That kind of work was very poor methodologically and its results are best explained as the Clever Hans effect, as the animals generating random strings unconnected to a current context, or simply as over-zealous efforts to assign meaning to the animals' signing, by their handlers.
See for example this article regarding the work of Francine Patterson with Koko:
On the Evidence for Linguistic Abilities in Signing Apes
https://www.sciencedirect.com/science/article/pii/0093934X79...
From which I quote (since it's Elsevier-walled):
>> One of the major problems with Patterson's study is that it presents very little data. While several large corpora of the utterances of children have been published (e.g. Bloom, 1973), there is as yet no corpus of any ape's signing behavior. That is, Patterson fails to provide a substantial number of transcribed utterances. Rather, she relies upon individual examples to support her interpretations. In the absence of a corpus of utterances, however, these examples are impossible to interpret. One cannot determine whether they have the functions wich Patterson attributes to them, or whether they resulted from the ape acting as a "random sign generator" which happened to emit sequences that could be selectively chosen to illustrate particular points. This sampling problem vitiates Patterson's claim that certain combinations demonstrated that Koko possessed the ability to creatively combine signs into novel utterances. Those who assert that apes have shown linguistic abilities have invariably relied upon examples such as Washoe's signing water bird for duck. In the absence of a large corpus, however, these examples are subject to multiple interpretations. Patterson's claim that cookie rock was a creative description of a stale sweet roll loses much of its force if Koko also produced utterances such as cookie tickle, cookie hat, and toothbrush cookie in similar contexts. The largest corpus of utterances from any signing ape, that of Terrace, Pettito and Bever (1967a,b) shows that their subject, a chimpanzee named Nim Chimpsky, did in fact combine each vocabulary sign with a large number of other signs. Although each of the resulting combinations could be interpreted metaphorically, a simpler interpretation is that he merely combined signs randomly. The correct interpretation depends on other information--an accounting of the frequencies with which signs occurred in combination with one another, the contexts in which combinations occurred, the content of the teachers' signing--which Patterson fails to provide. Without this information, the importance of her examples cannot be ascertained.
>> Patterson's discussion of her cookie rock example provides some measure of this problem. She states, "Although Koko has produced uninterpretable strings (as do some children), most of her utterances are appropriate to the situation and some are strikingly apt" (p. 88). She then cites some "interpretable" examples, including cookie rock; the "uninterpretable" strings are not described. It is the case that only "interpretable" sequences are ever documented in the reports on ape signing. Only by presenting an unedited corpus of responses, however, could Patterson's assertion be validated.
Note also that one of the two authors in the quoted paper is Laura A. Pettito, one of the researchers that worked with Nim Chimpsky.