Nature’s Words

Rube Goldberg machines excel at producing very long events, all part of a single causal chain. Like most events, Rube Goldberg events are built mostly out of hits, slides, and rings. Again letting b , s , and a stand for hits, slides, and rings, Rube Goldberg events sound something like basabababasababababababasabababasa , although the chains are very often much longer than even this. If events were typically like Rube Goldberg events, then even if spoken words have many of the auditory features found in events, words would be much too short to be event‑like. Events are, however, not typically Rube Goldberg‑like. Events are, instead, much more typically like a pen thrown on a table, the generic event we discussed in the previous section. Pen‑on‑table events may consist of a hit, hit, and slide. Or possibly just a hit and a slide. Or even just a lone hit. Most events have just several physical interactions or fewer, much nearer in length to spoken words than to Rube Goldberg events.

This is what nature‑harnessing expects. Spoken words across human languages are not only built out of sounds like those in solid‑object physical events, but words tend to have the size of typical physical events. Words tend to sound like events with up to several interaction sounds–plosives or fricatives–not, say, ten. And although words with a single interaction sound are allowed, two or three interaction sounds are more common, again like solid‑object physical events.

Words are not only approximately the size of solid‑object physical events–i.e., having several interaction sounds–words also take the amount of time for a typical event. This is something I have thus far ignored. But notice that plosives, fricatives, and rings do not just have similar acoustic characteristics to hits, slides, and rings; they also occur over periods of time similar to those typical of hits, slides, and rings. For example, although I described both hits and plosives as nearly instantaneous explosions, the notion of “instantaneous” depends on the time scale relevant to the listener–what’s instantaneous to a human may not be instantaneous to a fly. Hits and plosives are both instantaneous explosions as heard by human ears. This is why plosives sound hitlike; for example, if a hitlike sound were stretched out it would, instead, sound more slidelike (something we discussed in the earlier section called “Hesitant Hits”). Similarly, fricatives and sonorants tend to occur over time scales similar to the slides and rings of physical events. Typical syllables of human speech–e.g., of the form ba or sa –tend to have a duration approximately on the order of tenths of seconds, roughly the same time scale as is common for physical events involving macroscopic objects. In fact, you’ll notice in Figure 4 earlier that the physical and linguistic analogs (e.g., a hit and “k”) are on the same scale for the time (x) axis.

Words tend to be built out of the constituents of natural solid‑object physical events, and to have approximately the size and temporal duration of such events. But are words actually structured like solid‑object physical events? Are the natural‑sounding phonemes and syllables put together into natural‑sounding words? In particular, I’m interested in asking whether the sequences of physical interactions that occur in events–the hits and slides–are similar to the sequences of plosives and fricatives in words. My students and I analyzed the “event structure” of common words across 18 languages, and for each language we measured the distribution of six event types: hit (b ), slide (s ), hit‑hit (bb ), hit‑slide (bs ), slide‑hit (sb ), and slide‑slide (ss ). For example, “tea” is a b , “far” is an s , and “faker” is an sb .

Figure 11 . The freqency of the structure types found in words across 18 widely diverse languages (listed in the legend of Figure 9). (Standard error bars shown. See Appendix for details.)

To estimate how common these simple event types are in nature, students Elizabeth Counterman, Kyle McDonald, and Romann Weber counted the kinds of events occurring in a wide variety of videos. In deciding upon the kinds of videos to sample, we were not especially interested in having videos of, say, the savanna. Recall our discussion in the previous chapter, where we observed that there are “hard cores” of nature likely found in most or all habitats with solid objects crashing about. In choosing twenty videos from which to enumerate solid‑object physical events, we simply aimed for a variety of scenarios in which solid‑object physical events occur, including cooking, children playing, family gatherings, assembly instructions, and acrobatics. Each student acquired data on the events occurring, and did so using only the visual modality (that is, the videos were on mute); this helped to deal with a worry that our auditory systems are biased by speech so that we hear speechlike structure in events (akin to seeing faces in clouds). The three observers identified an average of 650 events across the 20 videos. Figure 12 shows the average results for the videos as a dotted line, overlaid on the language data from Figure 11. One can see the close similarity in the plots. (Notice that a simple model assuming hits are more common than slides does not explain why bs occurs more often than sb in the language data.)

Figure 12 . The relative frequency of simple event types in videos and in language. One can see their considerable similarity. (Standard error bars shown. See Appendix for details.)

Again, we find the signature of solid‑object physical events–of nature–in spoken language! Our final story in this chapter on speech concerns the sounds of speech above the level of words: the structure of whole phrases and sentences.