly to the clarinet-piano interaction in our study. While a common explanation for both studies might be that pianists do not often look at their partners, the set-up was insufficiently accounted for in order to enable a meaningful interpretation of the frequency of partner-gazes.

While the musical fragment and the set-up in our study may have caused differences between instrumentalists within the trio constellation, a comparison across trios seems to defy attempts at generalisation, confirming our expectation that differences in gaze behaviour would occur regardless of the specific instrument played. Between clarinetists and violinists, both the scenario of clarinetists looking more often at violinists (Trios 2 and 3) and the opposite scenario (Trios 1 and 4) occurred. In addition, the absolute frequency of gazing at the partner differed widely: In trio 2 the violinist only looked 10 times at the clarinetist, whereas the violinist in trio 4 did so 125 times. Also, while in some trios gaze occurred in all six directions (Trios 2 and 4), in other trios certain gaze directions were not represented in any of the four run-throughs. In extreme cases, the pianist was not looked at by anyone and/or did not look at anyone. Individuals thus differed to the extent that they looked at both partners, only looked at one partner and ignored another, or never even used (foveal) gaze as a means of communication.

As for the pianists’ infrequent looking at the partner, the need for seeing the keys may have to be taken into account. However, the pianists reported only few technical difficulties that required close visual attention. Pianists 1 and 3 reported a jump and a glissando. Pianist 2 indicated only the glissando and pianist 4 did not report any technical difficulties. Pianist 3 pointed out additional difficulties that required looking at the keyboard in two passages, each six bars long. We also note that they may have experienced a higher need for reading the score than the other musicians because of having to read two staves, but our self-reports did not cover this issue.

A substantial difference as to how often musicians looked at their partner can also be seen in Biasutti et al. (5). In daily conversations, too, participants’ amount of gazing at the partner may differ substantially. Specifically, Kendon (25) found that the amount of time spent gazing at the partner varied from 28% to over 70% of the total duration of the analysed samples. We note that, in our data set, the total amount of time spent looking at the partner was much less, as can be expected since the musicians also had a reading task to fulfill, and varied between 0% and approximately 15% of the entire duration of a run-through. As for the musicians who did not look at all at their partner (pianists 1 and 3), their situation mirrors the experimental conditions in Keller and Appel (24) and Kawase (23), whereby musicians could not see each other. The former study, using a fragment without metrical changes, found no remarkable effect of visual conditions on ensemble coordination, while in Kawase’s study there was an effect on the coordination of tempo changes. In line with Keller and Appel’s (24) study, our musical fragment did not contain tempo changes or changes in metrical pulse. Consequently, it may be that pianists 1 and 3 did not consider gazing at the partner, at least via central vision, necessary for the sake of temporal coordination. In fact, when listening to the audio, no disturbing asynchronies could be heard in any of their trio’s performances (as was the case for all trios). Thus, the musical sounds themselves may have provided them with the necessary cues to synchronise. Evidence from a study by Vera et al. (40) could support such an argument. While other studies have mentioned synchronisation and musical coordination as a possible role of gazing at the partner (11, 12, 21, 32, 41), this functionality has to be considered with respect to the musical characteristics and to competing coordination strategies that individuals may draw on to different extents.

We also looked at the distribution of partner-gazes across run-throughs to see whether a tendency could be observed. In the first run-through, partner-gazes were fewer and represented less gaze directions than in the other run-throughs. Since at this moment the musicians were still unfamiliar with each other and with the overall sound of the musical fragment, familiarity may well be a good explanation for the frequency of partner-gaze. Once the newness of the music and the social pressure to perform well in front of unfamiliar partners have been overcome, musicians are better able to let go of the notes on the score. Also, additional rehearsing after individual practice may have contributed to a diminished/diminishing need for close note-to-note reading in run-throughs 2, 3, and 4. Alternatively, the first run-through was not an ‘official’ one in the sense that it was not requested by the researcher. As it was an inherent part of the rehearsal period, there may have been less pressure on the participants to perform well, causing a difference in their gaze behaviour. Last, we note that the first run-through was played slower than the other run-throughs by trios 2 and 3, causing the conditions in which gaze occurred to be slightly different in the first run-through for those trios. (Specifically, all trios played run-throughs 2, 3 and 4 at performance tempo, with the total playing time varying between 1’45” and 1’55”. Trios 1 and 4 stayed within this range for the first run-through, while in trios 2 and 3 the total playing time was 2’51” and 2’20” respectively.)

A tendency across all run-throughs was otherwise hard to find. The last run-through was found to contain the most partner-gazes in the case of a few musicians, while sometimes the amount of partner-gazes also increased across all run-throughs. The opposite – the last run-through containing the least partner-gazes or a decrease across run-throughs – was not found. Although this is little evidence for a general tendency, this somewhat confirms Williamon and Davidson’s (41) suggestion that the increase of eye-contact at important momments for coordination should not be explained solely by a tendency for gaze to increasingly function as a coordinating device. By contrast, in our pilot study (39), no tendency for partner-gaze to occur more frequently could be found in any of the three duos that were analysed. This may be due to the fact that the four run-throughs in the pilot study took place on two different days and constituted ‘snapshots’ in the middle of a rehearsal process, which was initiated by the musicians themselves before they were asked to take part in the study. The tendencies found in the current study may thus be typical for a very first rehearsal when performers do not know each other and try out a new piece for the first time.

The authors declare that the contents of the article are in agreement with the ethics described in http://biblio.unibe.ch/portale/elibrary/BOP/jemr/ethics.html and that there is no conflict of interest regarding the publication of this paper.