Monday, April 23, 2012
DCU and SIPTU are on a collision course – finally!
It is worth first pointing out how much the world has moved on recently. Stanford and Berkeley started three new online courses today that will be in direct competition with the Irish universities, particularly computing at DCU;Sebastian Thrun reportedly got $25 million;As if the above wasn't enough, MITx will quicken its pace.Ireland's response has been the failure to agree a simple set of industrial relations procedures at DCU. I initially refused to participate in DCU's shenanigans in 2002; it cost me my job and my house. My view, now as then,was that SIPTU – who had a closed shop at DCU – needed to hang tough.In the meantime, Dr Howarth was effectively killed by DCU; several other workers lost their livelihood.My views on what really went on, and the cosy relationship between Bertie Ahern and SIPTU, have been rehearsed many times here. Now there is almost certain to be a serious dispute;
Subject: [DCUstaff]SIPTU Ballot Today
Reminder to all members of SIPTU:
Ballot on Statute 5 ( Suspension and Dismissal of Employees).
First voting time - today 12-2pm in the street area (Henry Grattan building).
Every SIPTU member has the right to vote - see information on ballots pasted below.
Marnie Holborow
Sec/Section Committee
--------------------------------------------------------------------------------------------------------------
DCU Statutes
The Section will be holding a ballot of all members in DCU on the statutes.
The ballot paper will read as follows:
I am in favour of Statute 5 (Suspension and Dismissal of Employees) in its current form □
I am not in favour of Statute 5 (Suspension and Dismissal of Employees) in its current form □
The two ballots will take place :
From 12 noon to 2pm, Monday 23rd April and the second one from 12 noon to 2pm , Wednesday 9th May. Both of these will take place in the Street area (Henry Grattan).
Every member of the union has the right to vote on this and everyone should try to ensure that they register their opinion on this important issue.
The Section Committee is calling for Statute 5 (Suspension and Dismissal of Employees), in its current form, to be rejected as it introduces redundancy, ’where the University deems necessary’ (5.1) as a reason for dismissal. Despite long negotiations, we were not able to persuade university management to remove this reference to compulsory redundancy.
We provide an account of the background to this issue below.
The Universities Act 1997 provides that:
33.(1) Subject to this Act and to the charter, if any, of the university, a governing authority of a university or the Senate may, and where required by this Act to do so shall, make such and so many statutes and regulations as it considers appropriate to regulate the affairs of the university.
(2) A governing authority shall, as soon as practicable after the making of a statute or a regulation under section 17 (2) or 24 (1), inform An tÚdarás and the Minister of the making of the statute and shall arrange for its publication in the Iris Oifigiúil.
The 1999 Statutes focused on Selection and Appointment procedures, and the 2001 Statutes ( Suspension and Dismissal of Employees) on dispute resolution procedure.
It was agreed, after rulings from a third party, that SIPTU and the University should discuss a redrafting of the statute, in order to establish whether agreement on the final draft of the statute could be reached.
Statute No. 5 of 2010 Suspension and Dismissal of Employees: Discussions on these Statutes began in 2007 but SIPTU withdrew from the discussions because they focused largely on redundancy. In 2010, discussions chaired by Alex White, between SIPTU and HR, resumed. HR refused to allow any changes or additions to DCU staff involved in the negotiations. The discussions continued over a number of months and we were pleased with some progress made on disciplinary procedures and dismissal.
However, we were most unhappy with the inclusion of a redundancy procedure in the Statutes. Other university statutes (TCD, UCD) do not mention redundancy. There has never been any compulsory redundancy in the Irish public service and, of course, under the Croke Park Agreement 2010-2014, there is an agreement that there will be no redundancies in the public service. We were also concerned about how, for academics, the introduction of compulsory redundancy would alter what has traditionally been understood by tenure for permanent staff. The discussions came to an end when it was felt that no more progress could be made. SIPTU repeatedly made the point that the Statutes would have to go to a vote by our members. Indeed, Sections 25, and 25.6 of the Universites Act 1997 states that any statute concerning dismissals, or dismissals by reason of redundacy shall ‘be established following consultation with trade unions’.
Despite the fact that SIPTU did not sign up to the Statutes, Governing Authority voted on them and a draft version appeared on the DCU website in November 2011. This was then taken down and replaced by the final version which says that the Statutes came into operation on the 8th December 2010. However, this is without the agreement of staff.
It is for these reasons and because Statute 5 represents the dismantlement of permanent employment status in DCU that the Section Committee cannot stand over this Statute and is therefore calling for it to be rejected in the upcoming ballot.
Dcu statutes * STATUTE NO. 1 OF 1999- Selection and Appointment Procedures
* STATUTE NO. 2 OF 1999 -Staff Appointment Procedures
* STATUTE NO. 3 OF 2001- Suspension and Dismissal of Employees (superseded by statute No. 5 below)
* STATUTE NO. 4 OF 2001 -Dispute Resolution Procedure
* STATUTE NO. 5 OF 2010 - Suspension and Dismissal of Employees
http://www.dcu.ie/info/statutes/index.shtml
Holborow's cant is sheerest humbug; this whole snafu is entirely her fault as she had an open goal in front of her in March 2003 as we were ready to issue a credible threat of a strike until she stopped it. (btw, Burns is right and H wrong in stating that the negotiations started in 2003, not 2007). Moreover, the new statute has for more serious problems with it than redundancy; again, I went into these in a previous post.
This was management's response. Remember that Brian Mac Craith's e-mail used to
bounce directly to Burns, who has a mainline connection to the real power at DCU;
Subject: [DCUstaff]Statute on Suspension & Dismissal
Dear Colleagues
I am writing to update you on the University's position in relation to the Statute on Suspension & Dismissal and to clarify some of the points which SIPTU have raised.
Since 2001 the University has consulted with SIPTU on an ongoing basis regarding the above Statute in line with Section 25 and Section 25.6 of the Universities Act. This section states that a Statute be established following consultation with trade unions.
The consultation involved a total of 20 meetings between SIPTU and the University, with Mr Alex Whyte TD (BL) facilitating a total of 15 meetings. These meetings took place between 2003 and 2010. From 2007 both parties agreed from the outset that the representation, as proposed by SIPTU and the University, would remain unchanged in order to maintain consistency and avoid disruption to the discussions.
A final draft text for a revised Statute was presented in a meeting, facilitated by Alex Whyte, between SIPTU and the University on 7th December 2010.
The final draft was then approved by the University's Governing Authority on 8th December 2010.
The section on redundancy has appeared in the Statute on Suspension & Dismissal since it was first approved by Governing Authority in 2001. It is important to note that the Statute simply outlines the procedure which the University will follow should a redundancy situation arise, similar to the clearly outlined procedure for dealing with a disciplinary situation.
The overall aim of the Statute is to protect the interests of staff and to ensure that fair procedures are in place to deal with particular employee relations matters.
The University, along with SIPTU is fully committed to the Public Service / Croke Park Agreement 2010 - 2014 which includes the guarantee of no redundancies in the public service. Notwithstanding this, the University has never made any permanent member of staff redundant, although the Section on Redundancy has been in the Statute since 2001.
I trust this clarifies the University's position regarding the Statute on Suspension & Dismissal, however should you have any queries or require further clarification, please do not hesitate to contact myself or Ms Emer McMahon, HR Manager on Ext. 5535 or via email at emer.mcmahon@dcu.ie.
Regards
Marian
-- Marian Burns Director of Human Resources Dublin City University Tel: (01) 700 5148
Readers can phone Marian to check that there is no way she wrote the above; she has the
IQ of an
iguana - like her predecessor - and God knows what garbage she has helped hire in DCU
In any case, here is what's really wrong with the new statute;
http://academictenure.blogspot.com/2011/01/dcus-new-illegal-statute.html
The redundancy argument may be a red herring. After the progress about which H boasts so much, the statute still includes provision for summary dismissal through the anti-definition of tenure that has been thrown out both by the High and Supreme courts. There is also a capacity for the “Andropov solution” - suspect the lecturer of mental incompetence, for example for his belief in Quantum mechanics.
It increasingly looks as though Ferdie decided to ignore IR, and take things through the courts instead. Who needs a redundancy agreement when you can summarily dismiss staff? When the court route failed, Arthur Cox told Mac Carith and his cronies at DCU that they could no longer give cover – so a team of useful idiots were kept on the SIPTU side negotiating this abomination.
Ireland may or may not be able to do universities; what is for sure is that north side Dublin cannot, and DCU should either be sold off or incorporated into the NUI.
Seán Ó Nualláin Ph D Stanford 23 u Aibrean 2012
PS
(Please note blogger has changed its format and I can't override it and go back to the old format
...........so this is a mess)
9u Bealtaine 2012.
Please note - SIPTU members at DCU voted 151-12 against statute no. 5
Total pool 163; the closed shop has been rescinded
Thursday, April 12, 2012
Genomics and natural language processing; how far can the analogy be extended?
This was published in 2007 in
O Nualláin, Seán and R. Strohman “Genome and natural language” in Witzany (ed.) “Biosemiotics in transdisciplinary contexts: Proceedings of Biosemiotics 2006. “
Helsinki; Umweb (2007) Pp. 249-260
It was my dear friend Richard’s last paper; he led the fight against the excessive claims of the HGP and greedy reductionism in Biology until his death
_______________________________________________________________________
Genome and natural language; how far can the analogy be extended?
Seán O Nualláin, lecturer, symbolic systems program, Stanford; visiting scholar, molecular and cell biology, Berkeley
Richard Strohman, Professor Emeritus, molecular and cell biology, Berkeley
1. Abstract
It is now almost universally accepted (Dennett, 1995; Kauffmann, 2000) that terms like “sign”, “signified”, and “semantics” are appropriate to genetic expression. Moreover, a field called “Biosemiotics” with an associated set of journals has come into being. However, how far can the analogy between genetic expression and natural language production and comprehension actually be pushed? This paper first looks at natural language, the best-known symbol system. It considers the various attempts that have been made formally to characterise human language. First, we broaden the context to consider language as one symbol system among others, including the genetic code. We then look at language through the prism of formal language theory before detailing some of the ingenious and thorough formalisms within the fields of formal and computational linguistics. We consider the layers within natural language itself, and that consideration leads to speculation about the role of context. Finally, we return to our original starting-point of genetic expression, and give a prognosis as to the likely progress of this field, as seen from a linguistic point of view. The consequences for biology and medicine are then spelled out. It is obviously vital that genetics learns lessons from the mistakes of another field. Just as the goal of eliciting meaning from parsing of strings of symbols proved infinitely more difficult in natural language than anticipated, so the goal of specifying production of proteins from nucleotide sequences is likely to exercise us for several generations.
2. Natural language (nl) and other symbol-systems
2.1 Layers of language and gene expression
Natural language (nl) seems to share with other human symbol systems like those that have been developed through music and art the following attributes (see O Nualláin, Seán (2003, chapter 7));
1. A hierarchical organization, whereby sentences and musical phrases both consist of top-level entities like sentence that can be broken down into subunits like noun phrases to be further analysed into words and so on.
2. Formal complexity of a certain degree (specifically for nl, that of indexed grammars,as we shall see), discussion of which will occupy the next subsection
3. A recursive structure. I state that (this is an example, because within the parentheses is a full sentence). Doug Hofstadter (1979) famously applied this notion, and that of self-reference, to music and art as exemplified by Bach, Goedel, and Escher.
4. Processing within micro-domains called contexts. This shall also constitute a full subsection of this paper, and undoubtedly applies to genetic expression
5. Metaphor. For Noyes in the Highwayman, the moon was a ghostly galleon; for Piet Mondrian, a jumble of boxes could be “Broadway boogie-woogie”.
6. Emotional impact
7. The possibility of self-reference. This sentence is false. All Cretans are liars, said the Cretan. Hofstadter (ibid.) argued that Bach died having finally composed a self-referential opus.
8. Ambiguity. The importance of this paper cannot be under-stated. The word “egregious” can mean outstandingly good or bad. Alternative splicing in the genome does seem to indicate exploitation of ambiguity by nature at this level.
9. Systematicity. We should be able to say “Cyrano loves the ‘precieuse’” because, regardless of semantics, we have already said that “ The ‘precieuse’ loves Cyrano”.
10. Duality of structure (in language, the phonological and syntactic levels) initially established, it is appropriate to predicate of language acoustic, semantic and pragmatic levels. We shall later concern ourselves with precisely how many levels to attribute to gene expression.
11. The notion of a native language, the phonotactic details of which become hardwired at the expense of all subsequent language learning. The distinction between “rue” and “roue” is audible; however, many of us enter a phonetic minefield in trying to distinguish “Caen” , and other soundalikes and sayalikes in French, some of which should not be said in polite company.
12. Creativity. Famously, we can all produce and understand an infinite number of sentences, as what we learn are underlying principles. The repeated statement of this idea underpins some of Chomsky’s reputation.
Of the above, 5,6,7, and 9 seem inappropriate for gene expression; the rest are at least up for grabs. It is appropriate to unpack 10 right now. It is always best to illustrate these phenomena with the aid of antinomial entities, which break the rules we are explaining. The German “zwei” breaks phonotactic restrictions in English. I will currently be breaking a syntactic rule that there is no future progressive in English. Semantic solecisms can be found aplenty in existential judgements about arms in Iraq made by our political elites in the early years of this millennium. Finally, were you to ask me whether I would like to stop here, and I said “yes” without understanding that you might wish to take a break, that would constitute a pragmatic error on my part.
Let us clarify. Phonotactic rules govern combination of phonemes into allowable sound-patterns for each particular language. Syntax deals with allowable combination of words. “Semantics” is a much more complex nexus of concepts. One the one hand, it is about meaning; yet elicitation of meaning often requires recourse to a further level, that of pragmatics. On the other hand, “semantic” formalisms are restatements of linguistic propositions, often in terms of formal logic. So “I got rhythm” may admit of the semantic correlate “Got rhythm (x)”. How far does that restatement advance one on the way to external “stuff”, to “meaning”, to specific proteins? There’s the rub. It be all that’s necessary, or it may be just one step on the road. In general, if context is severely restricted, then restatement in a semantic formalism may be sufficient. In actual fact, syntax, or indeed bare individual words, may be sufficient for full meaning (specification of proteins) if context is severely restricted. The pretext of some of the exaggerated claims that came from the HGP was that context was always so restricted. It is not; we have generations of work ahead of us looking at how nucleotide sequences differentially produce different proteins depending on the metabolic context.
While the spectacle of Connie Chung explaining on primetime TV in 1989 that the genome specified everything we are and could be now looks like a bad joke, it must be said that many diehards still have an eschatological hope that full nucleotide sequencing will reveal all. For example, Nick Campbell in Nature reviews Genomics in February 2004 states that biological development is readable in a deterministic, rather than epigenetic manner from gene sequences. Similarly Goodman et al (2005) imply an aspiration to hard-core genetic determinism. So the “one gene-one enzyme” paradigm of Beadle et al (1941) lives on.
In a sense, genetically transmitted diseases can show that the Word can be misspelled when it takes flesh. The inherited disease familial dysautonomia arises from a single-nucleotide mutation in a gene called IKBKAP (Ast, 2005). This corresponds to the phonotactic and orthographic levels in human speech and text; “Next” and “nest” are similarly different. Likewise, at a lexical level, Tay-Sachs is caused by a misspelling of nucleotides. At the syntactic level, Bcl-x, which governs cell death, can be alternatively spliced into Bcl-x (S), a promoter of cell-death, and Bcl-x(L), a suppressor thereof (ibid).
A linguistic analogy to this is the AI classic
The robot saw the hill with the telescope.
So did the robot need ocular aids to see the hill, or was it looking bare-sensored at an observatory? Does “with the telescope” attach to the robot or to the hill? This is structural ambiguity; to be more specific, it is a problem of prepositional phrase attachment, which needs access to the semantic or a deeper level to be resolved. The nature of this layer has not yet been established for the genome.
Strohman (2000) argues, with respect to the possibility of such a layer, that there exists a mesoscopic layer between the genome and phenotype wherein complex organisational states can exist and where there exist networks of regulatory proteins capable of organising patterns of gene expression and much other emergent cellular behaviour in context-dependent ways. In a later summary, Strohman (2003) refers to classic work by Veech et al. (2001) that established that the rate-limiting enzyme controlling which genes are on or off is a function of the entire mesoscopic system. In particular, different enzymes will be used to do the same tasks in different contexts.
And yes, semantics in natural language has proven almost that messy. In O Nuallain (forthcoming a) and passim below, this author contrasts the early Wittgenstein’s HGP-like sunniness about the prospect of a neat semantic description for language taken as a whole with his later wholesale rejection of that idea, and his insistence that language could be processed only within microcontexts that he called “Language games”. So is the case also for gene expression.
Let us summarize the elements of the proposed analogy between the genome and natural language. Nucleotides, the four letters, spell words, the twenty amino acids. These get linked together into chains of varying lengths, the proteins which can be words (let’s allow some recursivity), phrases, or sentences. Occasionally, in natural language, we happen on words, phrases, and “collocations” that are (almost) unambiguous in meaning. Technical terns like “Trilobite”, “formaldehyde”, “fratricide” and proper names like “Thailand” are examples. Collocations include “media circus”, “team effort”, “world record”, and so on. These are low-hanging fruit for natural language-processing computer programs. Interestingly, they are almost context-independent, in that they have only one widely-used metaphorical meaning that is distinct from a little-used literal meaning. In the extension of this analogy, the HGP picked up only technical terms, proper nouns, and perhaps those collocations which are radically context-independent. So it should come as no surprise that only 2% of diseases currently admit of a straightforward genetic explanation.
Jacob et al. (1962) focussed over a long period of time on the work of regulatory genes in their concept of “operons”. These genes were a key to unlock the actions of a set of other genes, which for example would generate trytophan. Regulatory genes, in turn, were switched on and off by proteins. So genes can be turned on and off to respond to the environment. Let us now revisit the analogy, which includes “context” qua the environment. Gene expression now includes feedback loops. We effectively need a new HGP for every context that organisms encounter.
Strohman (opera cit) expands on these points. The expression of the genome is regulated through biochemical mechanisms that sense the bioenergetic state of the cell. In particular, the metabolites NAD and NADH and other synoptic signals represent instant by instant changes in the bioenergetic status of the cell. Changes of metabolites like fatty acids and glucose result in differential gene expression through binding to transcription factors. Strohman wishes to stress the consequences for progress in biology and medicine. Metabolism can be altered by environmental factors like sedentary behavior as by gene mutations like amyloid production. Alterations in metabolism, in non-syntactic phenomena, are the proximate cause of disease, and cures can be sought without interfering with the genome, or formal language.
Specifically, he argues that gene expression may be regulated by NAD dependent histone deacetylase via epigenetic marking of chromosomal histones as during the life extension resulting from caloric restriction. Again, context affects meaning of a language string in the linguistics analogy. (Let us emphasise that the major problem in nl processing has always been ambiguity).Secondly, he continues that alteration in protein amount or structure, as in dystrophin or amyloid may alter the rate of metabolic reactions resulting in an altered phenotype. Finally, posttranslational modifications like phosphorylation modify proteins. The rather messy picture that is emerging reminds one of NL semantics, and indeed of life itself
Strohman is also attempting an oblique attack at the central dogma of molecular biology; the deterministic, linear, uni-directional, and encapsulated path from DNA to phenotype. Specifically, he wishes to frame the relationship of genotype and phenotype in terms of a complementarity between genetics and dynamics, between the language as a formal system and the context, to use the corresponding linguistic terms. He unpacks and extends Waddington’s (1966) notion of the “epigenetic landscape” by proposing linkages and feedback loops between the DNA, phenotype, proteins, environment, and behaviour. Kaufmann (2000) remarks that this nexus of gene and environment must be kept at a very specific state, the “edge of chaos”, for maximum creativity to occur. Therefore, behaviors might be genetically assimilated to give Lamarckian effects without violating the central dogma that Waddington desired; alternatively, as in the case of the scandens finch on the Galapagos, birdsong might become entirely learned down the patrilineal line if emergent properties from the nexus of organism and environment decides that this is a more economical way of storing it than imposing it on the genome.
2.2 Formal Language theory and NL formalisms.
Derek Jarman’s classic short film about Wittgenstein proposes that he was nostalgic for the days of “The ice”; clear, lucid theories of language that he later eschewed, perhaps coincidentally (or not) after living for some time in this author’s native island of Ireland. . This is particularly the case as there are deep and beautiful connections between formal language theory, set theory, and computability. The Chomsky hierarchy posits languages of different levels of formal complexity. At the top level, level 0, are languages that can be modelled by non recursively-enumerable sets. The next level, level 1, features languages that are recursively enumerable. Both of these types of language, and the automata that generate and recognise them, are too powerful to be appropriate models for human language. Level 2 languages are generally also considered too powerful to be adequate models for nl; however, Shieber (1984) adduced evidence that Swiss German included constructions that only formal systems as powerful as level 2 could handle. (Partly because Swiss German, even more than other languages, is a set of disparate dialects, this result has never been fully accepted in the linguistic community). It cannot be proven that any randomly chosen level 2 language can be recognised in finite time; likewise, computational problems at this level cannot have posited of them a solution in finite time. These connections are reminiscent of the unexpected connections between fractals and chaos; a priori, there is no reason to anticipate this type of phenomenon.
The final explicitly recursive level, level 3, is the consensus location for nl. Even Swiss German can be handled by a relatively trivial addition to level 3 called indexed grammars. Level 3 is also called the level of “phrase-structure” grammars. It is illustrated by sentences such as “The mat, on which the cat sat, which Séamus made, is in the hall” where we need agreement between subject and its verb often at a considerable lexical distance. We describe such agreement using the mathematical expression (a^nb^n); Swiss German apparently needs constructions like (a^nb^nc^n). Even a^nb^n constructions cannot be handled with any degree of elegance by level 4 grammars.
Of course, we are uncertain as to where the genetic code lies in this schema. Phenomena like alternative splicing indicate that there is some degree of ambiguity, and thus complexity. Hox genes indicate a degree of hierarchical organization, context-sensitivity as defined in the terms of the Chomsky hierarchy (see O Nuallain, forthcoming c) and long-distance dependencies, indicating that perhaps the phenomena Shieber remarked on are present also in the genome. In any case, the HGP treated the Genome as if it was similar to the parody of language apparent in the pattern-matching programs of the 1960’s. Lisp programs were programmed in these systems with preprogrammed scripts like “I have problems with my x”, to which they would reply “Tell me more about x”. It is likely that parsing the genome is infinitely more complex than this.
There has been much superb formal and computational linguistics since the 1960’s, though the general problem of parsing any arbitrary sentence in any arbitrary natural language and extracting meaning therefrom remains inviolate. Thus, 100% accurate machine translation will forever remain a pipedream. The original so-called Chomskyan “revolution” which at least had the virtue of adding the rigour of Chomsky’s teachers to the area, was attacked with respect to its dichotomisation of “syntax” with a “black box” called “semantics”. In particular, the early Chomsky had difficulty with the manifest difference that could be wrought by a quantifier like “some” in the two sentences “Formal linguistics is nonsense” and “Some formal linguistics is nonsense”. Thereafter, at one extreme, formalisms like categorial grammar shifted the burden to the lexicon; therefore, the lexical item’s (or gene’s) possible syntactic and semantic roles were assumed encoded in it. It is likely that genetics will plumb for a formalism like this, even though others like lexical-functional grammar are manifestly superior for language. Chomsky’s later X-bar predilection is in keeping with the MIT tendency to impose more on the lexicon in the manner of categorial grammar.
Compositional semantics demanded that the syntactic components should each generate expression in a semantic formalism like lambda-calculus, which could then be summarized through processes like beta-reduction. The resulting output, corresponding to the “meaning”, was initially presumed to require no further parsing despite its opaque nature. However, the formal semantic analysis of a sentence like “Would you like to keep quiet for a moment?” in whatever semantic formalism does not constitute its meaning in any real sense. That requires pragmatics; the perlocutionary impetus of the statement in pragmatic terms is a request for silence, not an expression of the speaker’s pleasure.
Yet nl processing (nlp) made massive strides for some time; indeed, it is fair to say that an asymptotoic state was reached in the early 1990’s. Syntactic parsing is a textbook affair; the national software registry allows downloading of many useful tools. Type 3 grammars are well represented by systems like the Alvey natural language toolkit. This system maps a huge subset of English onto lambda calculus, a semantic formalism whose origins reveal much about the origins of formal computational theory. A process called beta-reduction makes the resulting output more economical. What then? Well, for each specific application, as is mentioned below, the words to be used must be implemented as separate Lisp objects and taught how to interact with the semantic formalism. We are right back to square one; the omni-pervasiveness of context. Likewise, even the best natural language generator systems like Charon has difficulty in formalizing the interplay of linguistic and conceptual elements. Absurd claims dating back to Schank’s (1975) work have not helped and perhaps quickened the tendency within genetics to use what looked like a happy language metaphor. It behooves us to look at context more closely.
3. Context
While “culture” has claims to be the most confusing world in the English language, even its meaning will depend on context. “Context” is much talked about, and very little understood. Even the most naïve PR person knows to complain that his egregious statements are being taken “out of context”. We have gotten little distance with this term in decades of nl by computer, despite everyone understanding what it is in folk psychological terms. What then is the prognosis for gene expression in context?
Perhaps context can best be introduced by considering the project of the early Wittgenstein (1922), hinted at above. Wittgenstein sought a grand unified theory (GUT) of language. He argued that what was out there in the world was “Sachverhalten” states of affairs. The world, he argues, “is all that is the case”. The Sacverhalten could be decomposed into Tatsachen, simple elements. Language, he argued, consisted of propositions that could be decomposed into atomic propositions. Atomic propositions could be mapped on to simple elements by what he called a “private language”, idiosyncratic to everyone (Think of it as the instruction set of a computer). Wittgenstein later repudiated this “Tractatus” idea of language in favour of one that gave a pre-eminent role to context. Language, he argued, should be considered in context; language-games like guessing riddles, playing chess, and telling jokes should be the focus of study. Otherwise, he argued, one ended up asking ridiculous questions in language like “What is a language”? To spell it out, one is already immersed in a language-game as one formulates the question, and one cannot bootstrap oneself up to an objective perspective.
Wittgenstein (1967) presses the attack at this point. Language is like a city, he opined, with warrens in old sections contrasting with the rationally laid-out modern thoroughfares that are analogous to scientific discourse in language. The process of interpretation of the language is going to be vastly different as one goes from one section to the other. Relatively clean types of interpretation that even the HGP could support will work for the modern section, but not for the warrens. Of course, Wittgenstein should not be our main source here. There is plenty of other evidence for the hypothesis that language does not admit of a single method of analysis and, by extension, that we need to be very careful about positing a monolithic type of gene expression.
Even allowing for the Stalinist bias he was forced to impose on his writings, Vygotsky (1962) can responsibly be interpreted as arguing for different evolutionary roots for thought and language. Thought can be discerned in animals’ problem-solving; language originates in signalling-systems like birdsong. Piaget’s (1972) monumental research oeuvre gave pride of place to “operational knowledge”; conceptual and motor knowledge whose origin is an internalisation of our interaction with the world. For Piaget, language is a form of operational knowledge; Vygotsky, probably more correctly, would emphasise more the autonomy of the linguistic apparatus. (Indeed, the loss of reputation that Piaget has suffered perhaps can be ascribed to his stubbornness on this point, which made him vulnerable to the Chomskyans as Cognitive Science developed. His most prominent intellectual heir is ironically the renegade Chomskyan George Lakoff, who does not acknowledge Piaget’s influence).
Again, the musings of two deceased psychologists may be of less than urgent interest to readers of this journal. Yet their speculations gain impetus from a survey of the successes and failures of nlp, and particularly machine translation (MT). Talented researchers like Nirenburg (Nirenburg et al, 1991) and Roger Schank’s students (Schank, 1975) implicitly followed the path of the early Wittgenstein, as they sough the Holy Grail of MT; fully automatic translation between human languages by mapping onto a set of language-neutral logical atoms and then mapping from these atoms onto target text. So, for example, we go from “Is fear mé” in Gaelic through a formalism like Schank’s conceptual dependency, which claimed that all meanings could be expressed in a set numbering in the low teens (it varied) of logical atoms and we output a Gallic Muddy Waters “Je suis un homme” in French.
Both Nirenburg and the Schankian school later saw the error of their ways, and began to produce effective systems that worked within specific contexts. In particular, Schank borrowed a concept from Piaget via Frederic Bartlett of the “script”, a stereotyped set of actions. So there is a script for entering a restaurant, for diplomatic visits, for earthquakes, and so on. Again, the parallel with Wittgenstein’s “language-games” is striking. The later Schankian systems (Schank et al, 1975) were surprisingly effective within micro-context; yet, as we have described above for the Alvey systems, the Schankian logical primitives have to be mapped to specific words in the micro-context (for example, “open”, “ask”, and “sit” in the restaurant example) for processing to occur. Nirenburg’s altogether more detailed work began to distinguish between generally available semantic distinctions (we always consider whether entities are living or not) to context-dependent ones (is that a cheesy teenage song, or not?). Our own work (Ó Nualláin et al., 1994) confirmed this as we used the Alvey tools to construct visual scenes on a screen on the basis of nl input.
Let us recap. All processing of language, and indeed all cognition, is contextual (Ceci et al, 1994). (Indeed, this author would argue (forthcoming, b) that the cognitive role of self is to prevent information overload by keeping contexts separate; the alternative is mental illness. We narrate to ourselves continually to prevent this). The relationship between syntax and semantics was at least addressed in nl processing by computer (nlp) (Lesmo et al, 1985) and needs to be in genetics. Contexts, at first blush, seem to be idiosyncratic interactions between linguistic and operational knowledge, which require knowing the precise relationship between the words (the genes) and semantic formalism (the metabolic context) in order for correct processing to occur (in order to predict what proteins will be generated). It is likely indeed that all of this holds for genetics; access to entities like perlocution seems to demand knowledge of oneself as the object of another’s wishes, and therefore consciousness, and is not relevant to genetics.
Cognition is also massively hypothesis-driven, and it is likely that every act of gene-expression draws in some way on the experience of the whole organism. We can distinguish in language between acts of context-determination, which don’t involve any but generally available semantic primitives like +-alive, and processing within contexts where there is interaction between context-specific primitives and words. It is not unreasonable to expect such phenomena in genetic expression. Yet there is a deeper consideration still in language and context with which we will close this section.
Any language processing act involves evidence from orthographic/phonotactic, lexical, syntactic, semantic/operational, and pragmatic sources of knowledge. The classic “pipeline” model sees information travelling strictly from left to right in this schema. So we assemble sounds into words, parse the words syntactically, add some semantic interpretation, and finally take into account where we are in the discourse, what the interlocutor is expecting of us, and arrive at an interpretation and/or course of action. However, as argued in Jackendoff (1987) and elsewhere, interactions can be a great deal more complicated. The letters “I L Y” in a certain context, that of a marriage proposal a la the famous scene in Anna Karenina, can result in a rigorous business contract with punitive buy-out clauses. For Wittgenstein (1967), a labourer saying “slab” was being quite linguistically sufficient in the circumstances of a construction project; he wants a slab, rather than to discourse on the geology of the material. So we can have connections from the orthographic or lexical straight to the pragmatic if context is sufficiently restricted.
Specifically, context seems to deform the layers of language as it becomes restricted in much the same way that gravity deforms space-time as one approaches the surface of a planet. At a level intermediate between reading the New York Times and I L Y, semantic relations are appropriated by the syntax in “semantic grammars” used for natural language interface to public services. The HGP worked on the assumption that the context was always going to be sufficiently restricted for single words to work, and therein lies its failure. It is a valuable lexicographic tool, and therein lies its success. However, we also need syntax, semantics, discourse pragmatics, and enumeration of contexts if nl is anything to go by.
Barbieri’s work (1998, 2002) is a fundamental analysis of biosemiotics and related fields like biosemantics. He makes the point from a semiotics point of view that the distinction between context-independent and context-dependent semantic primitives needs to be emphasised in genetics. It may indeed be the case that the HGP, considered thus, has revealed only the low-hanging fruit; context-dependent primitives. Alternatively, it may be useful to consider the HGP’s findings as collocations, or indeed as purely a lexicon; in any case, low-level fruit with context rigorously circumscribed.
Conclusions; future research directions
So is NL anything to go by? The answer is “at least partly”. In NL understanding, we are often concerned with getting the gist of a full story, and that is when the full artillery of linguistic techniques is used. In Gene expression, we are initially concerned with building a full organism, so the analogy is appropriate here. Natural language is also used to elicit a simple answer to a simple question, or specify a protein in genetic terms; such elicitation may be available to the technique known as “semantic grammars” in language, no correlate of which has yet been found in genetics. Can it be the case that combinations of metabolic context and nucleotide sequence recur in ways which are useful to predict the generation of proteins? A first step is obviously to do a human genome syntax project in the way that Ast (2004, 2005) has hinted at in his analysis of alternative splicing. Only then can we see if fatty acids are been taken up by transcription factors in systematic ways that echo the appropriation of semantic roles by syntax in “semantic grammars” (Ó Nualláin, 2003, Pp. 121-126).
Sometimes we understand language in order explicitly to act; we map the language onto a set of Piagetian “schemes” (Schankian “scripts”), routinised sequences of behaviour. Similarly, we read in order to evoke a script that we may or may not act on. Such leisure is not available to the genome; undoubtedly, however, the task of organism construction is a s complex as anything in language (and of course itself generates the elements of the linguistic apparatus).
The prognosis for genetics from analysis of the history of nlp must be, then, that it must start research into the syntax of the genome at a level much deeper than mere introns and exons. While there has been preliminary work on this, it is urgent that we decide issues such as what level of the Chomsky hierarchy the genome lies on. Are there Swiss German type constructions for certain creatures? Could it be the case that the difference between human and chimpanzee brains, to take one celebrated example, is that between Chomsky types? The consensus from Salzberg et al. (1998) is that, whereas the genome is susceptible to description by finite-state automata, protein-folding requires dependencies at long distances, and thus the artillery of context-free grammars. Abe et al. (1997) comment on protein prediction.
It must then try and understand the “semantics”, the metabolic context for the particular case that Veech et al (2001) examine. Then comes issues of “discourse structure”; how tasks like building and maintaining the integrity of the organism act as high-level goals affecting the minutiae of protein generation. Finally come issues relating to how these elements come together in specific contexts. There are generations of work ahead, all of which will bear fruit for biology and medicine.
Specifically, the template-matching “one gene, one protein/enzyme” story is childish. When we arrive at a more sophisticated understanding, it will have huge consequences for biology and medicine. Gene expression involves interactions of genome, environment, and emergent characteristics of networks of proteins as Strohman (opera cit.) has rightly argued. In this framework, we can begin to understand how exercise and diet affect metabolism, how metabolism affects gene expression, and how health is maintained as the dynamic set of relations it is.
References
Abe, N., & Mamitsuka, H. (1997). Predicting protein secondary structure
using stochastic tree grammars. Machine Learning, 29, 275-301.
Ast, G. (2004) “How did alternative splicing evolve?” Nature Reviews Genetics, Vol. 5, Pp. 773-782
Ast, G. (2005) “The alternative Genome” Scientific American, April 2005, 58-65
Barbieri, Marcello (1998). The Organic Codes. The basic mechanism of macroevolution. Rivista di Biologia-Biology Forum, 91, 481-514.
Barbieri, Marcello (2002). Has Biosemiotics come of age? Semiotica, 139, 1/4, 283-295
Beadle, G.W. and E. Tatum (1941) “Genetic control of biochemical reactions in Neurospora” Proc. Natl. Acad. Sci. USA 27, 499-506
Nick Campbell “Building the complete toolkit” Nature Reviews Genetics 5, 81-81 2004
Ceci, S. and A. Roazzi (1994) “The effects of context on cognition” In R. Sternberg and R. Wagner (eds.) “Mind in context”. NT;CUP, Pp. 74-101
Dennett, D. (1995) Darwin’s dangerous idea. New York: Simon and Schuster
Eurotra (1988) Technical manual 5.0 Luxembourg: CEC publications
Goodman, M., Grossman, LI & Wildman, DE Trends. Genet.21,511–517 (2005).
Hofstadter, D. (1979) Goedel, Escher, Bach. New York: Basic Books
Jackendoff, R. (1987) Consciousness and the computational mind. MIT Press.
Jacob, F., R. Sussman, and J. Monod (1962) “Sur la nature du répresseur assurant l’immunité des bactéries lysogenes” Comptes rendus des Academies des Sciences, 254, 4214-16
Kauffmann, S. (2000) Investigations. New York: OUP
Lesmo, L., and Tirasso, P. (1985) “Weighted interaction os syntax and semantics in natural language analysis” In proceedings of IJCAI (pp772-778) Los Angeles
Nirenburg, S. and Goodman, K. Eds (1991) The KBMT project. San Mateo, Ca: Morgan Kaufmann
Ó Nualláin, Seán (2003) The Search for Mind; third edition. Exeter: England
Ó Nualláin, Seán (forthcoming a) “Subjects and Objects” Biosemiotics
Ó Nualláin, Seán (forthcoming, b) “Ask not what to do for your self” Cognitive Processing
Ó Nualláin, Seán (forthcoming, c) “Code and context in gene expression, symbolic cognition, and consciousness” In M. Barbieri (ed) Biosemiotics 2 (provisional title). Kluwer.
Ó Nualláin, S, Arnold G. Smith: An Investigation into the Common Semantics of Language and Vision. Artif. Intell. Rev. 8(2-3): 113-122 (1994)
Piaget, J. (1972) Principles of genetic epistemology London: Routledge
Steven Salzberg, David Searls, and Simon Kasif (1998) Computational Methods in Molecular Biology. Published by Elsevier Science
Schank, R. (Ed.) (1975) Conceptual Information Processing Amsterdam: North Holland
Schank, R. and Abelson, R. (1977) Scripts, plans, goals, and understanding. Hillsdale, NJ: Erlbaum
Searle, J (1980) “Minds, Brains and Programs” The behavioural and brain sciences
Shieber, S. (1984) “Evidence against the context-freeness of natural language” (TW-330) Menlo Park, Ca:SRI International
Strohman, R. (2000) “Organization becomes cause in the matter” Nature biotechnology, June 2000, Vol. 18, Pp. 575-576
Strohman, R. (2003) “Thermodynamics- old laws in medicine and complex disease” Nature Biotech, May 2003, Vol 21, Pp. 477-78
Veech, R.L., B. Chance, Y. Kashiwaya, H. Lardy, and G. Cahill (2001) “Ketone Bodies, potential therapeutic uses” IUMB Life 51:241-247
Vygotsky, L. (1962) Thought and Language Cambridge, Mass: MIT Press
Waddington, C (1966) Principles of development and differentiation NY: Macmillan
Wittgenstein, L. (1922) Tractatus logico-philosophicus Oxford: Blackwell
Wittgenstein, L. (1967) Philosophical Investigations Oxford: Blackwell
O Nualláin, Seán and R. Strohman “Genome and natural language” in Witzany (ed.) “Biosemiotics in transdisciplinary contexts: Proceedings of Biosemiotics 2006. “
Helsinki; Umweb (2007) Pp. 249-260
It was my dear friend Richard’s last paper; he led the fight against the excessive claims of the HGP and greedy reductionism in Biology until his death
_______________________________________________________________________
Genome and natural language; how far can the analogy be extended?
Seán O Nualláin, lecturer, symbolic systems program, Stanford; visiting scholar, molecular and cell biology, Berkeley
Richard Strohman, Professor Emeritus, molecular and cell biology, Berkeley
1. Abstract
It is now almost universally accepted (Dennett, 1995; Kauffmann, 2000) that terms like “sign”, “signified”, and “semantics” are appropriate to genetic expression. Moreover, a field called “Biosemiotics” with an associated set of journals has come into being. However, how far can the analogy between genetic expression and natural language production and comprehension actually be pushed? This paper first looks at natural language, the best-known symbol system. It considers the various attempts that have been made formally to characterise human language. First, we broaden the context to consider language as one symbol system among others, including the genetic code. We then look at language through the prism of formal language theory before detailing some of the ingenious and thorough formalisms within the fields of formal and computational linguistics. We consider the layers within natural language itself, and that consideration leads to speculation about the role of context. Finally, we return to our original starting-point of genetic expression, and give a prognosis as to the likely progress of this field, as seen from a linguistic point of view. The consequences for biology and medicine are then spelled out. It is obviously vital that genetics learns lessons from the mistakes of another field. Just as the goal of eliciting meaning from parsing of strings of symbols proved infinitely more difficult in natural language than anticipated, so the goal of specifying production of proteins from nucleotide sequences is likely to exercise us for several generations.
2. Natural language (nl) and other symbol-systems
2.1 Layers of language and gene expression
Natural language (nl) seems to share with other human symbol systems like those that have been developed through music and art the following attributes (see O Nualláin, Seán (2003, chapter 7));
1. A hierarchical organization, whereby sentences and musical phrases both consist of top-level entities like sentence that can be broken down into subunits like noun phrases to be further analysed into words and so on.
2. Formal complexity of a certain degree (specifically for nl, that of indexed grammars,as we shall see), discussion of which will occupy the next subsection
3. A recursive structure. I state that (this is an example, because within the parentheses is a full sentence). Doug Hofstadter (1979) famously applied this notion, and that of self-reference, to music and art as exemplified by Bach, Goedel, and Escher.
4. Processing within micro-domains called contexts. This shall also constitute a full subsection of this paper, and undoubtedly applies to genetic expression
5. Metaphor. For Noyes in the Highwayman, the moon was a ghostly galleon; for Piet Mondrian, a jumble of boxes could be “Broadway boogie-woogie”.
6. Emotional impact
7. The possibility of self-reference. This sentence is false. All Cretans are liars, said the Cretan. Hofstadter (ibid.) argued that Bach died having finally composed a self-referential opus.
8. Ambiguity. The importance of this paper cannot be under-stated. The word “egregious” can mean outstandingly good or bad. Alternative splicing in the genome does seem to indicate exploitation of ambiguity by nature at this level.
9. Systematicity. We should be able to say “Cyrano loves the ‘precieuse’” because, regardless of semantics, we have already said that “ The ‘precieuse’ loves Cyrano”.
10. Duality of structure (in language, the phonological and syntactic levels) initially established, it is appropriate to predicate of language acoustic, semantic and pragmatic levels. We shall later concern ourselves with precisely how many levels to attribute to gene expression.
11. The notion of a native language, the phonotactic details of which become hardwired at the expense of all subsequent language learning. The distinction between “rue” and “roue” is audible; however, many of us enter a phonetic minefield in trying to distinguish “Caen” , and other soundalikes and sayalikes in French, some of which should not be said in polite company.
12. Creativity. Famously, we can all produce and understand an infinite number of sentences, as what we learn are underlying principles. The repeated statement of this idea underpins some of Chomsky’s reputation.
Of the above, 5,6,7, and 9 seem inappropriate for gene expression; the rest are at least up for grabs. It is appropriate to unpack 10 right now. It is always best to illustrate these phenomena with the aid of antinomial entities, which break the rules we are explaining. The German “zwei” breaks phonotactic restrictions in English. I will currently be breaking a syntactic rule that there is no future progressive in English. Semantic solecisms can be found aplenty in existential judgements about arms in Iraq made by our political elites in the early years of this millennium. Finally, were you to ask me whether I would like to stop here, and I said “yes” without understanding that you might wish to take a break, that would constitute a pragmatic error on my part.
Let us clarify. Phonotactic rules govern combination of phonemes into allowable sound-patterns for each particular language. Syntax deals with allowable combination of words. “Semantics” is a much more complex nexus of concepts. One the one hand, it is about meaning; yet elicitation of meaning often requires recourse to a further level, that of pragmatics. On the other hand, “semantic” formalisms are restatements of linguistic propositions, often in terms of formal logic. So “I got rhythm” may admit of the semantic correlate “Got rhythm (x)”. How far does that restatement advance one on the way to external “stuff”, to “meaning”, to specific proteins? There’s the rub. It be all that’s necessary, or it may be just one step on the road. In general, if context is severely restricted, then restatement in a semantic formalism may be sufficient. In actual fact, syntax, or indeed bare individual words, may be sufficient for full meaning (specification of proteins) if context is severely restricted. The pretext of some of the exaggerated claims that came from the HGP was that context was always so restricted. It is not; we have generations of work ahead of us looking at how nucleotide sequences differentially produce different proteins depending on the metabolic context.
While the spectacle of Connie Chung explaining on primetime TV in 1989 that the genome specified everything we are and could be now looks like a bad joke, it must be said that many diehards still have an eschatological hope that full nucleotide sequencing will reveal all. For example, Nick Campbell in Nature reviews Genomics in February 2004 states that biological development is readable in a deterministic, rather than epigenetic manner from gene sequences. Similarly Goodman et al (2005) imply an aspiration to hard-core genetic determinism. So the “one gene-one enzyme” paradigm of Beadle et al (1941) lives on.
In a sense, genetically transmitted diseases can show that the Word can be misspelled when it takes flesh. The inherited disease familial dysautonomia arises from a single-nucleotide mutation in a gene called IKBKAP (Ast, 2005). This corresponds to the phonotactic and orthographic levels in human speech and text; “Next” and “nest” are similarly different. Likewise, at a lexical level, Tay-Sachs is caused by a misspelling of nucleotides. At the syntactic level, Bcl-x, which governs cell death, can be alternatively spliced into Bcl-x (S), a promoter of cell-death, and Bcl-x(L), a suppressor thereof (ibid).
A linguistic analogy to this is the AI classic
The robot saw the hill with the telescope.
So did the robot need ocular aids to see the hill, or was it looking bare-sensored at an observatory? Does “with the telescope” attach to the robot or to the hill? This is structural ambiguity; to be more specific, it is a problem of prepositional phrase attachment, which needs access to the semantic or a deeper level to be resolved. The nature of this layer has not yet been established for the genome.
Strohman (2000) argues, with respect to the possibility of such a layer, that there exists a mesoscopic layer between the genome and phenotype wherein complex organisational states can exist and where there exist networks of regulatory proteins capable of organising patterns of gene expression and much other emergent cellular behaviour in context-dependent ways. In a later summary, Strohman (2003) refers to classic work by Veech et al. (2001) that established that the rate-limiting enzyme controlling which genes are on or off is a function of the entire mesoscopic system. In particular, different enzymes will be used to do the same tasks in different contexts.
And yes, semantics in natural language has proven almost that messy. In O Nuallain (forthcoming a) and passim below, this author contrasts the early Wittgenstein’s HGP-like sunniness about the prospect of a neat semantic description for language taken as a whole with his later wholesale rejection of that idea, and his insistence that language could be processed only within microcontexts that he called “Language games”. So is the case also for gene expression.
Let us summarize the elements of the proposed analogy between the genome and natural language. Nucleotides, the four letters, spell words, the twenty amino acids. These get linked together into chains of varying lengths, the proteins which can be words (let’s allow some recursivity), phrases, or sentences. Occasionally, in natural language, we happen on words, phrases, and “collocations” that are (almost) unambiguous in meaning. Technical terns like “Trilobite”, “formaldehyde”, “fratricide” and proper names like “Thailand” are examples. Collocations include “media circus”, “team effort”, “world record”, and so on. These are low-hanging fruit for natural language-processing computer programs. Interestingly, they are almost context-independent, in that they have only one widely-used metaphorical meaning that is distinct from a little-used literal meaning. In the extension of this analogy, the HGP picked up only technical terms, proper nouns, and perhaps those collocations which are radically context-independent. So it should come as no surprise that only 2% of diseases currently admit of a straightforward genetic explanation.
Jacob et al. (1962) focussed over a long period of time on the work of regulatory genes in their concept of “operons”. These genes were a key to unlock the actions of a set of other genes, which for example would generate trytophan. Regulatory genes, in turn, were switched on and off by proteins. So genes can be turned on and off to respond to the environment. Let us now revisit the analogy, which includes “context” qua the environment. Gene expression now includes feedback loops. We effectively need a new HGP for every context that organisms encounter.
Strohman (opera cit) expands on these points. The expression of the genome is regulated through biochemical mechanisms that sense the bioenergetic state of the cell. In particular, the metabolites NAD and NADH and other synoptic signals represent instant by instant changes in the bioenergetic status of the cell. Changes of metabolites like fatty acids and glucose result in differential gene expression through binding to transcription factors. Strohman wishes to stress the consequences for progress in biology and medicine. Metabolism can be altered by environmental factors like sedentary behavior as by gene mutations like amyloid production. Alterations in metabolism, in non-syntactic phenomena, are the proximate cause of disease, and cures can be sought without interfering with the genome, or formal language.
Specifically, he argues that gene expression may be regulated by NAD dependent histone deacetylase via epigenetic marking of chromosomal histones as during the life extension resulting from caloric restriction. Again, context affects meaning of a language string in the linguistics analogy. (Let us emphasise that the major problem in nl processing has always been ambiguity).Secondly, he continues that alteration in protein amount or structure, as in dystrophin or amyloid may alter the rate of metabolic reactions resulting in an altered phenotype. Finally, posttranslational modifications like phosphorylation modify proteins. The rather messy picture that is emerging reminds one of NL semantics, and indeed of life itself
Strohman is also attempting an oblique attack at the central dogma of molecular biology; the deterministic, linear, uni-directional, and encapsulated path from DNA to phenotype. Specifically, he wishes to frame the relationship of genotype and phenotype in terms of a complementarity between genetics and dynamics, between the language as a formal system and the context, to use the corresponding linguistic terms. He unpacks and extends Waddington’s (1966) notion of the “epigenetic landscape” by proposing linkages and feedback loops between the DNA, phenotype, proteins, environment, and behaviour. Kaufmann (2000) remarks that this nexus of gene and environment must be kept at a very specific state, the “edge of chaos”, for maximum creativity to occur. Therefore, behaviors might be genetically assimilated to give Lamarckian effects without violating the central dogma that Waddington desired; alternatively, as in the case of the scandens finch on the Galapagos, birdsong might become entirely learned down the patrilineal line if emergent properties from the nexus of organism and environment decides that this is a more economical way of storing it than imposing it on the genome.
2.2 Formal Language theory and NL formalisms.
Derek Jarman’s classic short film about Wittgenstein proposes that he was nostalgic for the days of “The ice”; clear, lucid theories of language that he later eschewed, perhaps coincidentally (or not) after living for some time in this author’s native island of Ireland. . This is particularly the case as there are deep and beautiful connections between formal language theory, set theory, and computability. The Chomsky hierarchy posits languages of different levels of formal complexity. At the top level, level 0, are languages that can be modelled by non recursively-enumerable sets. The next level, level 1, features languages that are recursively enumerable. Both of these types of language, and the automata that generate and recognise them, are too powerful to be appropriate models for human language. Level 2 languages are generally also considered too powerful to be adequate models for nl; however, Shieber (1984) adduced evidence that Swiss German included constructions that only formal systems as powerful as level 2 could handle. (Partly because Swiss German, even more than other languages, is a set of disparate dialects, this result has never been fully accepted in the linguistic community). It cannot be proven that any randomly chosen level 2 language can be recognised in finite time; likewise, computational problems at this level cannot have posited of them a solution in finite time. These connections are reminiscent of the unexpected connections between fractals and chaos; a priori, there is no reason to anticipate this type of phenomenon.
The final explicitly recursive level, level 3, is the consensus location for nl. Even Swiss German can be handled by a relatively trivial addition to level 3 called indexed grammars. Level 3 is also called the level of “phrase-structure” grammars. It is illustrated by sentences such as “The mat, on which the cat sat, which Séamus made, is in the hall” where we need agreement between subject and its verb often at a considerable lexical distance. We describe such agreement using the mathematical expression (a^nb^n); Swiss German apparently needs constructions like (a^nb^nc^n). Even a^nb^n constructions cannot be handled with any degree of elegance by level 4 grammars.
Of course, we are uncertain as to where the genetic code lies in this schema. Phenomena like alternative splicing indicate that there is some degree of ambiguity, and thus complexity. Hox genes indicate a degree of hierarchical organization, context-sensitivity as defined in the terms of the Chomsky hierarchy (see O Nuallain, forthcoming c) and long-distance dependencies, indicating that perhaps the phenomena Shieber remarked on are present also in the genome. In any case, the HGP treated the Genome as if it was similar to the parody of language apparent in the pattern-matching programs of the 1960’s. Lisp programs were programmed in these systems with preprogrammed scripts like “I have problems with my x”, to which they would reply “Tell me more about x”. It is likely that parsing the genome is infinitely more complex than this.
There has been much superb formal and computational linguistics since the 1960’s, though the general problem of parsing any arbitrary sentence in any arbitrary natural language and extracting meaning therefrom remains inviolate. Thus, 100% accurate machine translation will forever remain a pipedream. The original so-called Chomskyan “revolution” which at least had the virtue of adding the rigour of Chomsky’s teachers to the area, was attacked with respect to its dichotomisation of “syntax” with a “black box” called “semantics”. In particular, the early Chomsky had difficulty with the manifest difference that could be wrought by a quantifier like “some” in the two sentences “Formal linguistics is nonsense” and “Some formal linguistics is nonsense”. Thereafter, at one extreme, formalisms like categorial grammar shifted the burden to the lexicon; therefore, the lexical item’s (or gene’s) possible syntactic and semantic roles were assumed encoded in it. It is likely that genetics will plumb for a formalism like this, even though others like lexical-functional grammar are manifestly superior for language. Chomsky’s later X-bar predilection is in keeping with the MIT tendency to impose more on the lexicon in the manner of categorial grammar.
Compositional semantics demanded that the syntactic components should each generate expression in a semantic formalism like lambda-calculus, which could then be summarized through processes like beta-reduction. The resulting output, corresponding to the “meaning”, was initially presumed to require no further parsing despite its opaque nature. However, the formal semantic analysis of a sentence like “Would you like to keep quiet for a moment?” in whatever semantic formalism does not constitute its meaning in any real sense. That requires pragmatics; the perlocutionary impetus of the statement in pragmatic terms is a request for silence, not an expression of the speaker’s pleasure.
Yet nl processing (nlp) made massive strides for some time; indeed, it is fair to say that an asymptotoic state was reached in the early 1990’s. Syntactic parsing is a textbook affair; the national software registry allows downloading of many useful tools. Type 3 grammars are well represented by systems like the Alvey natural language toolkit. This system maps a huge subset of English onto lambda calculus, a semantic formalism whose origins reveal much about the origins of formal computational theory. A process called beta-reduction makes the resulting output more economical. What then? Well, for each specific application, as is mentioned below, the words to be used must be implemented as separate Lisp objects and taught how to interact with the semantic formalism. We are right back to square one; the omni-pervasiveness of context. Likewise, even the best natural language generator systems like Charon has difficulty in formalizing the interplay of linguistic and conceptual elements. Absurd claims dating back to Schank’s (1975) work have not helped and perhaps quickened the tendency within genetics to use what looked like a happy language metaphor. It behooves us to look at context more closely.
3. Context
While “culture” has claims to be the most confusing world in the English language, even its meaning will depend on context. “Context” is much talked about, and very little understood. Even the most naïve PR person knows to complain that his egregious statements are being taken “out of context”. We have gotten little distance with this term in decades of nl by computer, despite everyone understanding what it is in folk psychological terms. What then is the prognosis for gene expression in context?
Perhaps context can best be introduced by considering the project of the early Wittgenstein (1922), hinted at above. Wittgenstein sought a grand unified theory (GUT) of language. He argued that what was out there in the world was “Sachverhalten” states of affairs. The world, he argues, “is all that is the case”. The Sacverhalten could be decomposed into Tatsachen, simple elements. Language, he argued, consisted of propositions that could be decomposed into atomic propositions. Atomic propositions could be mapped on to simple elements by what he called a “private language”, idiosyncratic to everyone (Think of it as the instruction set of a computer). Wittgenstein later repudiated this “Tractatus” idea of language in favour of one that gave a pre-eminent role to context. Language, he argued, should be considered in context; language-games like guessing riddles, playing chess, and telling jokes should be the focus of study. Otherwise, he argued, one ended up asking ridiculous questions in language like “What is a language”? To spell it out, one is already immersed in a language-game as one formulates the question, and one cannot bootstrap oneself up to an objective perspective.
Wittgenstein (1967) presses the attack at this point. Language is like a city, he opined, with warrens in old sections contrasting with the rationally laid-out modern thoroughfares that are analogous to scientific discourse in language. The process of interpretation of the language is going to be vastly different as one goes from one section to the other. Relatively clean types of interpretation that even the HGP could support will work for the modern section, but not for the warrens. Of course, Wittgenstein should not be our main source here. There is plenty of other evidence for the hypothesis that language does not admit of a single method of analysis and, by extension, that we need to be very careful about positing a monolithic type of gene expression.
Even allowing for the Stalinist bias he was forced to impose on his writings, Vygotsky (1962) can responsibly be interpreted as arguing for different evolutionary roots for thought and language. Thought can be discerned in animals’ problem-solving; language originates in signalling-systems like birdsong. Piaget’s (1972) monumental research oeuvre gave pride of place to “operational knowledge”; conceptual and motor knowledge whose origin is an internalisation of our interaction with the world. For Piaget, language is a form of operational knowledge; Vygotsky, probably more correctly, would emphasise more the autonomy of the linguistic apparatus. (Indeed, the loss of reputation that Piaget has suffered perhaps can be ascribed to his stubbornness on this point, which made him vulnerable to the Chomskyans as Cognitive Science developed. His most prominent intellectual heir is ironically the renegade Chomskyan George Lakoff, who does not acknowledge Piaget’s influence).
Again, the musings of two deceased psychologists may be of less than urgent interest to readers of this journal. Yet their speculations gain impetus from a survey of the successes and failures of nlp, and particularly machine translation (MT). Talented researchers like Nirenburg (Nirenburg et al, 1991) and Roger Schank’s students (Schank, 1975) implicitly followed the path of the early Wittgenstein, as they sough the Holy Grail of MT; fully automatic translation between human languages by mapping onto a set of language-neutral logical atoms and then mapping from these atoms onto target text. So, for example, we go from “Is fear mé” in Gaelic through a formalism like Schank’s conceptual dependency, which claimed that all meanings could be expressed in a set numbering in the low teens (it varied) of logical atoms and we output a Gallic Muddy Waters “Je suis un homme” in French.
Both Nirenburg and the Schankian school later saw the error of their ways, and began to produce effective systems that worked within specific contexts. In particular, Schank borrowed a concept from Piaget via Frederic Bartlett of the “script”, a stereotyped set of actions. So there is a script for entering a restaurant, for diplomatic visits, for earthquakes, and so on. Again, the parallel with Wittgenstein’s “language-games” is striking. The later Schankian systems (Schank et al, 1975) were surprisingly effective within micro-context; yet, as we have described above for the Alvey systems, the Schankian logical primitives have to be mapped to specific words in the micro-context (for example, “open”, “ask”, and “sit” in the restaurant example) for processing to occur. Nirenburg’s altogether more detailed work began to distinguish between generally available semantic distinctions (we always consider whether entities are living or not) to context-dependent ones (is that a cheesy teenage song, or not?). Our own work (Ó Nualláin et al., 1994) confirmed this as we used the Alvey tools to construct visual scenes on a screen on the basis of nl input.
Let us recap. All processing of language, and indeed all cognition, is contextual (Ceci et al, 1994). (Indeed, this author would argue (forthcoming, b) that the cognitive role of self is to prevent information overload by keeping contexts separate; the alternative is mental illness. We narrate to ourselves continually to prevent this). The relationship between syntax and semantics was at least addressed in nl processing by computer (nlp) (Lesmo et al, 1985) and needs to be in genetics. Contexts, at first blush, seem to be idiosyncratic interactions between linguistic and operational knowledge, which require knowing the precise relationship between the words (the genes) and semantic formalism (the metabolic context) in order for correct processing to occur (in order to predict what proteins will be generated). It is likely indeed that all of this holds for genetics; access to entities like perlocution seems to demand knowledge of oneself as the object of another’s wishes, and therefore consciousness, and is not relevant to genetics.
Cognition is also massively hypothesis-driven, and it is likely that every act of gene-expression draws in some way on the experience of the whole organism. We can distinguish in language between acts of context-determination, which don’t involve any but generally available semantic primitives like +-alive, and processing within contexts where there is interaction between context-specific primitives and words. It is not unreasonable to expect such phenomena in genetic expression. Yet there is a deeper consideration still in language and context with which we will close this section.
Any language processing act involves evidence from orthographic/phonotactic, lexical, syntactic, semantic/operational, and pragmatic sources of knowledge. The classic “pipeline” model sees information travelling strictly from left to right in this schema. So we assemble sounds into words, parse the words syntactically, add some semantic interpretation, and finally take into account where we are in the discourse, what the interlocutor is expecting of us, and arrive at an interpretation and/or course of action. However, as argued in Jackendoff (1987) and elsewhere, interactions can be a great deal more complicated. The letters “I L Y” in a certain context, that of a marriage proposal a la the famous scene in Anna Karenina, can result in a rigorous business contract with punitive buy-out clauses. For Wittgenstein (1967), a labourer saying “slab” was being quite linguistically sufficient in the circumstances of a construction project; he wants a slab, rather than to discourse on the geology of the material. So we can have connections from the orthographic or lexical straight to the pragmatic if context is sufficiently restricted.
Specifically, context seems to deform the layers of language as it becomes restricted in much the same way that gravity deforms space-time as one approaches the surface of a planet. At a level intermediate between reading the New York Times and I L Y, semantic relations are appropriated by the syntax in “semantic grammars” used for natural language interface to public services. The HGP worked on the assumption that the context was always going to be sufficiently restricted for single words to work, and therein lies its failure. It is a valuable lexicographic tool, and therein lies its success. However, we also need syntax, semantics, discourse pragmatics, and enumeration of contexts if nl is anything to go by.
Barbieri’s work (1998, 2002) is a fundamental analysis of biosemiotics and related fields like biosemantics. He makes the point from a semiotics point of view that the distinction between context-independent and context-dependent semantic primitives needs to be emphasised in genetics. It may indeed be the case that the HGP, considered thus, has revealed only the low-hanging fruit; context-dependent primitives. Alternatively, it may be useful to consider the HGP’s findings as collocations, or indeed as purely a lexicon; in any case, low-level fruit with context rigorously circumscribed.
Conclusions; future research directions
So is NL anything to go by? The answer is “at least partly”. In NL understanding, we are often concerned with getting the gist of a full story, and that is when the full artillery of linguistic techniques is used. In Gene expression, we are initially concerned with building a full organism, so the analogy is appropriate here. Natural language is also used to elicit a simple answer to a simple question, or specify a protein in genetic terms; such elicitation may be available to the technique known as “semantic grammars” in language, no correlate of which has yet been found in genetics. Can it be the case that combinations of metabolic context and nucleotide sequence recur in ways which are useful to predict the generation of proteins? A first step is obviously to do a human genome syntax project in the way that Ast (2004, 2005) has hinted at in his analysis of alternative splicing. Only then can we see if fatty acids are been taken up by transcription factors in systematic ways that echo the appropriation of semantic roles by syntax in “semantic grammars” (Ó Nualláin, 2003, Pp. 121-126).
Sometimes we understand language in order explicitly to act; we map the language onto a set of Piagetian “schemes” (Schankian “scripts”), routinised sequences of behaviour. Similarly, we read in order to evoke a script that we may or may not act on. Such leisure is not available to the genome; undoubtedly, however, the task of organism construction is a s complex as anything in language (and of course itself generates the elements of the linguistic apparatus).
The prognosis for genetics from analysis of the history of nlp must be, then, that it must start research into the syntax of the genome at a level much deeper than mere introns and exons. While there has been preliminary work on this, it is urgent that we decide issues such as what level of the Chomsky hierarchy the genome lies on. Are there Swiss German type constructions for certain creatures? Could it be the case that the difference between human and chimpanzee brains, to take one celebrated example, is that between Chomsky types? The consensus from Salzberg et al. (1998) is that, whereas the genome is susceptible to description by finite-state automata, protein-folding requires dependencies at long distances, and thus the artillery of context-free grammars. Abe et al. (1997) comment on protein prediction.
It must then try and understand the “semantics”, the metabolic context for the particular case that Veech et al (2001) examine. Then comes issues of “discourse structure”; how tasks like building and maintaining the integrity of the organism act as high-level goals affecting the minutiae of protein generation. Finally come issues relating to how these elements come together in specific contexts. There are generations of work ahead, all of which will bear fruit for biology and medicine.
Specifically, the template-matching “one gene, one protein/enzyme” story is childish. When we arrive at a more sophisticated understanding, it will have huge consequences for biology and medicine. Gene expression involves interactions of genome, environment, and emergent characteristics of networks of proteins as Strohman (opera cit.) has rightly argued. In this framework, we can begin to understand how exercise and diet affect metabolism, how metabolism affects gene expression, and how health is maintained as the dynamic set of relations it is.
References
Abe, N., & Mamitsuka, H. (1997). Predicting protein secondary structure
using stochastic tree grammars. Machine Learning, 29, 275-301.
Ast, G. (2004) “How did alternative splicing evolve?” Nature Reviews Genetics, Vol. 5, Pp. 773-782
Ast, G. (2005) “The alternative Genome” Scientific American, April 2005, 58-65
Barbieri, Marcello (1998). The Organic Codes. The basic mechanism of macroevolution. Rivista di Biologia-Biology Forum, 91, 481-514.
Barbieri, Marcello (2002). Has Biosemiotics come of age? Semiotica, 139, 1/4, 283-295
Beadle, G.W. and E. Tatum (1941) “Genetic control of biochemical reactions in Neurospora” Proc. Natl. Acad. Sci. USA 27, 499-506
Nick Campbell “Building the complete toolkit” Nature Reviews Genetics 5, 81-81 2004
Ceci, S. and A. Roazzi (1994) “The effects of context on cognition” In R. Sternberg and R. Wagner (eds.) “Mind in context”. NT;CUP, Pp. 74-101
Dennett, D. (1995) Darwin’s dangerous idea. New York: Simon and Schuster
Eurotra (1988) Technical manual 5.0 Luxembourg: CEC publications
Goodman, M., Grossman, LI & Wildman, DE Trends. Genet.21,511–517 (2005).
Hofstadter, D. (1979) Goedel, Escher, Bach. New York: Basic Books
Jackendoff, R. (1987) Consciousness and the computational mind. MIT Press.
Jacob, F., R. Sussman, and J. Monod (1962) “Sur la nature du répresseur assurant l’immunité des bactéries lysogenes” Comptes rendus des Academies des Sciences, 254, 4214-16
Kauffmann, S. (2000) Investigations. New York: OUP
Lesmo, L., and Tirasso, P. (1985) “Weighted interaction os syntax and semantics in natural language analysis” In proceedings of IJCAI (pp772-778) Los Angeles
Nirenburg, S. and Goodman, K. Eds (1991) The KBMT project. San Mateo, Ca: Morgan Kaufmann
Ó Nualláin, Seán (2003) The Search for Mind; third edition. Exeter: England
Ó Nualláin, Seán (forthcoming a) “Subjects and Objects” Biosemiotics
Ó Nualláin, Seán (forthcoming, b) “Ask not what to do for your self” Cognitive Processing
Ó Nualláin, Seán (forthcoming, c) “Code and context in gene expression, symbolic cognition, and consciousness” In M. Barbieri (ed) Biosemiotics 2 (provisional title). Kluwer.
Ó Nualláin, S, Arnold G. Smith: An Investigation into the Common Semantics of Language and Vision. Artif. Intell. Rev. 8(2-3): 113-122 (1994)
Piaget, J. (1972) Principles of genetic epistemology London: Routledge
Steven Salzberg, David Searls, and Simon Kasif (1998) Computational Methods in Molecular Biology. Published by Elsevier Science
Schank, R. (Ed.) (1975) Conceptual Information Processing Amsterdam: North Holland
Schank, R. and Abelson, R. (1977) Scripts, plans, goals, and understanding. Hillsdale, NJ: Erlbaum
Searle, J (1980) “Minds, Brains and Programs” The behavioural and brain sciences
Shieber, S. (1984) “Evidence against the context-freeness of natural language” (TW-330) Menlo Park, Ca:SRI International
Strohman, R. (2000) “Organization becomes cause in the matter” Nature biotechnology, June 2000, Vol. 18, Pp. 575-576
Strohman, R. (2003) “Thermodynamics- old laws in medicine and complex disease” Nature Biotech, May 2003, Vol 21, Pp. 477-78
Veech, R.L., B. Chance, Y. Kashiwaya, H. Lardy, and G. Cahill (2001) “Ketone Bodies, potential therapeutic uses” IUMB Life 51:241-247
Vygotsky, L. (1962) Thought and Language Cambridge, Mass: MIT Press
Waddington, C (1966) Principles of development and differentiation NY: Macmillan
Wittgenstein, L. (1922) Tractatus logico-philosophicus Oxford: Blackwell
Wittgenstein, L. (1967) Philosophical Investigations Oxford: Blackwell
Saturday, April 7, 2012
Ferdinand Prądzyńscy
In Polish if "a" has a tail , pronounce "n" - thus ond. Thus, it seems that we have also been conned by one Ferdinand Prądzyńscy, scion of the family business running the Oppole/Groszowice forced labor camp that comprised part of Stalag 8b. Poles with whom I've spoken say it is rare indeed for a polish name to be paired with "Von" and indeed there is no mention of the honorific in this correct history of his family's business;
http://pl.wikipedia.org/wiki/Groszowice
The official family history includes the Dyckerhoff connection and here it is definite that there was a claim for slave labour damages;
http://articles.chicagotribune.com/1999-05-11/news/9905120086_1_volkswagen-ag-slave-laborers-continental-ag
Look on the bright side; we could have been anally raped by agents of the Irish state like Mannix Flynn was. As it is, the Scots owe the world of academia a proper investigation into who this person is, what his family did during WW2, and why they were in Ireland.
Otherwise this is likely to get ever nastier; I for one resent the fact that not only was my life and career ruined, but hundreds of jobs were lost through this individual's incompetence and criminality
Of course, perhaps the Poles think the Prądzyńscy family are too modest to have their "Adel" status pointed out.
Seán Ó Nualláin Ph.D
Domhnach na Casca 2012
PS "adel" is of course the keyword to look for as the Scots check to see if they are unleashing something unspeakable first on their country and then on the world as it relates to the the register compiled in the wake of the recission of "nobility" (!!) the Weimar Republic. I am told by native German speakers that Ferdie's German "has something wrong with it".
My rather painful conclusion from all of this is that Ferdie is simply an attempt at corporate takeover of universities. The scene was set by the creation of a "CEO" role, first in Britain and then in the insane 1997 Irish act. They then had to find someone willing to destroy lives, and arrogant enough to imagine that - with his obviously limited ability - he could stand the heat as much better and brighter people than him objected to his policies.
Why universities? Well, because they have a lot of resources, because SIPTU had been bought off in Ireland( IFUT are thoroughly useless as the Evans case shows), and because they figured academics only want a quiet life. Not true; we want a life of intellectual engagement in civilized circumstances which only corporate whores imagine is the same as "quiet".
The viciousness of what went on was exacerbated by the Irish Times nearly going broke, and being mysteriously bailed out around a decade ago. The result is that this paper, which should have been our first recourse as academics to relate the pressing threat to academic freedom that even chief Justice Denham noted from the bench of the Supreme Court, instead had to be SUED to give an even-handed and accurate account of the Denham case. The Sindo and execrable British Daily Mail did what one would expect.
The only major article on academic freedom allowed by the IT was by Ferdie himself. Foxes minding hen-houses and indeed pimps managing girls' schools come to mind.
____________________________________________________________________
PPS
Summarizing what's below; the Prądzyńscy cement factory was merged in 1941 with the slave labour one in nearby Opole, which also had an SS-run ghetto and liquidated its Jews. Charming. It's incumbent on the Scots to confirm or refute this; as things stand, they have a university president with blood on his family's hands.
Excuse the bad translation (not my fault but Google's);
"
On 27 December 1941, the company is Portland cement and lime plants Saxony-Anhalt, Inc. Nienburg expired. The total assets of the Company was transferred to the Silesian Portland Cement Industry Ltd Opole. The plants in Saxony-Anhalt Nienburg, Concordia and Jesarbruch were transformed into works of the United Nienburg OMZ AG. Also approved the two subsidiaries "eagle" and "Prüssing" the merger agreement. According to the Annual General Meeting of 12 December 1941 the Silesian Portland - Cement - Industrial AG Opole renamed United Eastern and Central German Cement AG. The unification of the OMZ AG included the following works: Groschowitz, Oppel port and Giesel, Neudorf, Silesia, Frauendorf, Dornburg - Steuditz, Thuramentwerk Rosenberg and Sulzbach - Rosenberg, Göschwitz, Unterwellenborn, Berlin - Tempelhof, Rudersdorf, Saxony-Anhalt and Concordia and Jesarbruch "
This is the family company of the ex-President of DCU - by his own admission 0- and it was used for slave labour, as was Dyckherof, the company that his father then went to work for having done God knows what in WW2.
______________________________________________________________________
More research; it turns out there was a ghetto in Oppeln, with forced liquidation of its Jews;
And now this extraordinarily poignant note;
http://kehilalinks.jewishgen.org/Lackenbach/html/LOBL_YIZKOR.html
"LOEBL Benno born 21. june 1889 in Lackenbach, shoemaker
LOEBL Elsa, nee Zoller, born june 1895,
last adress: Wien 2, Grosse Schiffgasse 6/4
deportation with transport 1 to Opole on 15. feb. 1941
On February 15th, 1941 and February 26th, 1941 two deportation transports
with altogether 1962 Jewish men, women and children on board left Vienna
Aspang Station bound for Opole, a small town south of Lublin.
Opole had a long established Jewish community; when war broke out about
4000 Jews lived here, i.e. about 70 % of the population, a proportion
which rose further after the beginning of the war, as Jews from other
parts of Poland were forcefully resettled here.
By March 1941 about 8,000 Jews were deported to the ghetto which had been
set up in Opole. The new arrivals were either lodged with resident Jewish
families, or in mass accomodation, as for example in a synagogue or in
newly erected huts.
In the ghetto itself no restriction was placed on the freedom of movement
of the inmates, and there were no boundary lines, yet it was forbidden on
the threat of severe punishment to leave Opole without official permission.
Control of the ghetto was undertaken by the security service of the SS (SD), the Gendarmerie
and also, as may be concluded from witnesses' testimony, by German army soldiers.
The inhabitants of the ghetto were largely dependant on themselves
as far as earning a living was concerned.
From May 1941 about 800 men capable of work were deployed as forced labour in Deblin.
The liquidation of Opole ghetto began as early as spring 1942.
A transport to Belzec extermination camp left on March 31st, 1942,
and deportations to Sobibor followed in May and October 1942.
Of the 1962 Viennese Jews 28 are known to have survived. "
