I assume you are talking about this VCF file:

http://cdna.eva.mpg.de/denisova/VCF/hg19_1000g/T_hg19_1000g.20.mod.vcf.gz

The BAM file at that position contains 20 Ts on the + strand, 7 on the - strand. I am not sure as to why there is an alternative allele but I can tell you that a custom script was used to produce those from an older version of GATK. The reason being is that the older GATK could not call heterozygous for 2 alternatives but we could set the reference bias to 0. The newer GATK could call heterozygous for 2 alternatives but we could not set the reference bias to 0. The solution that what we had to do is: 1) call genotypes once with the older version of GATK on the normal genome, 2) mutilate the reference to add alternative alleles and 3) recall genotypes on this reference, 4) combine both calls into one. There could have been some errors along the way.