Patent law is full of rules that often seem somewhat contradictory. For example, we all know that claims are interpreted in light of the specification, but should not import limitations therefrom. This post addresses the connundrum facing patent prosecution professionals when drafting arguments against a rejection that try to point out differences between the claimed approach and the cited art.
As an initial matter, as discussed in a previous post, the arguments should not merely parrot the claim language (here). This is where the applicant asserts, without more, that a references fails to teach a quoted claim limitation. Of course, not merely parroting claim language leads to arguments using words beyond quoted claim language as to why the cited art fails to show certain features. And such additional language then opens the applicant up to the charge that they are arguing limitations that are not in the claim (here) and (here).
One way for patent professionals to better learn how to navigate the fine distinctions between these issues is through examples. A recent PTAB case illustrates a case where the applicant strayed too far from the claim language in making arguments against the cited art. The case is Appeal 2024-000092, Application 16/835,420.
As the specification explains, the invention relates to “computational methods and computer systems for determining an answer in response to a query.” This includes a large language model used by the system. Claim 1 on appeal is provided below:
1. A method of using a dialogue computer, comprising:
receiving a query from a user;
providing the query to an input layer of a neural network;
extracting one or more triples from a knowledge graph, wherein the one or more triples correspond to the query;
injecting the extracted one or more triples into a plurality of nodes of an output layer of the neural network; and
determining an answer to the query based on the output layer injected with the one or more triples.
The examiner cited art (Li) that was somewhat the reverse approach as claimed. As explained by the Applicant in their brief, the cited art found triples from a knowledge graph that correspond to the input. Thus the cited art was trying to find a proper query from which to extract answers using the triples. Claim 1, on the other hand, used the network to determine an answer to the query using the triples extracted from the knowledge graph. When making their argument on this point, the Applicant stated that the “query is the input, and the triples are extracted from the input.”
This argument seems quite reasonable at first glance in light of the claim language confirming that the query is received from a user and provided to an input layer of the network. But that did not deter the examiner or the Board from responding that the Applicant was arguing limitations not in the claim (citations omitted):
The Examiner answers that Appellant argues limitations that do not appear in claim 1. In particular, the Examiner finds claim 1 recites that the triples are extracted “from a knowledge graph,” not from the “input,” i.e., the query. Appellant’s contention fails from the outset because it is not based on limitations appearing in claim 1. Further, the Examiner finds that the extracted triples need only “correspond to the query.” Appellant admits that Li discloses this limitation.We agree with the Examiner. Thus, Appellant does not persuade us by this contention that the Examiner errs, and, in light of the discussion above, we are persuaded the Examiner adequately shows that Li teaches or suggests the “extracting” step.
When a patent profession finds themselves trying to make an argument that is not directly supported by explicit claim language (even when logic and proper interpretation seem to be on their side), there is the option to make further amendments to the claim to better match the argument. And while the applicant must do more than merely parrot the claim language, at least they can do so with further explanation that is still based exactly on explicit elements.