The SSi/Adobe case

Well, first the readers should check out the Adobe press release. The case involved SSi (owned by Paul King), Adobe and Emigre. SSi had made a CD with thousands of fonts, mostly (if not exclusively) cloned from major foundries and shareware fonts. This was done by subtly moving the control points of the Bezier curves that describe outlines of characters. Adobe and Emigre sued and won. And Paul King settled with them. There were reactions on the newsgroups and on private bulletin boards. Mostly there was applause, and while I think the judgment is morally right (after all, SSi made lots of profits from blatantly derived work), I think that this particular judgment throws the font copyright situation into turmoil. It is flawed and sets a dangerous precedent. Before I give my own view at the technical level, I will let you read the comments by Richard Kinch, a font software expert who created Metafog, TrueTex and other fine products. He has written software to manipulate, alter and transform Bezier curves for his applications.

Richard Kinch's comments

On June 4 and 5, 1999, Richard Kinch wrote the following.

I grieve at this mistake of justice. How ironic that the triumphant news release admits the issue was not one of "programs", but mere coordinate data, ineligible for copyright. To quote:

So it was the points, not the programs, that had to be protected.

The casting of coordinate data into Type 1 programs is the trick: turn measurements of uncopyrightable shapes into textual code that resemble works of human authorship, and it then appears that you have a work described by the copyright act. Can a machine mill a copyrightable text from a grist of uncopyrightable facts?

This result reflects one aspect of the broken and incomplete state of software copyright doctrine (such as Altai) in US law. Courts can't distinguish between the literary work of a human author and the code generated by a machine. How can there be depth and subtlety in such a judgment? The industry will come to regret this extension of a flawed doctrine. It ultimately must corrode, not strengthen, software copyrights.

I believe in strong software copyrights. Today they are all mushier than ever before, because their strength derives not from the originality of the author as an individual, but in the adroitness of a large and expensive legal apparatus.

Let us hope that future judgments will make better sense of the real issues, and that the appelate system can construct something rational out of the mess of conflicts. This trouble is much broader than fonts. After so many years, we are worse off than when Lotus v. Borland was thrown out with no decision.

(and in his next message...)

The judgment itself is indeed clear and direct. I applied "mushy" to the strength of copyrights, not the decision. The trouble is the lack of underlying doctrine. There are many unsettled issues in software copyrights in US law, typefaces being one corner of the swamp. This decision makes an abrupt and ad hoc judgment without theorizing. That's appropriate, I suppose, in a summary action of a trial court, but the problem is that the contradictions and questions of law that this case so fully developed were not answered.

Contradictions surrounding human authorship remain. We know that in *some regard* glyph shapes are not copyrightable in the US, that is, are freely copyable, even the most beautiful ones which come from much human effort. We have, in my mind, no authoritative legal definition of what that freedom covers.

A Type 1 charstring precisely defines a glyph shape. No one has an exclusive right to that "shape", in some regard of the word "shape". This decision does not tell me how the shape itself can be freely copied, while the unique Bezier specification of it cannot.

After much exegesis, I am about ready to admit that the 1992 policy is so confused that one could cite whatever you like out of it to prove anything, so doing so doesn't count for much. I don't find anything in the 1992 policy to disagree with, but much of what it says is muddled. What I would have more liked to have seen is an explanation of how this all harmonizes with the 1988 policy that still supposedly is in effect.

The nature of the product is crucial to copyright, not the sweat of the brow. If the points could have been derived by a machine fitting Bezier curves to freely-copyable shapes, then any authorship inheres in the unprotected shapes and not in the Bezier measurements. If you give me a big enough bitmap made from those Type 1 charstrings, I have just such a machine. Is perfect autotracing now going to be illegal?

(and in his next message in reply to David Lemon...)

This is one of the perplexing aspects. Software needs an instructional or programmatic nature to fit the statutory definitions, as opposed to a compilation of data. (Thus the grating newspeak of calling them "font programs" instead of "fonts".) The copyrights were asserted for programs. But in the end we were told that, no, it was coordinate data, not programs, that were protected and infringed.

I was terse and unclear to Mr Lemon. Of course there was a lot of effort. But the casting of coordinate data as programs was performed by a machine. People chose the coordinate data to define the shapes they wanted, but these elements of data were cast into Type 1 charstrings by machines. So what the people did was laboriously create (unprotectible) typeface shapes. The program code that has the look of a literary work eligible for copyright is just a wooden derivation with no new authorship.

(and on June 6, he reacts to a message by John Hudson...)

Tiro Typeworks wrote:

I admire this statement. It succinctly expresses what the debate should be about, and a lot of what was argued and apparently settled in Adobe v. SSi. In a short form it states a legal theory of font software copyright.

So let me try to, just as succintly, explain why it troubles me:

An outline from a digital font is a letterform.

There. Now let me comment.

I believe the definitional distinction between glyph outlines (such as are defined in Type 1 charstrings) and letterforms is false. Not all letterforms exist as digital glyph outlines (some are metal type, or hand drawings, or bitmaps, etc.) but all digital glyph outlines are letterforms (they are shapes of letters, leaving off any concern about non-alphabetic fonts).

One way to create a digital font is to scan analog drawings and manually fit Bezier shapes to the scans. Reading from the court's decision, it appears that this is the process in view. So the legal theory seems to be saying the the hand drawings are the unprotected letterforms, while the Bezier shape is some kind of new, original, and therefore copyrightable expression of the hand drawing. The hand drawing is an idea, if you will, while the Bezier shape is an expression of that idea. To this I object.

A Bezier outline from an alphabetic font is itself a letter form, as much as if it were metal, a photographic film, a bitmap, or a drawing. All of these are merely constructs that define each point in 2-D space as either marked or unmarked, inside or outside. In fact, a Bezier shape constitutes a "stronger" example of "letter form" than those of analog media because it is mathematically precise and unambiguous in its inside-outside definition.

This Bezier definition of a shape is also unique, since any moving of points will change the shape (if we omit cases of redundant points); slight moves will make slight changes to the shape, and large will make large.

So returning to John's original statement, we arrive at a nexus.

First, how can the letter form be freely copyable, if the one and only way to perfectly express it, namely as a Bezier contour, cannot be copied?

Second, if bitmaps made from copyrighted contours are freely copyable then what if the Bezier coordinates may be derived (autotraced) from the free bitmap? Is not the mechanical derivative of a free thing also free?

Let me close by stating my "sampling theorem", which if proved would destroy the basis of extending copyrights to outlines.

I assert: All non-redundant Bezier contours having rational coordinates are equivalent to bitmaps at some resolution. They are "co-derivatives": Each may be algorithmically derived from the other without loss of information.

So whatever legal status you accord to oulines must be equally applied to bitmaps. Either bitmaps and outlines are freely copyable, or they are both someone's exclusive property.

Well, Richard Kinch hit the nail on the head. First of all, in the USA, font shapes cannot be copyrighted. Thus, the action, and the war, moves to the representation of these shapes in software. It is perfectly legal to take character shapes, scan them in, and use software like Scanfont to create fonts from these. Let us take several examples.

As Richard Kinch points out, how can you copyright outline data and not outlines? It does not make sense! I would like to add a little technical oddity. Assume that the shape of an A is described by XXX's font as cubic Beziers, as in a type 1 font. Well, to get an identical shape, and describe it by other Bezier curves, one has to place the control points in certain places, and there are not many choices. One can always break one Bezier curve in two or more Bezier pieces, but generally speaking, two adjacent Bezier pieces cannot be merged, and the knot joining two adjacent pieces cannot be moved or removed. So, if the unique uncopyrightable shape is represented by Bezier curves, the data representing it are severely restricted, and in fact, the data must in all but a few instances contain all the original knots! So, then, how can one possibly ask that these knots be the subject of copyright?

I think that this particular judge was taken for a ride by the lawyers of Adobe and Emigre. He made an erroneous judgment, and Paul King was within his legal rights. Morally, Paul King's actions were deplorable, but that is another matter.

Contact and Copyright

Luc Devroye
School of Computer Science
McGill University
Montreal, Canada H3A 2K6